Vibe coding at production scale works when one orchestrator delegates to fresh-context workers while the human personally reviews architecture and payments, the parts where a mistake costs real money.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You run several Claude Code or Codex sessions at once and want a model for coordinating them through one orchestrator.
You are adding credits, subscriptions or usage metering to an AI app and want a checklist of what breaks.
You are about to ship a paid mobile app and want to know where Apple and Google let you link to web checkout in 2026.
You are deciding between Mastra and LangGraph, or between MCP and a CLI, for an agent system in TypeScript.
SKIP IF…
You want a tight tutorial; this is a six-hour unedited stream with long stretches of waiting on agents and chatting with viewers.
You need a step-by-step App Store setup; the store configuration happened off camera.
You have no interest in the AskLog tool itself; a fair share of the stream demos and promotes it.
TL;DR
The full version, fast.
The stream argues that AI-built software holds up when a human supervises the architecture and the money paths while an orchestrator does the typing. The host talks only to Chief, an Opus agent that starts Claude workers in separate git worktrees, sends their PRs to Codex for review, and keeps its own long-term memory. Over six hours it recovers from a 78-PR overnight mistake, designs a pricing page, and drafts a credit system the host audits line by line: one credit per cent of model cost, no overdraft, webhooks over browser redirects. It ends on a hybrid mobile plan of US link-outs plus slightly higher in-app prices, with both store builds submitted.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Holding card while the stream goes live on YouTube and Twitch.
03:19 – 11:40
02 · The 78-PR overnight mishap
An orchestrator left running overnight misread one instruction, split a feature into dozens of workers and opened 78 pull requests. Intro to the AskLog desktop app, the Chief orchestrator and the Noticing second-brain app being built.
11:40 – 23:20
03 · Stack and architecture decisions
Schema-less tables like Notion and Airtable, a space ID as the sharding key from day one, hybrid RAG with BGE-M3 picked off the MIRACL multilingual benchmark, Mastra over LangGraph for TypeScript, and a benchmark showing skills cut input tokens by more than half.
23:20 – 29:20
04 · Q&A: is a CS degree still worth it
College syllabi lag the industry by years. Skills you can show (shipped web and mobile apps) beat DSA grinding alone; generative AI and cybersecurity named as the two safest specialties.
29:20 – 44:00
05 · How the Chief orchestrator works
Chief is an Opus agent that talks to workers through the AskLog CLI (modeled on tmux) instead of MCP. Every worker gets a fresh context and its own git worktree; Chief keeps a long-term memory it queries instead of loading. Local dev servers run in visible tabs, not hidden background shells.
44:00 – 1:00:55
06 · Prompt caching, tool search and Redis
How to mark a system prompt as cacheable in Mastra with an ephemeral cache-control flag, why 30+ MCP tools bloat context, replacing them with one RAG-backed search tool, and where Redis or semantic caching fits in front of a database.
1:00:55 – 1:19:00
07 · Testing Noticing live
SSH tunnels to a shared staging Postgres add latency; the manual table builder ships with obvious UI bugs after the PR merge. Chief spins up QA, lint and smoke-test workers and is told to divide work by feature, not by tiny task.
1:19:00 – 1:31:00
08 · Viewer project: an interview copilot
Advice to a viewer building a meeting assistant: drive Codex or Claude CLIs through an embedded xterm terminal rather than OS accessibility hacks, plus how interviewers now screen for cheating tools.
1:31:00 – 2:03:00
09 · Designing the pricing page across two Claude accounts
Chief has a Codex computer-use agent export the Claude Design project zip, re-import it into a second account whose tokens had not run out, then hands the design back to a Claude worker. Credentials come from 1Password vaults, never .env files.
2:03:00 – 2:58:00
10 · Payments and credits review in an AskLog page
Line-by-line review of the payment architecture: one credit equals one US cent of model cost, rounded up per call; pause the AI at zero with a warning at 10 credits; never trust a browser success redirect, only signed webhooks; Stripe for the world, Razorpay for India.
2:58:00 – 3:22:20
11 · Viewer project review: a social network
Network-tab review of a viewer's Preact social app on Cloudflare Workers: optimistic UI on likes but not comments, no rate limiting, polling instead of WebSockets, and no front-end query caching.
3:22:20 – 4:26:30
12 · Mobile payments: Apple, Google and the 15%
An agent researches store rules: US iPhone apps may link out to web checkout with zero Apple fee; elsewhere link-outs get apps rejected. Decision: a hybrid with in-app purchases priced slightly higher than web to cover the store cut, plus Small Business Program enrollment.
4:26:30 – 4:50:00
13 · AskLog product tour
How to install AskLog, connect several machines without SSH, run Chief docked or detached, browse worktrees and memory. Claim: one orchestrator with fresh-context workers uses fewer tokens than sub-agents that pipe results back.
4:50:00 – 5:31:00
14 · Store setup off camera
Stream paused while Codex agents walk the Play Console and App Store Connect payment profiles, agreements, Google Pub/Sub monetization notifications and the sandbox tester.
5:31:00 – 6:07:29
15 · Admin panel, startup credits and wrap
Admin console deployed to a Kubernetes pod reachable only over Tailscale, a researched list of startup credit programs, CI tests switched off to unblock release, and both app builds queued for review before sign-off.
Atomic Insights
Lines worth screenshotting.
An orchestrator told to work overnight with multiple workers split one feature so finely that it opened 78 separate pull requests.
Tell an agent orchestrator to divide work at the feature level; one worker per tiny subtask produces a merge nightmare.
Benchmarked on the same task, using a skill cut input tokens from about 26,000 to less than half.
A CLI that can kill, attach and send prompts to terminals can coordinate agents faster than routing everything through MCP.
Every worker an orchestrator spawns should get a fresh context and its own git worktree.
Sub-agents that report back into one parent session burn more tokens than independent workers reporting short status messages.
Thirty MCP tool definitions in every request cost more context than one search tool that finds the right tool by meaning.
Design the sharding key on day one; adding it after launch means migrating every row.
One credit equals one US cent of model cost at list price, rounded up, so no AI call is ever free.
Never grant credits on a browser success page; only a signed webhook proves a payment happened.
Apple and Google take 15% under their small business programs, and only US iPhone apps may link to web checkout fee-free.
Being a tiny new app does not hide a store-policy violation; it makes you more visible to reviewers.
Charging $10.99 in the app and $10 on the web is allowed on both stores and covers most of the fee gap.
Optimistic UI without server-side rate limiting lets anyone hammer your database by looping a like button.
Use Codex for computer use and browser tasks and Opus for writing code; each model wins a different job.
Takeaway
Let agents type, but own the money paths yourself
WHAT TO LEARN
One orchestrator running fresh-context workers can ship a launch-ready app fast, as long as you set how it splits work and personally review architecture and payments.
02The 78-PR overnight mishap
An autonomous orchestrator needs a rule for how finely to split work; split by feature, not by every small subtask, or you inherit dozens of PRs to merge by hand.
Run multiple coding agents from one controller you talk to, rather than prompting each agent yourself, so a single instruction fans out into parallel work.
03Stack and architecture decisions
Store user-defined tables as rows in a generic table, column and row schema instead of creating real SQL tables per user, the way Notion and Airtable do.
Pick a sharding key such as a workspace or space ID on day one; retrofitting it later means painful data migrations.
Choose an embedding model against a benchmark that matches your use case, such as MIRACL for multilingual retrieval, instead of defaulting to the best-known API model.
Package repeat instructions as skills: in a measured comparison the same task dropped from about 26,000 input tokens to under half that.
05How the Chief orchestrator works
A small CLI can be a better agent-to-agent protocol than an MCP server for controlling terminals: kill, attach, send a prompt, read output.
Give every worker a fresh context and its own git worktree so parallel agents never step on each other's files.
Keep the orchestrator's long-term memory outside its context window and fetch it on demand, so the session never fills up with history.
Run local dev servers in visible terminals you control instead of agent background shells, so you can see crashes and logs.
06Prompt caching, tool search and Redis
Prompt caching pays off only when a large prefix such as the system prompt or loaded memory repeats exactly across calls.
Loading 30 or more MCP tool definitions inflates every request; one search tool that retrieves the right tool by semantic match keeps context small.
Put a Redis cache in front of read-heavy queries whose data rarely changes, and invalidate it when a write lands.
07Testing Noticing live
An SSH tunnel to a shared staging database is secure but adds latency to every request; a misrouted cluster once pushed responses to 15 seconds.
Merging many parallel PRs without a QA pass breaks the UI in visible ways, so schedule a dedicated test worker right after big merges.
09Designing the pricing page across two Claude accounts
When a design tool runs out of tokens on one account, export the project and re-import it into another so the new session keeps the design system.
Keep secrets in a password vault the agents can request from, and leave no .env files with live keys on disk.
10Payments and credits review in an AskLog page
Define a credit as a fixed slice of real model cost, for example one US cent at the provider's list price, and round every call up so nothing runs free.
Stop AI usage at zero credits and warn the user a few turns before; allowing overdraft invites abuse and silent losses.
Treat a browser success redirect as untrusted; only a signed webhook should grant credits or a subscription.
Brainstorm and approve the payment flow as an activity diagram before letting any agent build it; payments are the one area not to vibe code blind.
11Viewer project review: a social network
Use optimistic UI consistently: if likes update instantly, comments should too, or the slow path looks broken.
Optimistic actions need server-side rate limiting, otherwise anyone can loop a like request and hammer your database.
Polling can feel real time for chat at small scale and is far easier to run on serverless than WebSockets.
12Mobile payments: Apple, Google and the 15%
Apple and Google take 15% under their small business programs, and 30% above roughly one million dollars a year.
Only US iPhone apps may currently link out to a web checkout without an Apple fee; elsewhere, naming your website as a place to pay gets builds rejected.
Being a small new app does not hide a rule break; every iOS build gets human review and Google scans releases automatically.
Pricing in-app purchases slightly above web prices, such as $10.99 versus $10, is allowed and protects your margin.
15Admin panel, startup credits and wrap
Host an internal admin panel only on a private network such as Tailscale instead of exposing it to the public internet.
Cloud and AI startup credit programs vary widely; several require VC funding or a partner, so shortlist the ones you qualify for before applying.
Glossary
Terms worth knowing.
Orchestrator agent
An AI agent whose job is to plan work, start other agents, hand them tasks and collect their results, so the human talks to one agent instead of many.
Git worktree
A second working copy of the same repository checked out to its own branch and folder, letting several agents edit code in parallel without overwriting each other.
Prompt caching
A provider feature that stores a repeated prompt prefix, such as a long system prompt, so later calls reuse it at a fraction of normal input cost.
Sharding key
The field used to decide which database partition a row lives in, such as a workspace ID, so data can be split across servers as it grows.
Hybrid RAG
Retrieval that combines keyword search with vector similarity search to find the most relevant stored text before an AI model answers.
BGE-M3
An open-source embedding model that converts text in many languages into vectors, often chosen for multilingual search.
Mastra
A TypeScript framework for building AI agents and deterministic workflows, with a built-in studio for testing agents and viewing traces.
Smoke test
A quick pass over an app's main features after a change to confirm nothing major is broken before deeper testing.
Webhook
A server-to-server message a payment provider sends when an event happens, such as a completed charge, signed so your app can verify it is genuine.
Credit ledger
A record of every credit added or spent per user, used to compute balances and stop usage at zero.
Small Business Program
Apple's and Google's reduced 15% commission tier for developers earning under about one million dollars a year from their store.
Optimistic update
A UI pattern that shows the result of an action immediately, such as a like, while the server request finishes in the background.
Rate limiting
A cap on how many requests one user or IP can make in a time window, protecting servers from abuse and accidental floods.
tmux
A terminal multiplexer that runs many terminal sessions inside one window and keeps them alive after you disconnect.
xterm.js
A JavaScript terminal emulator that lets a desktop or web app embed real terminal sessions.
Tailscale
A private mesh VPN that connects your devices and servers so internal tools are reachable only by you, not the public internet.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphorstory
hello hello guys how are you just give me one second guys i'll make sure my voice is correctly set just give me one second just give me one second i'll be back right everything is working fine i guess the camera is having a little bit of lightning effect i don't know why okay let people join so i think um i can see people are joining hello shivam bhai kaise ho yo bro can you do a one shot live an idea to uh deployment in a single live when are you gonna do that probably tomorrow tomorrow in the daytime also we can stream guys like tomorrow is holiday and let's see what happens if we can do a live stream tomorrow like a full live stream taking a complete idea from scratch and then building it up to the deployment deploying it on wordcell deploying it on servers vps and everything right that will be crazy right hi hi aditya well hello i am on discord too amazing so i'm gonna also join discord so that we can discuss on discord as well just in a second looks like my live is going good on i'm just gonna verify quick thing whether we are live on twitch as well by the way from tomorrow guys like i'm gonna have uh i think i'm gonna take a subscription for uh widest even the creator one where i can actually live more than uh those so i'll check that if that is possible so we'll be actually doing live on twitch as well yep i can see
i can see we are live on um i can see we are live on twitch as well that's amazing do live on kick also why not why not we'll do we'll do live on kick as well yeah so are you gonna do that tomorrow bro i'm excited uh i'm not really 100 sure but if i get time definitely why not i'll have to think about the idea that maybe you can give me some idea what to build on and then we'll take that idea completely from scratch and take we'll take take that to prod okay so android uh has been since honor all right so we have by the way guys one good news that we have our application live on android as well but currently invite only there are a lot of issues here and there yesterday i'll tell you one um one bad thing which happened uh with the chief and it got a little hallucinated and it get one of my instruction it got one of my instruction wrong and it actually created around 78 prs guys literally so i asked it to basically uh do autonomously over the night but uh you know instead of you know working on uh single agent working on that feature you should ideally
launch multiple agents or sort of like multiple workers and it literally completed the work and i'm i was so happy when i when i wake up in the morning but then i realized that it actually has literally launched 78 prs and right now like i have to close all of these those prs manually i mean i have to literally tell it to do the uh do that manually do you do do live on kick also uh just build ask astrolog mvp by the way that's a good idea we can we can do that we can do that that's a good idea i'm gonna brainstorm what i'm gonna build on that ask astrolog and we literally will build that what do you think about ai fiesta ai fiesta uh i've seen his video dhruv's video on the latest ai i was like a little excited about it so i think i mean uh the thing that he's trying to provide anyone can build that to be honest like it's not that bigger task ultimately he's basically created our uh api router sort of and a chat router where you can route multiple agents that's not if you if you literally work on over the weekend you can literally build a project whole project so i don't think that it's so unique idea he's doing something different is it worth for cloud nope it's not worth worth it you can add a comparison feature uh to wire chief yeah i mean why not
and it's so easy to build something like that to be honest like okay i'm gonna actually open on discord also i'm gonna join the live hangout so guys if you are if you are on live hangout on discord you have you can have you can be there you can discuss up there as well so look at these guys like yesterday it was a very very uh bad thing this has done literally you can't even imagine how many prs i have to merch you know yesterday it was like literally a bad thing what are you building bro hi chetan first of all welcome to the stream so we are actually uh we have actually built one this tool called as clock let me actually do one thing let me put all of these so that on the streams it doesn't lag yep so we build this this tool called as clog uh which you can download for your mac you can download for your linux windows whatever and then you can also download the mobile application now the what that what this tool does it's pretty amazing so it literally uh you can run multiple sessions like you can see how many sessions i'm running on this on my machines i you can run multiple sessions on um your um like it can act like a terminal session and then at the same time you can
connect to these sessions through your phone so you don't need to be on laptop all the time by the way you can connect your multiple machines also like you can see i have connected my mac mini i've also connected my office laptop into it i work just on my one system mac pro macbook pro and then i can literally do everything and by the way this is the cool feature that we have a chief chief is basically like an orchestrator so you see all of these agents running right three uh codex agents running for the review and all these agents that we are running here and by the way we are building one application called noticing using as clock this is already built this is already live so we are building one application which i'll explain in a minute uh and if i tell you what this chief does it's it's like an orchestrator a very smart orchestrator it manages all these agents for me i just have to talk to this single agent and it does everything for me that's cool it is and by the way the application that we're building i'm gonna open the quick uh if i can i'll open the design of that cloud design yep this is the one so we are actually building one uh brain application which can literally you can talk to it you can store information in it
uh like this is the application right you can just talk to it you can store all your information on onto it and then you can ask about it it stores your data very nicely i think this is the old version let me actually open the latest one this is the ren mobile version yeah so this is the application that we are building it has a ren agent you can literally talk anything to it you can store any information you want to store anything you want you want to memorize it's like a super memory super brain kind of application and then what it does in the back end it actually stores all of your data in very strategically it has a rack in the back back end which stores this data very strategically not just simple rack by the way it's like hybrid rack on top of that we are having a schema less data management in the back and so whenever you want to ask about that data you can ask so you'll see the demo of also i'll start with the demo actually today and by the way while building all of these things right they're actually i'm trying to explain different concepts to different to to my viewers like they actually literally uh learn about uh llms they we have we discussed about evals we discussed about prompt injection guardrails even from the very basic front like react agent how does it work so
the idea of building this application is not just that you know building it just for the users but also uh to actually teach people the concepts of agentic AI right the best practices out there like talking about deterministic workflows versus agents talk about like memory layer talk about MCP so whatever we are building we are using s -clock and we are literally explaining people how things go so we discussed about memory different types of memory semantic memory episodic memory observational memory working memory and by the way we're using all these concepts in the application so in front of you if you're watching these streams you're literally learning about all these concepts by actually seeing the real life example and you know i'm gonna basically you can see how how much this escalator diagram has like completely messed up it's because we discuss a lot of things sometimes we just discuss about infrastructure deployment right sometimes we just discuss about which technique tech which stack we should use right currently we're using this stack for this application right ui is on next year's expo we are using for more
mobile application development, Postgres for database, Quadrant for vector DB, Drizzle ORM. So all of these things we are literally discussing on the stream. How cool is that?
and uh right now the status of this application is that it's already built in a very nice shape we have a complete multi -agentic workflow we have a couple of workflows couple of agents inside those workflows it's like multi -agentic uh you know the kind of workflow we also have created our mcp that yesterday and then that's actually really really working pretty fine i mean that we tested the mcp like the actual mcp server we've created so it's working really really great right so you're gonna learn a lot of things i want to use cloud uh on pro but tokens are not too enough that's true i mean with cloud it's not too enough but you can use open code you can also use like um you can optimize your token usage by using something like this orchestrator so one one benefit of using chief is the is this that it saves a lot of token one second guys one second one second one second i'm gonna just reduce the volume a little bit for now and then maybe we'll increase that later on yeah i think this is fine all right so
hi prodigy engineer hi bhai kaise ho yaar like welcome to the stream once again freddo india wrx hello welcome to the stream we can discuss how to connect uh my app with an ide or folder at the end of the stream when you have time yep so for sure like we can do that we can discuss about that that's a good thing to discuss right people will learn um sql type it it is like a sql type sort of you can under you can call it like that uh shadow king but technically we are not actually uh creating multiple tables so when i say like let's say for example if i use noticing .com it doesn't really mean that when i ask it to do something and it stores in the form of table but that we are not actually creating sql tables because look at think about it right you are having an application right and you literally have You're literally creating SQL tables in the backend, for example, right?
It is SQL type for sure, but not exactly SQL. Though we are using SQL queries and we are using SQL concepts to build out the database and interact with the database. But it's like, you know, if you start creating, let's say, for every single use case, for every single user, multiple tables in the backend, multiple tables in the database, your database will just bombard.
Like, I mean, it will just quickly fill up, right? So that's not something we are doing here. We're using some concept like, which Notion also does, Airtable also uses.
It's like schema -less tables you're creating. So we actually have a table for table. and then we have columns and rows and then we are kind of like you know you creating a relationships between them so that we can like technically create as many tables we can and by the way one thing we are scalable first so whatever way i'm discussing on these streams and making sure that the product is scalable so on the very very first day when we were designing the database of this we actually thought about sharding right normally people don't don't think much about sharding at the very beginning it's like they think that it is very very like over complicated or over engineering but i think that if your designs from the very day first is scalable it really helps you because later on if you do that it you'll have a lot of data migration issues you have to actually you know so right now whatever we are building in noticing application i'm going to give a quick demo to you guys everything is um sharding first so we have a space id which is our sharding key and that we are using for everything single place chetan frr damn bro the way you explained it
Truly amazing you gain my subscribes up will definitely follow you on future as well. Thank you so much Ethan And that's what I'm that's what my idea is. That's what I want to do during these streams That I share my knowledge Whatever I know about agentic system.
So currently if I tell you about a little bit about myself I'm working on a Indian company, which is working for a very big client like a very big name in Japan I'm actually leading a team of AI engineers over there. So I have a very good amount of experience in In it industry like almost like 10 plus years of experience you can say six years professionally But and then apart from that i've been like doing a lot of freelancing.
I've created a lot of projects killed those two like Thousands of users active users, right? Uh, so I have a good amount of experience I have I came from a background where I started with php laravel code igniter went to python based frameworks like zango flask and then now move to like not know how but like couple of years back four five years back move to node .js react native electron .js all of those so i have like a variety of experience around around it and that's what i'm basically sharing with you and not just about node .js and python i've actually worked with lecture as well if you know about this and you know you we all use discord do you know like what discord is built on top of discord discord is built on top of elixir language elixir is a very nice language which is really really considered very very good for scalable systems in terms of real -time communications right so all of that thing is basically happening here so whatever knowledge i have gave that guys
like whatever fundamental concepts about serverless uh server for technologies like we talked about a lot we locked we literally brainstorm every single thing like say like when we were actually building rag right we literally gone to the very fundamental level that what could could be the best embedding model for us right And think about it, people when they build RAC, they literally go with text embedding large, they go with open AI models, Claude doesn't have their own embedding models, but they go with like open, like they very rarely think about having something like BGEM3, which is a so nice, cool model, right?
So we discussed about our use case, our use case is mostly around multilingual, and we have a benchmark called Miracle. If you can search about it, you'll get to know. And literally, BGEM3 is considered a best, it actually, like we have done a...
You know, DAR means like comparing different technologies, libraries and coming to the conclusion which is best, right? We've done a DAR around BGAM3 and like, sorry, around embedding models. Then we decided about this, right?
That's how like, even for the choice of framework that we are doing, right? When I actually started this project, since I use Opus 5 .5 for development these days and Codex for the review, Astra and, you know, computer use and browser use. and literally when i started this uh this started with this uh agentic framework it actually recommended with me anthropic sdk can you imagine anthropic sdk the problem with anthropic is that sdk it is not model agnostic it only works with cloud model though you can tune it and kind of like do some hack to um work it may be possible with um you know open ai also but that is not something we want to do right then we had an option of using virtual ai sdk which is by the way model agnostic but not so mature then we have mature frameworks also like langchain langraph like one of my best framework like if i would want to go with python language i would suggest langraph obviously it has like you can orchestrate multi uh agentic systems workflows all of that can be done and by the way one thing but you know since we are on this project it is more like typescript heavy and i have like uh create like use this framework called maestra on production level see fundamentals of every single agentic libraries and frameworks are same only right
But some of them actually, for example, crew is really good when you're building something like crew kind of setup where you have one master orchestrator agent and then you have a couple of crewmates, right? Then crew is considered really good, right? When it comes to any kind of standard workflows, just deterministic workflows and non -deterministic agents, these Langrath and Maestra are really, really amazing.
The problem with Langrath is that it's really good when it comes to Python because it's very, very mature. i would not say it is not mature in typescript but like in our company we have literally done a lot of comparisons around maestra versus langraph especially around typescript based framework and we actually concluded that maestra is really amazing and that's what like i've explained during the stream also and one another reason of using maestra is that maestra comes with out of the box um Like studio master studio where you can literally see trace it and visualize all your agents, right?
And that's what we do also like for example if I am not sure if it is running or not But let me actually quickly try it out. So if I can literally run master studio I'm not sure it is literally running and it is running. So you see this thing If you have used, created any APIs in Express or, you know, used Swagger, sort of like OpenAPI, kind of like Explorer, you must have noticed that when you create APIs, out of the box, it gives you a platform where you can literally test those APIs.
Similar thing is with Maestra. When you are creating this agentic workflows or agentic framework, like multi -agentic framework, by the way, our agent looks like this. I mean, I just had a diagram of that.
and maybe i'll come to that later so master studio really literally have so these many agents that we have right now and these these are the three workflows that we have right i have also explained what are the difference between workflows and agents right but why workflows are considered good by the way we are using mix of both of those right so literally you can literally go inside any workflow and you can see what is going inside that right you can see the past runs of that you can literally go to any agent and actually can talk to that agent right that kind of functionality maestra gives us and it is really really good when you try to explain to people when literally when i'm teaching these concepts on these live streams it becomes really really easy for uh people to visualize stuff right here right you can literally see what is the system prompt of that particular and we can see traces we can see work i mean workflows is not really needed we have talked about skills how are skills like people know very less about skills whenever i you know i take interviews around skills i just like when i ask what is a skill people just say it is just you know giving expertise to an agent right it is for example marketing can be a skill back end but what you know what what we have in these streams we actually have created one devops buddy use case and i we actually proved this thing that how
skills not just only give expertise and additional stuff to agents but also helps in reduction of token and that's something which is very very important and we literally have taken this example of this devops buddy we benchmarked it with the skill with not scale right and we literally come to this when we are not using skills it just take like 26 000 tokens input token and then 831 output most likely will be same only it's more about input token But with the skills, it is very less, almost like less than a half, right?
That kind of functionality you get when you use a skill. So we talk about these things, right? I think that's enough for the introduction.
So I'm going to just check quick thing with my chief. Before that, I would really like to answer. Damn, brother, you explained truly amazing again.
Okay, thank you so much, man. Thank you so much, Ethan. If I tell 20 things and pick up...
any from anything which will pick up first if i tell 20 things and pick up any from anything which will pick up first i didn't understand the question uh yes you are very good thank you so much man thank you so much shadow king i really love these kind of feedbacks coming from you guys is b tech computer science worth it in 2026 i would always say b tech 20 b b tech computer science is worth it always because i'm literally like not like i'm not that kind of person who is pessimistic about ai to be honest like i know it is very difficult and to be honest one thing i'll tell you guys like the time has that time has now gone in which if you just do computer science you'll get a job now it's about getting skilled it is about even if you're in college right for example i when i was in college right my college was in i think many of you already know about it but um it's like it's it's in jalandhar lpu it's called lpu so you know lpu gives us has given us exposure right to uh work on uh like there were a lot of talks going on and all of that right but i can tell you the syllabus of work lpu or the syllabus of any college even iit i've seen the syllabus of curriculum of iits as well they really suck i mean they literally when i was in lpu
The cutting edge technology at that time was like MongoDB and React and people were using Next .js and you know what was in our syllabus? It was like Angular. It was freaking Angular.
And then we were actually using, you know, we were dealing with a database. We were dealing with MS SQL. Can you imagine, right?
How bad was that syllabus, right? So literally I had to, and I was like, before even going to college, I was doing freelancing. And you know what?
I actually, just scrap all like because learning angular is not so difficult right uh learning even and they were they were also teaching android studio they were like ai ml but very little little right but after even if you're in college right if you're right now in college or if you're thinking of doing b tech and computer science i'll tell you one thing if you are just going to rely on the curriculum that you'll get in college you're not going to do anything because ultimately what will happen you will just learn whatever they are teaching you and you will think that Like, there were a couple of friends of mine who were just doing DSA, competitive programming on CodeChef on like, okay, I also used to do that.
I also used to love that. But the problem is that that's not enough. You need to get some technology, you need to learn some skills, which can...
create some output you can visually share right for example creating websites creating web apps creating mobile applications that was i was like doing so i think that was like taught me all of these technologies and when i was like out of my college even before getting out of my college i was earning pretty damn cool uh with freelancing itself and you know what when you go out of your college right you get a job like junior software engineer or like um very very junior roles i immediately got a role of senior software engineer right in a decent company so that really if you you're skilled right then you can literally get any job and ai is actually a lot of people says that you know ai has taken job but i would say ai has opened even a lot of ex like it has opened the whole new industry of this agentic ai look at this like you can literally you know have something new to learn i mean you can literally learn these concepts and tune your uis that this is just one part of it right this is just one part of it which is generated where there's ml core core ml as well so you can lit it has actually opened up a huge industry for you altogether and then you're saying that you're not getting jobs so that's something which
you have to work on skills can you tell me the name of the alternative of cloud artifact i forgot the uh name uh was it something like lavish so lavish is uh the uh it's not cloud artifact if since you already have a subscription i would suggest uh you use as clock pages just just tell it that you need to generate in s clock page cloud design is something else because cloud design is basically um for designing stuff and if you want to go with artifact then you can use s law page and since you already have that subscription so you can if the guys who are not on s club subscription if they want to use some some sort of uh free alternative of s clock pages you can go with lavish editor as well Just joined, have you made the orchestrator already?
And if yes, can you show me and explain me once? And if not, please tell the goal on it, your view combining two or more models together. I mean, I can literally tell you, we're already having that orchestrator in front of us.
You can see literally that that is called chief.
I tried your app, but there are some UI issues like when I try to go back from any page, it just exit the app. instead of going back uh okay so srinath can you do one thing maybe you've tried a very like long back because recently there are no such bugs and we have completely revamped the ui of it i can literally show you in front of me like this is my application and you can see so you're saying that you there are some ui issues like when i try to go back from a page which page you're talking Latest version of Android.
Oh, you're talking about Android. So guys like Android app and iOS application of s clock They are still in like very early development. I would not say early but they have certain issues So I thought you're talking about the desktop application But yeah, if you if you if anything is blocking you should not just ping me on discord.
I'll help you out Two skills in CS are more most safe sir The first one is generative AI right now. And second is cybersecurity.
Gotcha, thanks. Latest version of Android. Got it.
So, explain your orchestrator. And if you can show me your protocol file and how you're running. I mean, that's a very, very nice question.
Let me actually show you, Botman. Okay, guys. So, let's see.
Let's see. Let's see this. So, basically, as you can see, a lot of agents running here, right?
I can literally see how many agents. I can even count. 28 agents.
Like 15 workers are running. Then a couple of them. Okay, these are not agents.
actually there are 15 agents running are there any review okay one is like testing so there are 15 agents running right literally i have not pinged all of them so you know how does this how does this orchestrator works works inside as clock this is chief this is also a uh codex agent by the way sorry cloud agent by the way it is running on cloud opus 5 .5 right so i'll tell you one thing can you quickly summarize things for me and um tell me what is the latest update have you merged all the prs and started the tests already so guys what happened yesterday just to give you a background um like there were a lot of prs it opened which is good i mean when you want to ship things very precisely and always run unit test cases and always want to make sure every single feature doesn't break anything but It took one of my command wrongly.
I don't know how and why. But then it opened a lot of PRs. And I actually have to do a lot of work.
Like I was like. have to literally tell it and it automatically fixed it as well. It basically then did some fixing.
So right now I'm talking to this chief. I can have this chief on a detached mode, which I really like because when I sometime, you know, go to the application, for example, it remains here. So that is something which I like and it is communicating with those with as clock all the time.
So one thing I also have an option, like I can put this inside this, right? So let me tell you. Okay.
So I think it has done something. So, okay. So PRs almost none are open.
46 merged directly rest in this now so let's see what it what it will do right it was running a merge stream by the way like merged in literally merged in okay so i want now uh chief can you run uh all together a lint and inside a worker lint and uh just a second guys just a second i'm actually getting a phone i hope so i'm not getting Use my phone for camera.
So sometime if I get phone, so I had that's a that's a problem unless I get my new camera So I so basically I'm talking to it. I won't now achieve. Can you run all the together a lint and inside a worker?
So that we are sure about that. Nothing breaks. Can you also do a high level sanity of the application, right?
So I just told it this now what it will do if I can go like I can go in a detachment also not and by the way this chief if I'll quickly show you one thing bot man this chief has its own memory like a very very big memory like a very graph level memory in which we are using all the nice concepts of memory which we are actually teaching and that's why even if i start a completely new session of this orchestrator this chief it will work because it knows everything about shubham it knows everything like you can see shubham it knows about a noticing project about shubham it knows about micro lm that's a different project of mine so it knows everything about every single project right So now you can see, now what just happened, if you have noticed, it has literally, if I just close this, minimize this thing, it has literally started this main sanity check, right?
So this main sanity check is a worker which is orchestrated and it's started by this agent. And now it has already given it a prompt. Now you'll tell me, your question was around, how does it communicate?
What is the protocol you're using, right? That's important. So what, when I started building Ask Clock, i was actually working a lot with tmux right tmux and having these kind of setup which you're seeing right now here i was actually doing this using tmux and tmux is a very nice tool to multiplex multiple terminal eight sessions inside a single terminal session and the best part about tmux is that even if you're ssh into a machine and you run a run a particular terminal pan it just remains inside the process even if you close that it just works and you can connect to it anytime right that was the two things that i always used to like about tmux right and when i started building this chief was not in the idea so i actually started building by if i open the shell like you can literally open a shell also uh inside i don't know why i i think because of these many sessions running it's a little bit slow but you literally i have created the cli of as clock minus so when you install as clock it installs that cli and the cli i've taken a lot of reference from um like tmux so you can see it literally has these options like as clock layout it has all the cli and we have i have also created a skill which that this particular your um this
Chief knows and then it communicate with the with these agents in the inside as clock it does everything through this cli now I could have also done done same thing through mcp server but in one of my stream I explained that how and i've actually benchmarked this how cli's are better than mcp in certain areas and this is one of the area right so it literally knows everything like it can It can kill a pan.
It can attach a pan. It can literally do send a data inside a particular terminal session, it can send a prompt, it can get a feedback.
So these, all these workers that we call, right, all these workers, let me actually close this. So all these 16 workers that you're right now seeing here, these workers sends stuff back to Chief and Chief communicates them using AskLockCNI. right so i'm gonna do one thing because i think a lot of lot of like 16 workers are running right now maybe some of them are not required so hey chief can you do one thing the workers who has finished the work inside the workers just kill them because i think they are not required anymore and also i've i'm seeing that you have a preview workers preview panel like terminals running on inside preview tab just kill those as well because we don't really need a preview i've already done the preview now we'll be directly running locals on dev environment and that's how we are gonna basically do so can you literally do that uh kill all these previews and all those uh such workers which are not right now running have you completed noticing yes we have completed noticing app almost like we are around 60 70 percent uh fun and enjoy and today i'm gonna give you guys a demo about that so i mean yesterday also i've given a demo it was pretty amazing today i'm also gonna give a demo with a different use case and by the way it is live on app store play store it is just
right now in invite only mode but i'm gonna make it public very soon like maybe today we're gonna make it public once the payment integration is done so you can see literally it has closed all those workers it has closed all that preview right so that's how it integrates with it why is the chief not sending comments on stream today have you switched that off what kind of comments ah okay okay yeah i turned that off that was like pretty hilarious so guys like this chief not just uh it can uh do this management it can do other things out of the box like literally this chief has built a complete content pipeline for me i've shown that to one of my stream this chief is so powerful Are these vector graphic memory?
So it is combination of vectors also. We are using semantic recall in there. We are also using a little bit of like working memory and then abstract memory as well.
So it's a combination of vector memory. By the way, I am actually moving it towards more and more graphical graph memory. So that's kind of progression is still going on.
Are uncensored local models capable of connection in these? See, first of all, whatever models you can see right here these are frontier models from cloud and no model stops you to run command it's like you know running commands on local right even if you're using local models for example you're using open olama and you're running let's say quan kimi k3 on your local machine they should also be capable because ultimately what what this chief is doing it is literally running commands on my system every single model can do that right okay bro sorry can't watch uh your life because i'm building something but i will always soon as i definitely man fun god 795 you're the like one of the viewer who keeps coming regularly even if like you're not getting time that's totally fine i'm so happy that you're building something keep going like keep up keep it up and make your product live as soon as possible i don't know how you're handling the cash of for ass clock and how much it is costing on into you okay so thing is that
i do not literally have to manage cash for asclock as of now because the reason is that And you'll see that in the stream. And I think that you'll notice that also when you're actually, because since Aditya is a paid subscript, he has a paid subscription of AskLock.
So he knows that. So you'll get an idea. So when, whenever chief assigns any task to, let's say we are working on noticing application right now, whenever it assigns any task to any of the worker, it starts a complete fresh context.
It gives it up like whatever information, right? And whenever like chief wants to know about something. inside its memory literally i can just literally compact this right now but this isn't 30 maybe i'll compare around even if i compact it uh right now since it has its own memory which it can fetch programmatically it's just fine for it because it literally ha doesn't have to load every single thing in the context itself right so we don't really need caching over there as well right and whatever caching logic these agents because these all are basically cloud agents codex agents whatever they have by default that is something which is uh like happening back but yes the application that i'm building which is noticing application which i'm going to give you guys a quick demo and that's pretty amazing you'll really like the demo of that first of all i'll actually do one thing i'll actually actually make sure that every single no uh thing is running on my local so you can see even my local servers right when you use like just normal cloud right what happens inside that when you start a cloud session and you keep you tell it for example okay so i'll come to this question so when you run a
Normal a clock session and when you start it to you know, can you run my application on local? So what normally these agents does these workers does they actually? Start inside their background shell.
So you must have noticed something like this You can see here here one of the this agent is doing something inside shell, right? But it's not local server for them But the bad thing about is that you lose control over those local servers, right? And what I prefer like for example Right now, I also want to run this noticing application locally.
So what I can tell Chief directly, and it is, by the way, it's already running, but I'll just restart them so that if anything is wrong, so it can fix. So I want to do one thing, Chief. I want to run all my things on my local, all my servers.
I can see inside noticing local tab, you're running all of those two different tunnels for database and clam antivirus. But I want you to run all and restart all of these servers. and make sure these are running on main branch so that we can do a quick test of the application right and i hope that sanity checks also going on so this is how i interact now it will do what it will do if i go to this noticing this this particular tab now it will literally uh restart all these for me so now the best part is that i can see my stage tunnel if anything is getting locked there if i'm getting any issue i can see my tunnel is running here stage db by the way i'll explain what this tunnel and why do we use that and then we can literally see my api mcp auth my noticing web which is the web version next .js version my api of our attachment worker and my main noticing uh mcp and i can see literally all of these agents all of sorry all of these local servers and i can see if any log is coming any error is coming i can literally see if anything got crashed and all of these workers can literally access all of these through s clock cli once again whenever they want to
So, can you explain what's the difference between KV caching and Redis caching and which is better for which use case? So, KV caching and Redis caching is kind of very similar because Redis also fundamentally use KV caching only. So, KV caching is basically key value pair sort of caching and Redis also stores it in caching.
It depends, right? So, like ultimately Redis also same. So, for example, if you're building your front -end application, you're building your like application around your uh web application or mobile application and you have you're seeing a lot of latencies uh bad latencies around your databases because every time let's say for read example is a read query every time a read query on your database is taking a lot of time and creating a lot of load on your database you and you know that that data is not getting changed very frequently you can put a redis cache in front of it if you're having let's say you know you're like for example if i have my this uh local application i have my local application and i know that right if i want to ship it to other users right i why would i just uh if i have to use redis for example right and i know that my application is client side more client side rendered more i mean this is literally a client application desktop application even the mobile applications are client right
If these are on the API side, you should go for Redis caching. Otherwise, you can go for KB caching. Now, KB caching has many libraries, many ways to do that.
So, that's what my recommendation would be. How is AskCloud different from Paseo? I don't know.
I've not used this Paseo. I've used Orca. I've used T3 Code, but not used this one.
So, I'm not really sure how it is different. Maybe one thing I can tell you guys, this chief kind of experience that I'm sure no one is doing. everyone is trying to build orchestrator but the idea is different like the engineering i've put into the memory part of it the memory layer i don't i'm not sure if anyone is doing that and the especially uh doing it through s clock cli or their internal cli is literally different so i've seen orca also does orchestration but it it installs a lot of skills onto your and literally directly communicates to pants not not using their cli so i'm that is what i'm doing different and that is why it is fast you can see like i trolled it to restart it just restarted right so one since i have done a lot of uh like i said prs yesterday i'm not sure whether it will be literally working right now so i'm gonna just quickly do a like a smoke test and i before i i do that smoke test i want to just let this a one of the agent of maybe a qa tester run the smoke test in front of us right so we can see that
at least couple of things are working because we have literally merged around 78 prs guys and this chief was able to build a complete there are huge list of functionalities not really a very architectural level because we already brainstormed around rich text editor we already brainstormed around which uh like uh editor library we should use which is we've gone for tip tap editor which is pretty nice we also uh brainstormed about a grid library because this Grid is something which is very important.
So it has finally decided the 10 sec tables. So the PRs were not like very, very critical in terms of architecture, but there's a lot of work that needs to be done on. So I want to do a quick check.
I want you to make sure to start a Qubit testing agent, which can literally do a complete smoke test right in front of me of noticing application. And I can see at least the main functionalities are working.
Since we have merged a lot of PRs.
How do you designate a specific portion of a request such as system prompt, tool definition to be eligible for prompt caching? See, when I am building this as an application, prompt caching is very, very important. But you need to be very, very cautious, guys.
I'll actually explain this through an example. So when you should do prompt caching, right? So it's very, very important to understand what is prompt caching, first of all, right?
okay i think it's gonna start a smoke test so i'm not sure so it's gonna i think it it might have started the or it is it has not yet started what's wrong so let me actually see i've started a live smoke test worker uh okay but where it runs visible chrome tab you can watch it when as soon as the local servers report unhealthy on main i think right now something is breaking is it like what is breaking Is something breaking on local right now?
Can you confirm that?
Okay, so I'm going to give you the explanation around what is prompt caching. In general, how do we do it and how do I do it? How do I think about it?
Okay, so first of all, one thing in your comment you mentioned about like Tool definition list to be eligible for prompt caching. I have a little different Idea around it, but I'll first explain what normally You know when we should do prompt caching and how do I think about it?
And how do I do it? Especially how do I do it with maestra? Okay, so let's say you have a like you have a like you have a really long prompt.
So you have a really long prompt I actually do Yep Okay, so now that if you let's say if you have let's say an agent you've created an agent, right? Maybe just you have an LLM, right?
You have a model, right? So if you send a huge text to this prompt, so if you send a huge prompt No, it can be a prompt basically literally literally when when we send a model It really don't identify it as a prompt a system prompt. Maybe it identifies but you just specify you send a system prompt and then if you have like within the system prompt you can also send like if you want to load some memory around it maybe when you do it with your cloud agent you can literally do it with using your memories and then you you send uh if you want this is by the way optional memory some some sort of memory and then you also send let's say if you're a user prompt right and you know that for some fact right if this this is very huge For example, this is very, very huge.
And if you send that to any LLM, right? Any LLM. If you send that to any LLM.
And that LLM will respond with you some response, right? I'm going to basically say. I'm going to actually make it like this.
If this is really, really huge. If this is really, really. I mean, it's basically like 500.
Like 200 or maybe that will be too much. But maybe just 20k tokens, right? Let's take an example.
If it is all combined as 20k token, right? And. when you have when you basically are uh sending this to any model for example you're sending it to gpt astra let's take an example of that right it's going to respond with you some output right basically output tokens you're going to basically send some output tokens to it now that if you know for a fact that every single time, the same sort of system prompt, the same sort of memory, and the same sort of user prompt, you know for a fact that, let's say you've sent a request number one, right?
And you got some output, right? And you know for a fact that next time also, if I send the same sort of prompt to the same model, maybe there'll be slight changes in output tokens. but mostly the essence of that is going to be same only right if you know this for a fact right and i think it's quite understandable for example if i'm sending uh remember my name shubham and then i'm sending a prompt that what is my name 100 this is going to be sending the same thing right same if you know that this is not going to change any part of it for example even the even this part of it is not going to change this part is going to be same only if you know that then you can literally Instead of sending that prompt again to the model, you can literally store the previous output somewhere in the cache and then can use that.
Now, Maestra does that by default. Maestra has one way, like Maestra prompt caching. I think the way it does is send some additional to the provider.
So all these models support prompt caching at their end also. So you literally have to just send some configuration. Just a second You need to send some Configuration, let's say cache control equals to mferal something like that in order to do that Let's actually see that because that's what I do because in most cases when you build these agents There are very less chances where you literally want to do prompt caching because I I don't have much good experience in it definitely saves your token, but most of the time it just you know it's it's better to just always get those token tokens but i'll show you how does we how do we normally can do with maestra and any other so it's basically some settings inside model itself so uh maestra m options for prompt caching let's see let's see let's see i don't know why they have don't have this in the documentation Yeah, you can see this thing right you can see this thing system and here we are using clouds on it model and then we are setting max token to 1 0 2 4 and then we are setting system prompt system and then we are basically sending that this type equals to M Farrell so this basically literally tells that you need to cache this particular part of the your Input token.
So next time when it will basically try to use that it will just Send it directly from the cache. So that's how you can literally set setting these settings in maestra and The second part of your question, which which basically was around Tool definitions, so if let's say you have MCP tools, for example, right if you if you're talking about MCP tools Then I have taken this example also So let's say when you have around 11 to 15 MCP tools for example stripe has that even 11 to 15s are fine, but if let's say you have 20 -30 tools, tool name, description, input schema, so every single tool has these kind of information.
What is the tool name? What is the description of that tool? What is the input schema of that?
What is the output schema of that, right? And when you literally have like 30 plus tools, right? And every MCP server, the way these client works, they actually have two operations.
One is listing all those tools, I mean, getting all these details. And then at the same time, calling up a specific tool let's say for example we have a tool inside that get invoice details in a stripe then we basically have an option inside msql server to call that tool but in at a very state when when it has to load all these tools if you have huge number of tools like 30 40 50 100 they're going to load all of these tools in the context and that is going to increase your context now you are you can fix that context i mean literally i've not tried with prompt caching if that can help I may try it someday but what I normally use I create this remove all these 30 tools from my master mcp server and only use this one single tool which is search tool which in the behind the scene use rag to basically programatically find out which is the not very close for example user has sent I want to get detail of invite right so if you have already loaded these all that's 30 plus tools right you can directly call this that is one way but that is basically going to increase your contact because you're going to load all these tools on your system prompt but the other way is basically you know you just have one single tool so no matter what is going to happen if it is related to invoice payment you can just have a higher level description of this search tool
That if anything related to stripe all you need to just call this tool so that will call with the same so I can send the same sort of Initial prompt that I want to get that invoice details of this invoice ID 1 2 3 It's gonna first do a search and that's gonna do a rag around those tools that These are the tools at the closest match and when you find the closest closest met it uses Algorithms like cosine similarity all the drag algorithms right to find out the nearest match by the meaning and the semantic meaning and then it returns that tool and then it basically calls that tool and return back so that's how you can save tokens in terms of usage all right so my two cents uh kv caching is something which is which you don't do at application layer it is handled by interface engine and uh for prompt caching red is more uh specifically reddish lang cache is used for semantic cache okay that's fine so see the idea is that like i said the prompt caching is something which
these models handles by themselves but there are ways you can also do your application level caching right you can also do your manage your cache caching at your end specifically for reddish i'm not sure about this reddish lang cache what you're talking about but we use reddish a lot because in your application layer you can literally use reddish for example currently all my apps for example if i go to this noticing application if i just actually open that noticing application for you for a second if i literally can open that i don't know where it is doing that smoke testing by the way but yeah i'll show you that so let's say for example this is the application right so i can i can if i inspect this uh api right so if i have this api called list api oh by the way it just have now one api it has done all those performance optimizations so for example this list api right this list api returns all the list Okay, right now it is not returning anything because it has literally, yesterday it was returning a lot of things, a lot of lists and calling multiple times, right?
So this list APIs does basically fetch all these lists and there is very less frequent chance that these lists will be updated. So what we can do, instead of calling this list API directly, we can cache this list data inside Reddish at the application layer. I mean, the way it works is basically you have a database, let's say here, and this is pretty standard concepts of So you have your Redis here.
I'm going to actually agree because yesterday I was watching my own video and I was seeing that for the stream it is very, the zoom level is not so good. So let's say you have your Redis right here and then you have your database, right? And I know for a fact that database is the most critical part of any infrastructure, right?
So for example, I have this inside my database, right? Inside my database.
I have this table called Let's say list Right, I have all the list right so on the from the front end So from the front end, this is my front end application. For example, right if I'm calling an API to my database, right? So my back and literally does what select a star from?
From this table, right? This is what it normally do it typically do right Select a star from this table, right and then what happened? from this table right and then it returns the list right and technically if you if your database are so busy if you're getting a lot of hits then sometime it's good that if you know that for a fact that this list is going to not going to change much so you can literally store all these lists let's say list is one two and three you can store that and instead of your system going directly here it can literally get it from cache it can literally get it from cache When it knows that you know now this a new list is added because then this cache is kind of Invalidated then it can literally fetch again from the same location.
So that's how typically you can actually Build your system, right? Yes, yeah Redis can be used for normal key values cache which all traditional traditional system use but you can provide your semantic caching for that use uh we use ready slang cache that's awesome like i if if if you are able to use redis for your semantic cache when i when you say semantic cache uh this guy varma ankit can you tell me like um you meant that you're doing some rag around uh like semantic rag and then you're fetching that information then you're caching it using redis i mean that can also be done that can also be done for sure but if that's what you're doing that can be it's very similar to db no right here we are replacing that db in now with a vector database okay let's move on example if you ask what is the refund policy in how many days i can i understand your point now so basically you're saying you ask we send a prompt to our uh our agent what is oh yeah so yeah i understand your point now i understand you completely understand your point basically it's the same thing only so if i replace this with vector vector database so
you have let's say you have your company documents you have uh like created chunks and created the vectors of those right and then you're doing uh search uh like semantic search around those semantic matching around those using maybe pg vector milvers quadrant any any database you've used for vector database and maybe you also have used some maybe not for sure you must have also used some vector embedding and then you're doing some semantic match so instead of doing all those semantic search every single time if somebody already asked in some employee already asked that um how many uh what is the refund policy so you will just store that somewhere inside your redis by using lang lang cache and i i believe this redis lang cache must be a a plug -in for lang chain or maybe the lang graph and then you you have used that and that have solved your problem it's the same thing the concept is fundamentally the same only understand your point okay thanks for uh letting us know so now that guys like i will go to go back to chief
I'll go back to chief and I'll see what's going on there By the way like Where's chief let's click on chief. I like to get it back But I guess I think I have to some like I think I have to Have a button somewhere that can bring this chief to me Okay So let's see what it has done.
A live smoke test is starting now. I would set it to wait for the strat to work or to report. Okay, it's actually started this live test, but I have not seen that.
Password for each step and gif at the very end. It started the local servers. Okay, let's actually, let's, we, I think we've already seen that application running live, right?
So let's actually check what all changes it has done from after yesterday, right? so let me actually press try if my uh like model is running by the way guys here you might see this model is slow because to uh since i do a lot of work from my mac mini also and i wanted to find out for infrastructure where i can literally uh have a stage database inside my Kubernetes cluster, I already have a Kubernetes cluster so I kind of wanted something like where I can literally like for example this is the kind of current architecture of our local database also which I am literally hating maybe I'll just go back to the local database so what I'm trying to say initially I used to have my Postgres for my local development on my machine inside docker which is the best practice by the way if you're using working on multiple teams you should do that but Since I sometimes work from my Mac mini and as clock gives me that functionality that I can connect my Mac mini office laptop every single place, right?
So I thought of why not just using a common database because we already have a kubernetes cluster just hosted inside that and but I'm actually now the one way is very which will enhance the latency is by Literally opening up a port on this postgres um, maybe on five four five four three two or maybe if you don't want to open a port you just uh link with the link reverse proxy that with a domain let's say for example pg stage pg dot noticing .com that is one way by the way but the other secure way which adds little bit latency is using tunneling so you can literally you reuse your 22 number port which is used for ssh using ssh keys you can tunnel a port on this device but that tunnel adds a little bit less in latency right so that's why the response are a bit slow here so it still remembers my name is let's change this name so from now okay let's use its own transcription let's actually not use this transcription here okay from now onwards uh you should know my name is shubham not aditya one second guys one second
damn my memory is gonna go now okay so now onwards you should know my name is shubham not aditya okay i think it's it's better now it should understand me as i don't know what it is filing so maybe it is storing updating that in your preferences table also so the idea of this noticing application guys like literally you can just talk it to anything you can store any information and see how slow it is literally like if i show you the network calls by the way these are in milliseconds on prod so you can see they're literally taking for a small read one point one second but i'll fix this issue because i'm not really liking i would really go with the back with the Maybe I'll open a port on that.
Let's actually do that today. Let's actually fix this latency further today by using a poor opening a port or maybe Reverse proxying it to a domain. You'll also learn about that and in that way and also will reduce this latency as well And by the way initially it was like 15 second because I will like we misconfigured our kubernetes that it was first going to Germany To Mumbai then back Germany to our local which is also in India.
So that's what that was really really bad Okay, let's go over to the application so now that i'm going to test a new different use case of noticing first of all i would really like to see if it has actually changed the experience of our data i think it has i am it's giving me that let's actually go to expense table now oh i think now no no no okay no no no i i think it has seen the experience for sure uh okay it is showing some here i can literally now edit these things directly so it added those functionality though these all things are not yet tested by the way and we have a view option also like we can literally add a new view here and i don't know why but i think that is something so all of these i'm showing you right now nothing is basically tested right so we can literally see here and yeah i mean this is good i can literally uh go to that table also link table also i can see the data inside this so this is really storing data is now nicely and let's actually go to okay this is what
this is ren's memory for life oh actually it has it's it's uh you know this like main working memory here so i think now if i go to um any packing let's say let me actually see the image one i think we've stored that inside what inside expenses only that was a 3000 yeah this was this oh it has an image also nice nice nice nice perfect so yep i'm gonna actually do a different uh scenario test all together and then by the way we also have a mobile application for this i'm gonna actually see if it has changed things in mobile also so let's actually see that okay again this is not zoom so i'm gonna ask my chief so where's my chief okay here it is i'm gonna open the mobile application right here so chief again the same problem i can still see i'm not able to See my mobile.
Can you also make sure first of all your the mobile application is also running on main and main branch and I can literally I can literally see that and I can see that in a zoom view still I'm not able to it's coming very small for me. So meanwhile it fixes that thing I'm gonna actually go back to my This thing so we are gonna do up this proper QA testing guys because whenever like we uh do it's like a big implementation literally you can see that you know it has like and by the way we can also explore the rich text editor it has all these fields option now so you see how messed up it is because we have not done the qa so don't like think that it is something like broken this application we have done a yesterday complete revamp of this these grades and all of that so that is why it is not looking that good so obviously needs to needs to be improved and by the way we have this like uh okay we have this can we update from here here from here we cannot update it's like okay so it's taking me to the chat where it actually has i actually so it's like that cross relationships is really good in this application but it's completely messed up in terms of ui for now and especially for the data side okay i'm gonna fix that uh but before that i'm gonna just test one thing yeah itinerary i want to test one thing
i wanna can i create a data structure completely from manual so let's try creating a table okay we can do that so let's create something called my what is called so deployments let's let's for example deployments and then i wanna actually oh it has a start from option okay now this is nice let's say uh release as gonna be text and this is basically um release nodes maybe release nodes alexis we call it releases why not the hell so these kind of bugs it is having right now but it will eventually go because like i said it is not qa tested yet i'm gonna do a one fresh round of qa testing around this and then i'll test and then if i find any bug i'll fire that so the best part is that when i fire those bugs through uh chief it also just gives some additional information to the qa tester or the agent first reproduce that and find related bugs also so that's really nice so i'm gonna just do a fresh test so deployment what the so it's basically completely taking me away this is a big biggest bug i think i think i'm sure that cuba tester will also find this bug deployments
and then one will be let's say uh like i said release release one will be release nodes and i would really like to test the rich text editor also so i don't know from where i can do that i don't have an option for rich text editor more types or we have an option for um we have an option for long text let's try that and then release nodes and then i'm gonna have a release date I'm going to make it a date.
Let's create this table. So by the way, I do not have to do manually these things. I'm just testing out this functionality.
Nothing else, right? It doesn't call calculate option, by the way. Okay, that's nice.
It started copying a couple of things from Notion, which is good. This is very bad. This UI is very bad.
Oh, it's actually completely messed up here.
Can I just edit it? oh can i edit it here oh i can edit it here oh i think this has messed up here completely i mean hey it it should be a rich text editor though okay i think it's it's messed up completely i'm gonna get this fixed right so let me actually first prompt um this uh chief around this and then we will uh do the test around ai agent which is the most important part by the way so i'm gonna first do a quick test around this so let me actually do one thing let me actually open the as clock and then see what is going on smoke if the smoke test is going on manual edit editing parity so let's first ask what are these agents right now doing can you quickly tell me because i just did a test and it really a lot of things are failing so can you tell me what manual editing parity and what are other these agents which are working are right now doing and you can close the agents like which are done like for example restart local servers merge train all of those
so i'm basically asking what are these agents doing and then i have to start some dedicated qa testers which will basically do the qa testing of the application by the way when we will be doing qa testing i need to kind of tell it that you know do it in the background because i'm doing live stream i also might be testing myself so that might you know set up break scenarios in bid so the testing will not be around ai agent from for from these ai agents it will be most around the manual functionalities awesome and i have not even touched the so let me actually see if it has ran the mobile application okay it's still not ran why is that this slows okay running the app on mobile so i think trying to run that simulator is still on the main actually minimize this one now but literally guys we are very close of this application to be in a very nice shape trust me it is going to be a very very valuable application so it has now given has given me a brief summary
that uh one manual mode on web and one feature going through every manual screen against your design and fixing what's broken ah it's actually doing the same stuff right the thing that what what i have done and i have to do it because i did not have any other option is because i literally have to merge everything inside a single pr because they have done a lot of parallel work it has created multiple like 78 prs guys can you imagine right can you think can you think about having giving instruction to these agents manually to different 78 agents and then getting the work done so that's where the chief's value comes but by the way that's a problem because i should have actually added some instruction that you don't spawn okay you spawn divide the task but divide at a feature level not like completely every single for even for a small test column let's say there are 10 test column types it has created 10 different agents and raised 10 different prs that's good for some certain cases because unit test then it can write in itself but not always good so uh but then i told it manually to merge all these prs and that is what it has done and now it is basically doing a full testing around it which is which is why the issues that we are seeing might also go away and then we have a mobile also like mobile on main
the same job it is paused it ran out of usage account resumes at 1 10. so i'm gonna just tell it that resume because i've already switched the account and then we have a link checker i think this is also important all of these agents are kind of important so the agents are stuck due to the mobile usage uh sorry usage you can tell them to continue the work because i've already completed that but i would suggest that let's focus on web first and get the work completed and it should speed up the work because i can i can see a lot of issues out there um not really like too many but i think a proper qa testing needs to be done around web that's how like a client i this is like my project manager guys like i literally speak it like my project manager it does everything behind the scene for me and that's cool about that's the cool thing about uh this chief and it really if you are watching my stream from regularly from last week up to here you would have noticed that
see within one or two weeks we are able to build this application completely from scratch and we have done a lot of brainstorming around it right so my focus for today let's actually set focus for today because we have already we've already spent a lot of time but yeah i'm gonna set a focus for today because yesterday as you know like let me actually go back to the yesterday we created a mcp server and then uh we la okay launched payment uh okay payment is something which i need to still launch we need to still discuss about notification and reminders brainstorming workflows i think that's this this part is actually done you can remove that actually remove this also remove this also this part this section i think this section is tied with this i don't know why okay so mcp is done this part is done okay by the way mcp is done but this search tool implementation is still remaining i mean the test is still going on so i'm gonna actually have it like this that mcp search tool only let let it be like that so today my goal is guys like we already have payment integration but i think there is something we have not tested and i'm gonna actually make sure that the payment happens and we actually create a credit based system so that is very credit based system
based system now also we also have an admin panel to track and see my system is getting slow and manage applications like manage our users in us in a way manage app right so manage app whatever whatever say so admin panel is very important the security of this admin panel is also important so mostly i just uh give access to my plus personal intranet and my vpn to that yeah oh yes i think now properly these agents are actually doing the complete web testing and meanwhile it is doing those we can do what we can actually uh let me actually open the mobile version first let me see one thing if we can do a quick check on mobile also i think it's not running so let's actually close this because i think that's going to consume a lot of memory and compute let me ask the chief to do that uh so chief let's do one thing let's focus on mobile later on just stop okay let's keep do one thing let's stop this mobile worker as of now i want to focus more on web first because once the web work is done that will be great and i'm live streaming i don't want my i don't do not want my system to lag so for now just kill the simulator and uh it's metro tunnel also
so guys we literally need a metro tunnel in order to communicate with react native application but i think for now i'm just going to close that because i want to focus on what the web's web version of it but i also at the same time do not want it to take a lot of time though this is important so we'll be doing that that will happening in the background meanwhile we'll be doing some chit chat on discord so you guys can join discord i think aditya is already joined Yeah, so I think you were I think you're there Guess tell me one thing when I did and me are communicating There shouldn't be any echo Coming to you guys because I think with this kind of setup echo comes sometimes So just make sure that's not happening When you are able to hear are they told so let me actually reduce the music value even low So I did there has a query.
So by the way Guys, if you have any projects that you want to get reviewed, drop those inside Discord right now. I'm going to review those.
And meanwhile, this testing is going on. I'm going to basically, let's say, I think it would be great if we can see it testing live, right? But I don't know why is that not happening.
Let me just quickly check a couple of things. Where is it testing? What approach?
So let me actually go to that particular agent. put it back in the s log okay so i think this is the one which is doing i think it's already uh trying to fix that so it's trying to fix that from code from first then it will do that so all web tests green so it's kind of like it now it now will give chief that kind of message that it is now all green then we'll do okay so what was your uh question and what was like what do you want to wanted to know So normally if I'm taking an interview of, depends like if I'm taking a front -end interview or if I'm taking a like just, which interview example can you give me?
. . .
. .
And after that, it will directly say that there is a bug here. You can present it like this.
Basically, you will follow these steps and this should be your thinking. And you just keep reading it like this. I am thinking that I should go to this file, I should do it like this.
And it will just make your process a little easier. Okay. For this, I asked that codex and like cloud etc.
You have done direct integrate in your slog. Yes. So, how can we make it connector type?
See, you can have one of the CLI installed and that CLI can, you can have your project CLI on your desktop application. That is somebody if install that by default that CLI gets installed, right? So, for example, if you have But you will actually get to a problem, but I'll explain that, right?
So let's say you have Gupt AI. Okay? And then when somebody installs it, you can do it like this.
One Gupt AI. This will get automatically installed. That's what I do with AskLog also.
So when you install AskLog desktop application, this gets installed by default. Now this, you can have many options to this. but the problem is that if you want to communicate with a terminal then you need to get like one way is you just go with like the way computer use does computer use what they do normally uh they get access to your accessibility on recording those i think that that you're already doing right accessibility and accessibility and that recording access you're already asking for yes right so there are ways in macbook i think you need to do a little little bit of research which can literally open terminal for you the default terminal right if they have something installed like ghosty it will be really helpful because ghosty you can do a lot of things but you can literally install open terminal uh and run a command on their terminal and another way if let's say you don't want that if they have terminal or not you want to make it generic but you can also do like which i'm also doing with asclock asclock runs a terminal terminal emulator using this called technology called xterm so we i use xterm and all these things that you're seeing right here right so asclock cli i definitely have
but these are actually the terminal instances that is using xterm and it is actually running directly on your kernel so you can literally but your application will become little heavy so you can just have it like you can run your xterm emulation on top of your application and then inside this xterm application it can directly on your unix it can run the commands it can literally run any CLI command For example, I say that, you know, create a react to do application.
Now tell me there is a bug in that. And let's say we are using global storage. Let's say we are using redux in that.
And in that redux, we have a bug that this redux slice is not working. So how do you do that? So now what will my AI do?
It will listen to you. It will make a query on it and it will only do this much. It will make that query a prompt and directly send it to AI.
Codex or whatever you are already tooling. And nothing else. So it should build a query and go.
But the interviewer will not know. The interviewer will not know that he is cheating. How did this prompt go to him?
How will he know?
Guys tell me one thing on the chat if the voice is not echoing. So we are discussing about Gupt AI which is created by this guy Aditya. And nice tool if you are preparing for interviews.
What is the domain of that by the way?
now tell me one thing can you debug the okay i'm giving you this code which is having some bug in it download this code on your ad ide maybe vs code and find out the bug and fix the bug issue in the bug how do you do that because that he will be screen sharing and he will expect expecting that you go inside that file and change that code that will be my expectation from as an interviewer and that's by the way when i am taking interviews these days guys i have literally first of all most interviews in my organization they are actually doing now in person and even if they're because i work remotely so when i do i take interviews there are a couple of rounds i'll give some interview tips also to you guys so what we are doing we are actually in very starting telling them to open their camera to you know make sure they are not connected with multiple desktops they are actually uh not that they have to share their screen and we also make sure that they they do not have uh tools like team viewer and index install so that's also by the way i'm gonna share one incident of one interview with you guys um which caused a company six billion dollars i'll come to that but before that let's actually finish this topic
Okay, so Aditya, I'm taking your interview. Let's say I'm taking a React interview or a front -end interview, for example, right? So I have given this piece of code to you in which there is a bug inside, let's say, Redux layer, right?
We are using Redux and Redux has some bug in it. So what do you want this Guptai does to you? So what Guptai will do is it will just listen to the question, right?
Yep. it knows what we actually have to do so it will build a query and it will just uh send that to like connect that to like codec cli and ask it to like search in the folder if there are any files related to this like this solution that i am thinking of right now it will only go with the read only process okay okay and then we'll then we'll guide the user key fix this way yes right so that's the i i got it so you just want to read the files right yes and what but it could be the case that there are no codecs and clouds cli installed on the system user system yeah so that's why i think that it's an optional feature that if a person has this he has an advantage yeah then then what you can what you can do what you can do you can try if you can best way would be if you can literally without installing xterm xterm is the best way you can have a small lightweight xm xm is very lightweight by the way you can have xterm installed by default and xterm can run any commands on your unix and by the way if you don't want to use xterm i think one way would be if you if your guptai cli can literally run commands directly on terminals so this guptai will use this cli
and through the cli it will turn all those commands and uh it will navigate to that particular project and maybe you just don't even need good up like you can even do this without the cli your guptai will directly be running commands on terminal if you if you can do this that is something which uh is going to be uh doing no so the problem here is that uh so the architect of gupta is like right now it's like currently like for free uh like if you use grok api you know that in free mode you only have like 12k token per minute limit right so you can't just do like searching the code but i am saying i'm not saying using so your tool can use codecs only codecs or cloud only but the only thing is that instead of having a cli in between it will basically i will directly run that codex command so it will basically run that let's say uh maybe the project folder is let's say you know uh interview let's say you know test app test app so it will basically identify that test app for you in from the text if it is inside desktop it will directly run command that you know cd test app something like that on the terminal maybe if this is possible i'm not sure because i use a x term for s clock so it doesn't work like that so
you can just literally do cd set test app and then you can run a codex inside that and send a prompt like minus p equals to the prompt and then it'll once this is done but here you will need that how do you how would you read that from that terminal right for that i think you will need a cli so you can have a guptai cli in that case because you will somehow need to run that and i think The best part would be to go with Xterm only because Xterm can literally do everything Xterm would be the best way for you No, it's just free it's open source you can also use ghosty but ghosty would be overkill for this So Xterm is like like you can see in my project as clock you can literally see this every single pan are x terms only right they are performant i also tried ghosty ghosty is even performant but there are problems with ghosty it's like over engineered so with x term it just works pretty fine harsh they were saying what is guptai like what are you building or what are you guys building so uh harsh i like uh there's a there's a viewer inside our discord channel right now
Aditya has helped build this group there which is a real -time like meeting assistant assistant and I am actually building I've built this as clock where you can manage multiple agents you can connect your multiple machines like Mac minis and you can access all these terminals from your phone Also, so I built this asklog .ai and that's how I'm actually creating an application which is basically a second brain kind of application like noticing which is like a memory application.
So yeah, that's what we are doing right here. So I think it has it's basically doing I'm just going to ask a quick update. So the way I communicate with the asklog and all these agents inside this project is just I just keep communicating with this single chief and this chief manages everything for me.
So this chief is an orchestrator that I've created. so can you tell me the quick status of these uh how many issues found like what are the test pass what are the lint issues i think it's already giving me i think it's already giving me lint is passed type is passed api tests are passed shared contacts are passed web test failing i mean literally because you know we merged a lot of pr so yeah that's why i'm gonna just actually ask it again so can can you just tell me the quick status and create an as clock page where i can see a real -time stage status of all of these uh qa bugs that we are finding that failures that we are happening because i can see there are a lot of failures so now what will happen guys this will create an asla page maybe there's an artifact which aslog uses to visualize and communicate and i want to make sure that you are listening on that um aslog page so you can see it uses skills in between right that's what i discussed with you guys that
skills are very important that we use those skills extensively with uh with our s clock tool in order to um basically make sure that we are saving some tokens so skills reduces token usage right okay so meanwhile this is happening why not let's if someone has a project that they want to review guys like this is the time you can send your project in project showcase i can see one project inside that i'm going to open that i think it's shivam is there in the stream shivam are you there i would really like to review here but i'll only review because last time also when i reviewed a project you were not there so if you're there i will review your project right away right now otherwise we'll review that later when you're here when you're there i think he is not there so we'll review his project later okay so next thing guys i can i want to do one thing inside this noticing application uh you know what we don't have a landing pay in the in the land since now it's already live i think someone asked something i mean actually i missed some message uh hey can how can i try noticing you will be able to try noticing in one or two days it will be live for public right now it's only available for the users so yeah we can try we can you'll be able to try that after some days
There's a user hi, hi, welcome the stream user pH 6 w e 9 z, whatever Okay, so we are gonna add a pricing page here and the credit management this is very important guys when you are building So I'm gonna actually let let actually I'm gonna ask this Chief to do that for me 17 open 3 fixed 3 being fixed 31 fixed today so it's kind of like finding the bugs automatically and fixing it that's good and it has created a live as clock page for me can you open this page on the browser for me for these small task calls i'll just i just let chief to do things for me all right it has opened so here is the real time progress this is called as clock pages guys where you can see and you know you can actually i can literally uh see what are the bugs notes edit don't save reliably save save conflict so all of these bugs basically are happening we can see these are the uh status of these agents running by the way i mean it's actually created a complete chief sort of like session and this is fixed this is kind of running and it'll keep fixing these issues and it is finding right so meanwhile we can actually start with something called uh so i want to know
i can see we don't have a pricing page right now on noticing landing page can you start a worker and start working on that and i also actually want to evaluate what is the payment status i mean is the payment already integrated and also i think we have already integrated razor pay and it's right but there was some problem so i would really like to fix those issues today so i want you to start a worker which can analyze the current state of all the payments related stuff and then we can work on that we can work on that and also i want to understand how is the credit system going to work like for example someone's credits are completely uh they have got uh consumed how will they how how will they be buying new credits is that system ready or not if not let's actually start working and evaluating on that so i want you to create a ask lock page uh which will research around the existing payment stuff and at the same time it will also start about this uh credit system
so this is how i give my prompt to this guys like uh like a completely vague prompt but it is fine because it is chief it has its own memory about the project it knows what's going on you can look at the look at the memory how this memory is growing with time right and by the way this memory does not get consumed every time on this chief it it has a very smart logic when to use that memory that is why you can see that its context doesn't get explored with time it just takes time to fill And when this chief starts any single agent, right?
It's very important. It automatically assign them a work tree. They use in their independent work tree.
It automatically send them dot whatever ANB variables are required from my like one or like one pass. And then it does everything out of the box. I can see it is calling as clock commands.
Awesome.
amazing so guys like a lot of new people joining the stream thank you so much for joining the stream i urge you guys to subscribe in this channel we are sharing a lot of knowledge around uh generative ai we're actually building an application so that you guys can see and by the way tomorrow the day after tomorrow we are thinking of doing a stream all from like uh in the single stream long stream maybe a little bit gaps in between breaks in between and build a complete application end to end i'm i'm just thinking about which day i can do that uh because as libraries are sorry diwali is also there holiday holidays are also there so i'm just thinking which day i should do that so you can literally see right what it has done it has started this agent which is basically one one it has started worker for creating the pricing page right and actually i want to say okay get the ui done from um cloud uh artifact right use chat gpt uh worker to access the cloud artifact because it is on a different cloud account so use chat gpt to open that in as jagudi profile and that profile access uh like but i think let me actually see if i have a token raiser on my another account give me one second guys give me one second i'm gonna just quickly check one thing
i have i think token reset session because in the the the account that i'm using for i think it's not it's it's going to be reset at 1 140 am so 140 will be okay i think that will be too late so um hi hi hi again uh ak whatever like user ph welcome to the stream okay so i think uh let's actually just see let's see what happens like crowd artifact it is on different cloud account but maybe just get the context we already downloaded you can re -upload it to this account also and then get this design done on plot uh and then basically uh then only go with pricing page don't just create random ui so i told this that pricing page this i started this worker and this worker is also live testing by the way so it's trying to access i think i'm going to give it access i think by the way it can automatically does it but yeah let's see So there's a testing going on.
I've already started couple of workers to literally do some work on the pricing page and also create a current status of payments and credits. So it's basically going to do that. That's amazing.
I'm going to make payments live today. So I'll be giving some of you guys access to noticing application especially the one who regularly views some credits with it. and that way you'll be able to actually test noticing application and give me feedbacks around it i'm using it for my personal use and it is really helping me out okay Let's see what our agents are doing.
Think it's going to start this manual editing. Okay, so I think this is fixing the bugs pricing page work bug So it just done swim I have told pricing worker to build UI from scratch to stop building UI from scratch use already downloaded designs That doesn't have design by the way re -upload them to a cloud account and then design the pricing page first.
That's amazing Barely a functional engineer say may i know what is the tool that you are building this chief thing is a is this bridge mind or what no this is not bridge mind guy like this is my own tool which i built from scratch and this is basically as clock and uh if you have you're following me you'll you'll understand that this ass clock it is similar to bridge mind but it has a lot of different functionalities right it is available for like windows mac linux and it also like we also have our applications live so it is not bridge mine and by the way uh the concept of chief is not available in bridgeman i don't think say he has built any orchestrator he i think uses some grog models to uh maybe grog kind of architecture but i have a little different architecture i'm using uh this chief is itself is a opus and it has it has a protocol uh which it uses using as clock cli and it has its own big memory which i have optimized a lot to make sure it works so i never have to uh basically do anything like i never have to communicate with any of these agents right what bridge mind does if you must have seen that he communicates with eat in each individual agents right i just keep talking to this one single agent and that does everything for me
Like, we literally built this application all together from scratch in a couple of my last streams and using this chief only, right? And this chief is, by the way, guys, like, it is not just, you know, orchestrating these agents, right? It is doing a lot of things.
For example, I'll show you one example. It has, I told it to a vague prompt by, like, I want to see the current status of, like, pricing. what's going on with payments and also at the same time i want to create a pricing page because on the application that we are right now building we don't have a pricing page right we don't have a pricing page here so i told it it automatically so now what it does under the hood it if i click on this pricing page and if i show the git also you'll see this is a work tree that is dedicatedly created for this pricing page right you can see this is the pricing page work worker and it's work tree and now every changes it will do inside this work tree i'll be able to see that so that's how that's cool that's how it is so it does it automatically yeah that makes uh hi yeah that makes welcome to the stream how are you i'm good how about you thank you so much for joining the stream so that's how i work guys so literally now i told it to uh i want the complete information so it has now you can see what is right now happening
Actually, let me actually tell this thing one tell one thing to this. Okay. So one thing I want to tell that this for the designing since this as your goodie account Profile has a different profile.
It has to use this profile in order to upload that design again.
I Would say I think it's a slog That's log So right now, what is this chief is going to do? Basically this, it has told this, uh, because we, I use no, uh, like, you know, for my designs, I use this application called Claude design.
And currently my, uh, it's already, uh, hit the limit of token. I also have a different account. So it's, it's going to basically switch that account right now.
And I've just told this chief. Now this chief will automatically manage that for me. It'll just guide this agent.
This, what is this? This is called this pricing agent. Yeah to just get that where is the pricing agent by the way, where's the passing agent?
Actually This is the pricing agent the pricing page agent so it'll guide so It has got some feedback. Okay. So, okay.
So basically there are some problem with that account I'm gonna just bring this down guys one second Can you use a different profile, which is basically I've shared with you Let me actually get some information on that profile real quick That is this one Okay, so I've told this By the way guys, if you want your project I review your project you can put those in discord You can join our discord server.
You can also Join I'm live on discord. You can have I can have a talk and also So now it's gonna do one thing That basically gonna do that thing Steampunk you are the good. Thank you so much, man.
Like thank you so much. That's a great feedback just a second i'm just going to do some configuration changes a little bit of okay so basically i've i've told this chief to send instruction to this pricing page that it should use a different chrome profile which will basically have the new cloud account which which it can feed that zip that we already have so we all we use this uh inside this account we we basically use this no uh and this is a different one yeah so inside this account we were using the designing now because i have it in a different account and that account is going to reset up some time so i just literally told this chief that uh just do that for me right so it's going to do that for me it's going to basically ask this pricing page now that you have to use a different account so whatever zip we have will actually upload that right here so we'll actually be seeing that getting uploaded right here i hope so let's see let's just see that let me actually see let's actually see that let's actually see that happening and if it
Face any problem. Maybe for the time being we'll just go with a custom design So that's why that's the reason guys like that's the problem that I face with Cloud design and cloud artifacts that when you switch accounts, you cannot retain the state That is why I wanted to build something like this inside ask log itself so that I don't have to literally but you know It's very like they are doing a lot of things through So when you get that this things design, for example, if when I get my application design with it What really happens they they actually follow the industry practices, right?
It first created a design system. I mean, not talking about the UI, but I'm actually a little bit concerned about the UX. It literally created all the colors we should use, all the typography components we should use.
So I like the concept of actually doing that way. So it's actually still going to this. I'm going to close this.
You're going to the same profile. Use codex agent if needed to switch between profiles, but you're actually going to the same profile. so fix that okay so that's why like i always like to basically have that right so meanwhile this is doing this let me actually do some more testing so that you'll also get an idea what this application is about uh if i go here i just told this maybe it'll just control let's control this control x control s so if you want to send your message immediately just do this control s control x so it just sends it to the existing thinking in between an existing thinking process of our agent let's go back here let's actually do some testing around because i'm not launching so we'll just do testing around mobile app right so what all functionality we have because then we might know another bug meanwhile this is anyway like fixing this okay so what and scenario noticing application can be useful right so we have tried multiple scenarios right it can literally manage any kind of data and now it is even extensive like it has like a lot of manual editing also everything which we can you can do inside notion you can do here as well and it is ai first right we've created that master workflow uh that ai workflow and those ai agents right that we also discuss during the stream so that you you guys can also learn from learn from it
So if I think it's now open this one, which is nice. I think it's now going to like inject and upload those zip files on this. Let's actually see what it is doing.
It is able to access that this one, this Chrome file, by the way. Now I can do this thing manually, but actually the reason why I don't want it to do manually because I actually want to teach it. Right.
So understand one thing. You need to start a completely new project inside the. this new chrome profile because there we don't have that account so that way you will have to inject the existing zip in that so that that account will also have the context and then according to same sort of ui it can create pricing page for you i think it's not that big a deal i could have waited for i'm feeling right now i could have waited for some time when the this previous account gets reset then i can i could have used that but awesome so let me actually go back ask log and i hope so oh it has created this existing this is called this design guys like this is called ask log page i told one of the agent to do a complete in -depth analysis of the payment integrations that we have right now and also the credit management that we are doing so it actually has created this payments and credit as clock page as clock page are nothing but like a normal pages so i can interact with them inside my s clock itself but i can literally interact in the page so that gives me a complete idea what are we doing right now and what is our idea so shubham this is where payments on uh stands on main tested today in the test mode on your local api so it basically tried testing it on the local api
um and database no real money and move like just you try test it that's this is good because when it tests tested those things it tests those things just basically do that now i think it's trying to do that now now now i think uh codex has got an idea so i'll do one thing you know this these opus models are really good for coding and all but codex models are really really good in terms of like computer use so in my mac mini where i'm building as clog and other projects I use my codec subscription only for the QA manual testing and literally by actually using it for my computer use and browser use.
These kind of operations. Right now it's going to actually create a project, I think. Now you can see this is Astra, by the way.
This is not something like... I think it's going to add a project inside that. See, now all of these things, if you see, right?
Imagine that I if I have to do that manually, right? I just sent it a prompt to my if you would have noticed I just sent my prompt to my chief now chief since it has memory So if you go to this this this particular hero profile designer If you go to this this particular prompt if you go to this particular agent, I think I'm just disturbing it But let's fine it has given all of that detail back to this agent, right?
So that's how chief works and now it has context so it's basically able to now do that so if we go here it's now able to do everything because it already had have that kind of context so it's literally going to upload those zip files over there and then then in the second cloud account also we will get all those information which we can literally use for building the pricing page design so we are right now designing a noticing .com pricing page so you can see it is uploading the context so everything it is doing for me so that's why i like astra and codex that's the only reason i am sticking i have not cancelled uh the codex plan on one of my subscriptions so i cancelled one because their coding models are really not that good so it has given initial contacts and now it will basically it will send this this information back to the another agent which will then give it a prompt to get the design of pricing page done so that's how Chief works guys and it by the way, it uses all your existing tokens All your existing cloud and codec subscriptions.
You don't need to Like some people say that do you buy and like I we don't need to do anything, right? So it's literally uploading that noticing right? It's literally doing all that thing for me and This is the reason guys I am able to launch my application so fast now that it has uploaded it has told that cloud agent So this uploading task is being done by a different Astra agent Now, once it is done, now it has given that to Claude back that now you can literally go and debug that using Claude and Chrome.
So that's how your multiple agents are getting orchestrated through this chief. And that's the beauty of using chief and as log.
So it's literally adding all the context, all the different information on this account so that we can literally use this account also for sometimes.
So I didn't even know that we can actually do some import. I just told my chief and chief knows about it. It actually has created the agent and it is helping me doing that.
Like literally, right? So if I see the chief, what's this doing right now? So the web PR fix is merged.
Notes, no more false. changes on another device conflict no more lost words okay all of these are fixed now and looks like the thread scroll at end to end the worker finished the full smoke test list reached it so yeah i think looks like most of the bugs that were there it is now fixed let's see how can you imagine like multiple things are working on so you can see now it it is now using cloud to give prompt to it See how amazing it is.
So you can look at the prompt that we need to create a pricing page. Now, if I have to do this all this manually, it would be like a disaster because I would not be able to provide all these context and how it is possible that it's able to do that because chief has its own memory. So it knows from the very first starting when we started development of noticing from then it basically started doing this thing.
Now it's basically going to use this. By the way, I think I have, I think it's not using artifacts. So I want to make sure, I'm not asking these things to Chief, right?
It looks like that it is not using Cloud Design and Cloud Artifact, Chief, because I can see that has opened in the correct profile, but seems like it is not using Cloud Design and Cloud Artifact. I want to make sure that it uses Cloud Artifact and Cloud Design. So just make sure, evaluate that, and it should happen like that only.
I can see that it is doing stuff, but it is not using Cloud Artifact. It should use Cloud Artifact and Cloud Design the same way we were doing inside the first account so that we can get the similar kind of results.
I don't care about anything. I just talked to this agent. Now, it's his job to send this information that I talked to it to the correct agent, and then at the same time, it will now send...
uh information it using s clock cli protocol to any agent that is working already on that and that that will now fix it so look at this so now it's not like it has a design option but it has not used it right so now that you know it now it has started using design look at this right now it has this using design so this is how it works right by the way i want to check one quick thing whether it is using astra in fast mode yeah it's using astra in fast mode this is something which i explicitly told it to do reason being i have a lot of astra credits and i just want to do and research also so i just wanted to use those uh by the way it's using 6 .1 so but i have a good amount of credit so that's why i just wanted to use them that's why okay now let's actually mean while it is doing that because it will keep doing that we can just go so this is what i do i just go to the agent when it's completely done and i just have to just review that's that's how i just do so i'm going to do one thing i'm going to review the payment because that is something which is very critical for me to review and that's open there so now it has we have a ask by the way cool thing about ask log page is that agent is listening on it so i can literally select anything if i have a doubt around this what
is this can you change this to ten dollars let's say i can literally queue my and send my message right here and then quickly it will update it here so that's the best way you can brainstorm with your uh artifacts and with your um agents right when you are brainstorming and that is why every single decision in this project this noticing project we have done we have brainstormed using this only so let's say what is uh where do we where do we stand in terms of payment integration So this is where the payment stand on this main tested today in local.
I think it has tested. The gaps are around the edges. A payment can be lost.
Okay. Something is wrong. I think it's basically using that.
It's going to upload the design. Let's actually do that. And let's actually send a prompt.
I'll be back in a second. I'll just bring some water for me.
Look at this guy's like how it is doing things for me like You know, if if you're following these streams, you would notice one thing we've taken an idea completely from scratch and We literally build this whole idea and this is pretty decent idea I always wanted to build something like this right and it literally has created the whole application and made it live on Multiple places right?
So this is the power of agentic ai this is the power of wipe coding if you're doing supervised wipe coding like i said i always say don't just blindly go for wipe coding but you always need to think about you know what is going behind the scene what the architecture is going on and what everything is going on right that's basically going to create all of those things and i'm going to upload all of these things right imagine if i have to do this manually and i really really love their uh this this op this designing tool that they have this cloud artifact this is so amazing i mean and i always create designs using this only so you can see it is now going to basically upload the correct thing see how fast it is doing because i'm using in fast mode the astra as i have like enough tokens for that they are to consume let me actually increase the little bit of music volume let me let's review this okay so uh the gaps are around edges payment can be lost if the confirmed step fails two plus subscriptions are possible nothing happens to add zero credit okay because meeting met metering is a still shadow only the decision so i think there are a lot of problems because we just created that payment but nothing works tested end to end today 500 credit nine nine dollars 500 credit pack seven four nine okay plus
monthly is ten dollars okay the stripe hosted checkout page and then we can buy 900 with nine dollars we can buy 500 credit packs and that literally okay and the subscriptions we also we have we have two subscriptions also and then we have a way to add credits also that's what i wanted because this is a token heavy application i mean ron will consume a lot of tokens but we may need to add when you're doing a lot of stuff then we have catalog and prices free we are giving free for 50 credits per month plus ten dollars seven nine nine and ninety six dollar okay meanwhile let's just keep doing my work in the background as well so i think that's cool so this is the tool i use guys like every time this these agents requires my uh credentials so you should actually use something like one password i always talk about it because one password is pretty amazing tool it can literally give your agents access to all the credentials right now if i quickly show you my workflow i never stores anything inside dot env file for example if i let's say open any agent which is actually working right now this payments and credit let's say right now if i open this agent uh it has its own workflow by the way it has its own uh work tree
in that it is doing some work but if i go to the folder structure you see every single application i have be it my api my admin so this is the kind of repo structure this is mono repo i follow and you can see it it doesn't is that it does not have any dot env file because putting any credentials locally on your system these days is really really very risky so don't do that because you know use some sort of like vault tools if you don't if you have one password that is amazing if you don't have one password go with some open vault some sort of uh alternatives open source alternatives there's bitwarden which i think gives you some free uh option as well to just restore your tokens uh store your keys and everything so i have multiple vaults inside my one password i use that i mean this is one of the most important uh in my use case and sas application that i use because it manages all my tokens on my private keys ssh keys everything and by the way it has mobile application version also so that makes it very very easy for me to manage everything in one place so all right so now that uh now that i think my backend qa testing is also finished so we can do a quick testing of the application as well and just give some more feedbacks for the
chief also uh so that it can assign those qa tasks and similar bugs to qba as well but actually go back to our payment integration because that is something which we need to really really read i think now this is now working on my the design of that pricing page which makes sense i can go back to the my other designing work okay so give me one second guys i'll just bring my coffee okay i'm back okay i'm not seeing your message guys like are you not watching right now or what's going on is it like youtube limit hits again watching aditya thank you so much man all right so let's see guys like okay so we can see that this design work is going on we can go back to the uh agent uh the the one okay thank you man okay so let's see uh right now okay so let's actually review this and actually give it i think maybe some agent is requiring that access so we have status per price so the idea is that noticing is going to use some apis in the back end right so because it requires that run agent to to be that run agent to be running on certain apis right so it definitely requires some sort of apis in the back end right so we need to be make sure that we are also having a little bit margin maybe in a starting negligible to zero but at least some sort of margin and we want to make sure that it is consuming less tokens at the same time it's uh
at the same time it's not basically let me actually do one thing let me actually pull this down so that it doesn't interfere in with my this thing right okay so we need to make sure that it is doing some sort of profit also at the same time if let's say we are giving ten dollars plan right and if they're consuming all their tokens they should have an option to buy a credit back and for that we are giving a nine dollar additional credit pack in which they'll buy 500 credits and say in for if they're coming from india they with 749 they can buy 500 uh credits for 749 which is equivalent to this amount right and plus plan is of 10 monthly and it is 799 which kind of makes sense for me and with 10 they can get around 500 credits every month makes sense this is okay for me so this is like good for me i'm hoping that i'm seeing your messages here all right so now that let's actually go more so status per piece catalog and piece uh fifty dollars per month uh plus uh this same indian amount and if they want to buy for in like for a whole year they can also go with this and for indian users obviously gst will be there
stripe checkout confirm portal so it's actually i think ready host to check out so no data card reaches us indian cards can't pay in rpi so the page says so so basically everything is integrated for both india and like outside of india i think laser pay orders and subscriptions also you are using we are using and we need to make sure because this is this is a complete payment layer so everything is proper in this that's what is very important right all the web hooks and everything then we have webhook then we have credit ledger so uh monthly reset also uh free gets uh 50 every month so every month we are thinking of giving them free but i don't know maybe um financially how much would that be possible for me to do depends on that i'll be doing that And then metering is also something where every rent, turn, filling, answering file, reading, transcription is priced.
One credit is equals to 0 .01 cent of model cost and debited. Nothing is refused. I would like to do one thing.
I want to make sure that if we are saying that one credit is equals to 0 .01 cent, that means it should be metered and should be reduced from the balance according to the cost right and it should be a solid so i want you to can you create a quick um diagram activity diagram or maybe a diagram for me so that i can understand how does this work so i'm not sure maybe it has created that diagram already or it has not created so it's better to just get that diagram created because this is very important to understand how is this going to basically uh do that for me like how is this going to um use those credits right and how is this going credits get going to be reduced from the balance so that is very important what happens when zero gets so nothing then keeps answering an account gods into overdraft now this is not something which i want nothing then keeps answering an account goes into overdraft no message no i think that's not good thing as soon as the credit goes to zero
uh including their monthly credits plus additional credit it should immediately uh let user know and uh they should not be able to use ren after that because of the price issue so this is something which is critical guys otherwise you know if we do overdraft people can misuse the tool so this is also important feedback that i would like to queue and send send that back to the agent so this is how like i can literally brainstorm the complete thing within this application within this uh s log right that's the best part of it and we'll be able to complete uh payment today i'm sure for that and this is the most critical part that is holding me to make it live right i wanted to make it live but before that i need to make sure that this is something which is available so um buy more prompt in the app so packs can only be bought by going to slash billing there is no low credit auto out of credit notices anywhere okay this kind of notice should be there send it back recovery when okay live balance setting to show a live balance than a minute or more old ren never uses refuses it
We need to fix this because it should use they should be able to see their current credit balance whatever credit they are using We need to create a solid architecture and solid Functionality around credit usage which can literally tell them where their credits are getting used Send this back to agent receipts and emails stripe sends its own invoice and by the way here comes the thing that we should okay it's good that we are sending razor pay and their own email for now it is just fine but then at the end we also have to integrate our mail service so we're going to also create our mail service i'm going to actually add that to my backlog as well problems found so here are the problems it has found uh with the testing i believe a payment is lost billing confirm fails after the person paid network drop api restart right so we kind of like this is now fixed it has actually working on it so it is kind of like fixed that in in its pr so all of these prs are fixed um a zero credit nothing is refused there is no message send and top of the app that's wrong because ultimately those will be going from some account so it's very important that they should be able to um there has to be a cap on a user also so i'm actually building an admin version of this application as well which can really help us to manage all these right manage all the users which user consuming because problem with
These applications guys where the money is involved in the back end. You have to be very very cautious about How do you handle and maybe some? User can try to exploit that your system should be solid solid enough that it can't be exploited I've actually seen this that one of my project that because of very minor bug we were able to lost a lot of money in dollars thousand of dollars because of when we were using something called tango for giving rewards and so you know that happens right one developer can create mistakes right although when we when you're when you're doing your supervision around it it the chances are very less because i'm actually understanding every single functionality and i'm going to like it might have created this diagram for me i'm going to understand every single diagram how is this going to look like the whole architecture and everything for it but i'm going to ask you i want the complete architecture uh diagram in this how are the payments working for both stripe and reservoir
because this is very important right this is the most critical part of your application if you don't understand this that means you're like if you you cannot wipe code this completely you've got to be a little serious about your payment integration especially and meanwhile guys by the way this design is i think uh like this is coming up very nice design so once that is done i think i'll the chief will let us know because it's totally fine for us for the time being awesome so you can see agent is working sometimes on this page from here directly and it is just a normal html just agent basically you can even use it with your normal agent as well like not just on the only on the workers but any other agents also it can work so it's kind of like i think these are the just bugs which already fixed i don't have to much think about it but this is something which i would like to see how metering takes credits of balance so one credit is equals to u .s cent so basically one credit is equals to u .s cent that that's a straightforward calculation and i actually want to use us um sent or us usd as a main currency even if someone is watching using it from india it's better just to have a single currency as a standard currency and i think usd can really help us
So the of the model cost at the providers list price Not zero point one zero one cent. Okay, that's something which I understand that one user I mean, that means 100 cent 100 credit is equals to one dollar And the math really works because if we are giving it for $100 100 if we are giving it for $10 subscription we are actually right now having 50 of gross margin but i'm gonna reduce that further because i wanna make sure i'm gonna give discounts to users at least in the initial phase so the ledger keeps thousand of the credit so a cheaper call can cost 0 .3 credits okay every cost is rounded up so no call is free by rounding this means today on every ai call and is exact only refusal is at zero which is like new and behind the billing in four so it's basically trying to say that somebody will say hey like hi right so at that i think they should not be built so gate if the credits left only if billing in force is on an ai call okay right now i'm just discussing on this document so don't take and don't work on any of the functionality that i'm discussing i just want i'm just brainstorming we'll just send this feedbacks back to you and then you can start the development actual development after that so i want to be very specific that right now i'm just basically sending because i'll be doing a lot of brainstorming and then once that is done then it will basically go we'll go back to the agent and that's how you should also do guys like that's what that's what called supervised uh
wipe coding i always say right awesome so let's go back to this ai ai call usage so basically if the credits are left only if billing in force is on then usage input cash and all obviously all those two according to that usage uh we calculated the price and then we calculate how much credits to be consumed and reduced and then debit once per call and then locked i think this is pretty much fine and then update that updates the balance and when the not when not enough the call already ran the rest over overdraft that's good so a read call shivam i'm back okay shivam we'll review your project by the way like you were not there so that is i didn't review so we'll review that new project that you shared looks good pretty awesome so i'm gonna review that in us in a couple of time i'm just reviewing this document then i will come to that okay so a read call meter today uh on my test so it's basically a fresh input which is basically out of which i think 19 741 are cached only so six tokens are only used and that cost was pretty pretty low that's cool
so how how credits flow this is something a quick diagram a monthly allowance 50 uh on first utc and then 50 uh plus 50 per paid invoice unused one month and with with a month so the same concept that the chat gpt on these also works that if let's say you cancelled your plan and you already paid for it you keep you can keep using that up to a particular month spend is basically each ai call priced at model cost and then credits zero credits may be when you run out of today nothing happens in the pr flag of uh capture answer and say no out of credit so it's basically important that we should say out of credit this makes more sense buy more is basically a proposed card where they can buy more that's completed or wait for the next month until you can get that now it is asking a couple of questions to me i think it has not created the diagram architectural diagram i would like to see that but let's see let's just answer the questions right it's very important that you keep answering your questions to your ai agency so that they can get information about you so what happens to zero credit pause ai to add zero capture uh ask and run and out of credits so it's basically gonna do that that's important
One them before zero so at least actually at ten credits left about two three turns. This is recommended This is nice idea. Why not?
Pack sizes and prices so currently these are the one Plus work out of credit with the plan top of your skip as they are Where does buy more up here the card in that chats can also help since they are communicating everything through ai so it will be best that if they get that to get to see on ios app android app at zero credits you have used this try come back on first no link no price this is something which is now important to discuss because the reason i i don't want them any payment related information because the otherwise these guys enforce you to use in -app purchases but that's something which basically i don't like thanks my honor yeah man why not like i i i can review any any project that is being shared on discord okay so never refuse on mobile until in -app purchase exist what do we do about mobile guys i literally want to talk about this now i'm going to start a different discussion around this from with my chief because i'm a little concerned about what do we do on mobile how do we fix this issue one problem is with whisper flow literally here
and this is not i want that you know they just go they cannot just use it dollar seal tax fix the price uh our final no tax added for now that's fine yeah i want to discuss about this concept of um how do we go with about this in -app purchases thing because i literally want to uh not you know let them know if they're on mobile application they should be knowing that their credits are enough and they can buy more credits so some something should be there in mobile application as well i don't know how do we do it but this is very important let's send it to send it back now i'm gonna basically also want a architecture diagram of the current system payment flow and then by the way we'll be testing this on live today i'll be talking about one more thing which is basically fast checkout so currently if you come to noticing let's say if i come to noticing right now if somebody just want to get started right so they should they okay right now invite code is there one that will be removed for sure if i open my local right now
The very first thing that should happen. Yeah, it's not running on the local if they want to just get it started They they will be creating their account account creation because immediately they should be asked for premium because that's called fast account I mean somebody already on pricing page and they are just trying to basically get that They can simply be taken the credit so that those sort of things are important very so I think the designs are created Let me actually see what chief is going on, right?
So uh i have told the codex to cloud artifacts as these looks owner owner and nursing project upload your design chips and i think okay what is the current status what's going on right now why are you listening to these guys why aren't they sending you response back through s club cli why is it listening why is that is it like doing this work by himself which is also wrong in a way what are currently running inside your shell what is currently running inside your shell is it like payment page or something else i think it might be payment page yeah it's actually the ass clock page that makes sense that makes sense it's not like i always start thinking that it is polling these guys but yeah this this kind of makes sense polling should work for ass clock page google sign up is already there google sign up is there just one thing before going to before it downloads the pricing page i would like to review it so let me know if the pricing page work is done then i'm gonna actually review it and i'm gonna give my feedback then only start working on the pricing page development so guys like i just told it that i want pricing page and it literally might have
actually uh given all the important information around that so you can see it has created the whole pricing page it has taken it from one account then gone to the second account and then did that so that's gonna let's actually see what kind of pricing page it has given oh actually it has updated the home page also that's nice no i think though this is the pricing page okay sorry sorry sorry i'm gonna click one thing let me first see what it comes with it has created it's only created one page that's nice here also one page only that's only one page that's nice i would like to add add the pricing page component on main browser also but let's actually review this i think it's already open that's just asking for my review it it must have told the worker not to work on this as of now so let's actually review this i'm gonna click on present uh desktop light that's fine oh nice not bad we close this we close this how do i close this let's close this let's also close this okay so this is what it designed guys like wn a w and l in the chat this is simple plans ren runs on credits pretty straightforward we have an option here we can switch between yearly and then we have some saving option which um
we'll see if we can give then there is a free plan where basically they can get 50 credits every month so all of these information is feeded by that unlimited manual use which i wanted to give in initial time 2 gb storage and then they can literally use 50 credits which is fine and then for the plus plan we have 10 a month Then They everything in free so they get 500 credits instead of 50 very straightforward Tables list and everything they get At a pucer casserole Views are my stream in guys So pucer a application I'm banana noticing .com kirk a And the idea is that we are making it using AskLog.
So we are using AskLog for the building. First of all, thank you so much for joining the stream. And GPT 6 .3 Astra, Shubham is saying.
So yeah, so we are using AskLog. And in AskLog, there is an orchestrator, Chief, with whom I can talk. And all these agents running in my background, it actually manages all of those agents, right?
so right now it's actually running you know you can see it's running like how many agents it's running around fixation but there are instances where it is running multiple agents so it's and it actually automatically assigns these uh tasks to these agents and it is it assigns these tasks to these agents and um all it does when when whenever i have to basically give it a task let me actually close this it also creates work trees for them so you know okay conflict and everything do a tweet about as clock will repost amazing product yeah yeah sure sure sure sure And by the way guys, I'm using Videst, I keep saying this multiple times that I'm using Videst.
Because of Videst, I'm able to stream on Twitch as well. I'm able to stream here as well. It's a pretty amazing software.
And by the way guys, have you seen one thing? I'll show you one thing. On Videst .com, Piyush sir has put our testimonial as well.
Look at this. And the first testimonial is ours. It's so nice.
Right. So this is really great. Thank you so much sir for this and amazing tool.
I am using this only. I am going to record some certain YouTube tutorials and some tech related videos. I am going to use this only because this is an amazing tool.
Right. So if I just go to. Now I will tell you what I am actually making.
If you are watching. So using Asklog, we are actually building this application called noticing .com. Where you can literally...
Bro, I listened to your Hindi today. So, I feel blessed.
Cool, cool. Sorry, I... No, no, sir.
How are you talking? You didn't disturb the flow. It's actually very motivating.
Actually, you know what? One day, Manu sir also came. Manu paaji also came on the stream.
And it is really, really motivating for me. I implemented all the feedbacks that he gave on AskLaw. So my idea is that while doing these streams, at the same time, whatever we are making in the noticing application, I also explain it to people.
agentic engineering because my main domain is that right the company I'm also working for so I try to teach people that how to use evals how to use evals how to use evals how to use evals how to use evals how to use evals how to use evals how to use evals how to use evals how to use evals how to use evals getting good feedbacks from people as well.
So, that's it. I mean, it's trying to walk on your path.
So, but thank you so much for coming on the stream.
One doubt. What is the difference between your project and Claude's own harness? How Claude creates sub -agents and what does your product do different?
So, see the... one thing that i'll tell you uh like cloud is like if cloud you're using plot it can create sub -agents of cloud only right the way i in the way chief works is that it uses something called as clock cli which i created i think i've talked multiple times that when i started building this as clock uh i actually created as clock cli right It's very similar to Tmux because when I was using Tmux earlier, I was actually using Tmux only for managing these multiple pans and everything, right?
Which is good, but it has certain limitations, which I wanted to use with my application, especially around Orchestrator. So the main difference between Clot CLI, Clot's own desktop application, and this, as Clog is this, that here you can actually even if you are using let's say i'm using orchestrator as opus i can start a anti -gravity agent i can start a codex agent doesn't matter and it has its own memory which i have kind of optimized right so it uses semantic recall all those concepts right and it is smart smartly handle things right and it can literally you can literally visualize these agents you can inst in like if you if you are working with sub agents you cannot go inside those sub agents maybe you can but it is not that good ui friendly right now here you can just literally see you can put this put this uh chief on the right hand side you can have this different kind of views where it is like tab view great view can when you can when you will be very very because i am actually doing a lot of stuff right so you can literally do one thing you can literally see and go to even individual agents let's say if you're not happy what chief is doing and then i mostly use grid view and then it also manages your local server so for example if you ask claude
to run something on your local if they normally what they do they actually start inside their shell which is not a good because you cannot see what's going on right so right now if i see if i show you for this noticing project i am running these many shells uh these many local servers right it is running a this clam virus antivirus tunnel it is using running my web i can literally see that and i can actually you know tell anything for example uh chief can you do one thing can you close this master studio it can organize it can do all lot of things other than just a normal plot right and it does it pretty well and one thing with the sub agents i have noticed this personally that when you use this it will consume a lot of tokens because every time it spawns a sub agent they the sub agent context gets back to the uh you know the main agent again and that that way you can lose a lot of tokens in case of chief it just assigned tasks these agents get the task done it automatically creates multiple work trees for them and dedicate give give them a dedicated work tree so you can see it has just closed that right so everything it can do so it just creates these um work trees they work on their work trees they just raise the pr now what the chief does after that it literally creates a review agent from codex so the way i configured it depends on you how you want to configure it the way i configured it is just i
told this chief that always start development with opus 5 .5 and then once you're done just go to a codex agent and start a review agent and everything happens within the same work tree and then they raise pr they communicate over the pr and then they raise pr like if it is a critical pr i also review that and then once that's done so i basically merge that so that's how it works right and that's why i've been i've been teaching in these streams that how you can use these uh chief and that's how it is different from normal that's one like code clock agent sub agent workflow right that's that's not that's very dependent on models you can only use cloud models here you can use any kind of models so these are the differences that we have here arman right so let's see i think it has already designed it and now oh all right all right nice work amazing thank you so much man okay so we can go back to the chief let's see what it has done i just told you to you know kill the master but i'll just tell you to start it again can you start the master studio once again and also like can you
Let's now focus on payments only And then then we have this as clock pages through which you can literally it is connected to an agent, right? I can literally using this document only you can see how nice Diagram it has created for me to make me understand right and I can actually annotate any sections. For example I am not understanding this particular point right this particular point like What is this?
Right? I can literally set it from here also do the note i'll send it back send it to the agent agent will then work on it so that's how it normally works and i use this uh one password guys like in my workflow which can access my one password for it has dedicated vault for every project and whenever they need credentials they can literally ask ask it to me i know what's wrong wrong that's how like you don't store any credentials and critical secrets on your machine that's also very important these days Bullshit, we are we are not interested to see how you wipe code show me what you have done for your company with the wipe coding so Like When I'm here, I'm doing wipe coding here.
I'm not just trying to explain you how to do wipe code Right. I'm trying to explain you how to do supervised wipe coding because you must be seeing only The wipe coding that I am doing but that's not only thing I'm doing i'm also doing explaining different concepts right and if you're interested what what i have done for my company why would i tell you that there are nd assigned got it so i think first of all i cannot share what i have done for my company and like i'm literally sharing my knowledge what i have done for asclock the way i built it i'm building this noticing if you're not interested then you can definitely leave and i'm not interested to answer to your questions also in that case open code use karta so mohit i'll tell you one thing if you don't like the concept of my videos if you don't like uh the way i should you know whether i should show you wipe coding because a lot of people gives feedback around this they want to see this thing if you're not interested go and watch some other channel man it's not here something you have to come here and see how do you wipe code right if you know everything already then you should not need to come here you can just watch some other videos i hope you understand so i use open code sometimes but mostly not because i have these subscriptions so that's why
bro relax dev from hyderabad relax bro first of all you should think and you should see what language you're using right and then you say you should if you do any negative comment obviously how can i be chill right and don't this is my warning for you don't do that again otherwise like i'll just remove you from my channel okay so let's see what's going on with this payment worker and if you really want to see you can watch my last couple of videos you will understand whether we are doing just wipe coding here or doing something else also right if we were just doing wipe coding this this escalador page if you see it would not be this much filled with these many discussions and these many uh stuff right like i said i don't use kiro and open code as of now i've used them but i don't use them as of now okay so let's go back to the topic guys i mean these kind of comments literally distract me and uh i couldn't focus because of these guys and that's why i i i want my audience to be very you know positive not negative and all of that if you don't like man just leave i don't want my audience to be like this uh this is a stupid person like mohit right who's just coming here and saying i don't might want my audience let me actually remove this guy it's not i think i don't really want this kind of person in my
let me actually first remove this i don't want his view also if i can ban this guy i hope he gets banned also yep i think it's better because i literally lose my okay so let's let's actually ignore these like i'm actually you know i'm little new to youtube so i'm trying to ignore these guys because um sometimes like people like puser comes and they motivate people like you know shivam pritam these guys come and they try to motivate and then these kind of guys come Okay, so let's see Okay, I think it's basically made them my master studio once again up Let me actually do one more thing that subscriber mode also on subscriber mode i will ask this can you in my live stream right now which i am doing right now can you access and uh make the subscriber mode on on the chat let's actually go back to our noticing project okay let's first see the flow here stripe dollars outside india hostel checkout so browser slash billing checkout checkout session happens you can see this activity by the way it has uh turned it into vertical but normally when we see um like these activity diagram it should be like
horizontal so I can actually ask it to generate it in horizontal can you generate this activity diagram in horizontal because it is not really looking good but we can see what it is doing basically when we the way stripe will work basically uh it will start a checkout session and then it goes to stripe and then they basically uh do the payment and then that goes to the hosted page right so oh it's like what i wanted the same oh it's actually i wanted the same i wanted the same sort of activity diagram but um main component should be horizontal because activity diagrams are the best way to understand the flow of application guys let me actually do one thing okay what the hell it is trying to do okay let's see let's let's try to understand this i can just revert this i would i would go to the first first first one that you build yeah this is fine so we do check out and then session and hosted page back to the page confirm and then that fetched the page yo preetam i'm i'm late was busy today no worry man thank you so much for joining again welcome to the stream so we have we've done a lot of progress on noticing application and right now we are just discussing its main payment flow
and this is better this is better this is this this is better i mean if this could be like just rotated by 90 degree that would be great but i understand that i think let me actually try to ask this is perfect but this should be uh rotated to 90 degree because i want to say stripe noticing api browser on like a horizontal line and all these activity lines on uh like on a vertical i hope it understands now because that normally we have a habit of seeing activity right diagram it's like actually has made it like this this is not good yep this is what i wanted this is i literally wanted now you can see that i think i have to zoom in a bit understand the flow of stripe and yeah so you can see right so where is it yeah so browser and then browser will start Checkout page and then it'll just create a checkout session.
So browser will basically tell what is the back -end price? We need to create a checkout session and that goes to hope it then it will basically pay do the 3ds payment So normally here what happens it goes to the browser again if they have you if they are using 3d secure card and then once that is done it basically fetches Never trust the browser.
So this is something where it's very important that when you are building your payment flow Even if you look if you are confirmed the payment is done You should never trust the browser session because you should always trust a webhook later on So this is very very critical and I think it's doing the same thing. That's what I was expecting and in case of This also it should be trusting these webhook, but I want to make sure that these web web hooks are secure So I want to basically do a prompt right here Because normally what happens sometimes these webhook should not be like public.
They're signed they should be literally like so both of these work webhook can you just confirm both on staging and live using maybe codec session and see if these are literally secured with a secret key and that secret key is stored securely in kubernetes i want to make sure that happens can you please confirm that whisper flow issue i don't know maybe what's what's wrong with whisper flow it's like browser is blocking it or what i'm gonna fix this uh on this then this is not happening on s club pages but multiple places so this is also important guys that you need to understand that when you're dealing with uh payments that you should never trust a browser success you should always check for webhook so let's actually send this and then it'll may it may start a codex agent and do then eight whistle i think this is get and this is i think we have already explained how does this metering works so i think it looks good um and where you uh what happens when you it has created some diagrams what happens when you go out of credit so if i just quickly open these diagram if i quickly show you that new chat
three credits used this month so it's like going to give you a brief idea about where our credit is getting consumed maybe we can get it redesigned but for now it's just fine no air credit left so it's like that and then buy credit will take you to the buying buying more credits that's good that's what i wanted and here we have like a like a customer we can see the balance on that this looks perfect now we have this mobile okay so i want to discuss about the mobile uh how do we handle payments on mobile because if we just handle the payment on mobile like web app these guys will crush us these guys from app apple and android they want us they want us to pay 30 of our revenue to them which i don't know how does it make even sense It's basically it can see that it has tested, but I'm gonna want to double make sure that these web books are Good I've got someone.
Oh, I dropped from Yep, especially Apple yeah, that's true. I mean Apple just kills you for launching your app without Purchase so I'm gonna literally handle it and that's where I'm gonna have a discussion with chief separately let me actually ask chief one thing this page that i'm currently interacting are you the part of it or is it some other agent which is listening to it i think i think this chief is only listening to this page anytime like when you use as clock pages there is a chief which is there's an agent which is listening to it and in this case i think chief is only listening directly so it depends which page you're working on so i'm actually talking about payments and credit i actually it is listening on this keyword status not the one i'm talking about okay if i publish my app on app store and play store do i necessarily have to take the payment through them i also don't want to give my earnings to them yeah that's the problem that's there you have to pay them but then there are a way there are ways and i'm actually going to explore that today
because in asklog what i'm doing i'm just not taking payments on phone phone apps are secondary for the payments right so uh literally we're going to learn that today that what's the best way to do that right so um okay on payment so it's done uh worker so it's basically it's not listening it's worker uh because you know what i what i want you to listen on that project First of all, I have already given and understood the architecture.
I think that's fine. You can just close that. And I want you to start listening on that because I want to discuss something around payments, handling those payments on web pages, sorry, mobile applications, in -app purchase guidelines and all of that.
So I just want to discuss with you directly through that page. So I want you to start listening on that.
Let's see if it can listen on that.
they directed the users to pay using a browser but then what they say in their terms that you can even you cannot even uh show something that is unblocked but i think in case of uh us we can do that because we have we are showing the credits not we are we are telling them to buy credits they're not telling them to buy a subscriber let's see let's see let's see what happens right now it's listening then now this this agent is listening on this right that's cool if i redirect they won't get it they're cut hi i tried to visit your site uh 127 .0 but it is not loading for me which side you're talking about so if you try to access this site then bro this is a like a oh my god this is my local host this this this as clock runs on local host so and if you're trying to access noticing .com you should be able to access it but you should not you'll not be able to get access to it because that is currently uh invite only mode so i'm going to give certain people like maybe preterm shivam these guys like who regularly joins my a little bit of access and some credits on noticing to use for beta testing but you can access noticing .com by the way you cannot access um
This one to seven because that's on my local that's running on my local post like localhost So so go hotel X Learn about some some fundamentals of STTP, how does this work internet works?
Shubham I'm listening on the payments. Okay, let's let's create so now I'm gonna bring this now as my chief is listening to this now let's see mobile direction and now balance plus back on then in -app purchase of pack it's basically saying we use in -app purchases right uh for for additional credits only for now straight away before turning in for enforcement on enforcement on mobile hi sir may i know uh what you are working on uh is that for noticing payment yeah that's for noticing payment so decision only you can make what happens to zero credits so the the idea is that with ten dollars they'll get 500 credits and once they consume the whole credits they will uh basically can purchase more credits When will they make a new model for GPT?
My plan got wasted. I'll learn good manual coding from GPT 6 .1 Sol.
Good, good, good. That's true, Preetam. I understand that.
I agree with that.
Okay, decision only you can make. Pause AI at 0. Capture and run reading out of credits.
I think this is fine. I think this already we have given that. when when to warn At 10 credits left.
I think this is also recommended. Why not? Pack sizes and prices keep as they are This is fine Where does buy more appear a card?
Okay. This is also fine. I think I've already answered these questions, right?
Why are where do I need to answer these again? Let's actually yeah, let's actually kill the chief Let's actually give these answers and we'll focus completely on phones on iOS and Apple apps at credits you have used this month they they come back i think this is fine for now let me actually send it dollar sales tax this is going to be fixed for now actually so i'm sending all these options just remove these decisions because these are final from my side but don't not the one which are involving the in -app let's let's completely focus on in -app purchase now because i want to discuss about that in whisperflow issue i just heard my voice back for some reason what just happened let me actually it's local website he said give me one second guys something is wrong oh i i accidentally clicked on play button of that noticing i'm gonna go back to the chat so it's local website once only he said i think he has to pay and then it goes online no no i'm just developing it on my local i'm just having a conversation and it'll when once it will be live it will be uh definitely behind auth and payment you have redesigned the full noticing ai with cloud design theme you've created yep that's i mean
uh i was having a uh new application uh i have i was having a on my different cloud account i was having this noticing project and that's the problem with cloud design that when when i was wanted to use because that that plan already was on reset like a token already used so that's why i i basically kind of like um Imported that that download those zip files and then I'm imported in another cloud account so that they can all that can also Work on the pricing page by the way that has actually worked on the privacy But I have not given my feedback because then a stupid guy distracted me.
So This this is basically I'm not talking about you. So there was a guy like who came to the stream and then talking shit. So So go hotel X.
This is not this is local bro. This is not this is not something which i'm gonna make live this is as clock pages as clock pages are basically just a development tool which i will be i'm just communicating with these so that it can send feedback to my agents that are running here and then they can work on that uh those feedbacks it's just that by the way let me actually do one thing first let me actually put my pricing page worker on working because i think that's already done it and we were reviewing that those things so i'm gonna quickly open the pricing page designs that cloud has created just a second just a second guys where was those designs where are those designs where are those designs yep these are these are the one when it is live this is not going to be live this noticing .com will be live this as clock pages is already oh you're talking about as clock then i'll tell you as cloud .ai is live you can download this and you can use this as cloud .ai this is the tool that i use for development so you can download it for your mac that's what you're talking about that's that by the one second guys just give me one second now let me check one thing okay i'll just go quickly
i was just checking something okay that's fine so okay now look at this this is the page that it has created how cool is this so i'm going to uh play this i'm going to play the first the desktop version you tell me how do you like it mac is more better than windows for app and absolutely correct if you have money just buy a mac windows suck i think linux is better than windows so simple plans then runs on credits so everything you do hand is free and ultimate unlimited on every plan the plans differ so by the way we have revamped the complete manual mode where we can literally do a lot of manual stuff and i think i have shown in the starting of the stream So this is the basic plan.
You get 50 credits every month. I'm not sure how much I'll be able to afford these. But you get 50 credits for every month.
And then you can literally use these. And then I'll make sure that people cannot just... I bought a gaming laptop last week.
It's like, I think you should have gone for Mac. If you're really serious about... If you like gaming, I think that's fine then.
But if you're really serious about development, I think that would be good if you can... go with mac this is a nice nice nice pricing page it has created guys like i'm gonna i'll actually do this thing let me actually let's see full view of it so this looks nice and then we have like one credit is one cent of model cost so one basically i think it's explaining too much here wow ui looks amazing yeah i think this this basically is telling that thing 500 credits you get and 2000 credits you get with this sign in to buy and one thing i like i would like to have a fast checkout but i think for now it is fine i mean i'm gonna go and go with uh this directly and then we have let's say now the cool thing about this is that you can literally um switch these and you can actually go for let's say uh light mode it's phone so i'm gonna go to fill so this is how it's gonna look like on phone pretty decent not bad oh this is bad by the way we'll fix that maybe it's looking bad on my phone i'm actually gonna actually see the desktop view of desktop yeah okay i actually i have this feeling in my mind that
the text of view of noticing is not not that great it doesn't give that vibe by the way the color graded if you see here this looks nice but again overall i would not like it so yeah we'll change that later on but i think i'm gonna just give the feedback to uh my ass lock that uh i just saw the view the the pricing page it looks really nice start implementing that tell that agent to start working on this immediately and let's discuss about um yeah let's discover let's discuss about payment on phone how can we handle that because i was like if let's say i'll be running ads on instagram and you know like meta ads and google ads people will be downloading applications right they'll be directly not going to website so they'll be doing payments on on phone only right It's minimal simple and clear.
By the way, I like minimal simple and premium designs I don't really like over complicate but the landing page cannot can be little nicer Like the freedom creates really nice landing page. I've seen that so My idea is that let's actually focus on this. I'm gonna actually ask this agent Can you do tell me one thing?
What's the best way we can do for? these let's actually do one thing let's actually close this existing this payment uh and credits page and let's start a new analysis inside a worker that was the best way we can start this problem of in -app purchases because i want that people should be able to uh access my application um and they can they should be able to do payments on phone be it android ios on both so i want to do that And at the same time, I don't want to give my 30 % revenue to these guys.
I mean, I think Play Store is still better, but App Store are not really good, right? So, can you come up with the best way I can handle this situation? Maybe something like taking them to the website for the payment.
But, like, just check it out. Like, what can we do best, right? In order to just be a little diplomatic in a way that we can take payments also and then we are also not...
violating their policies right so can you please get that and also get the pricing workers start working on pricing page immediately because that really looks nice and yeah so it's going to basically start now it's going to close this page i'm going to actually send an end it's going to close this page now it's going to start a new page because payment wise i'm sorted right it is it is able to do that uh payment integration but i'm not sorted uh in terms of like actual payment flow actual payments using on app so let's i think it'll start a new worker now which will have that discussion like it has already started mobile payment strategy that's cool noticing ui is best according to its product market fit yep that's true you know full stack development 70 % good Yes, give me 5 -10 minutes I'll be back and the first thing we'll be doing after that we'll be reviewing Shubham's project and Somebody wants to get their project review.
We'll also review that so we'll be back in couple of minutes. Meanwhile cheap is working It might do a research around this. I want this mobile payments.
Okay, I think it's already saying researcher workers I have not mentioned much but yeah i mean like that's a good thing right so i'm gonna basically put this on chief on a little bit on work and i'm gonna i'm gonna be back in five six uh five ten minutes Yeah, so I'm back guys so yeah, I mean like Even don't worry like will not go Very hard on your project.
Obviously. I understand that If you want you can review the network calls also like normally I review or if you just want that to review just the functional part that also I can do tell me what do you want to me to get review like review only the functionalities or like Little bit about the technical architecture also one second guys it's in the project showcase section What are network Claws, I mean like I I normally I like I think you're not there when when we reviewed your project last time So you didn't know so like basically what do you using super base the basically the architecture by actually seeing the network calls There also I do some time So if you really want to get review completely that's also fine if you just want me to review that the basic functionality also That's also fine i've actually opened this page on the browser that would be great all right so let's review uh shivam's project and then we'll uh continue our work it has opened this thing oh it's 15 percent not 30 percent i mean that's crazy i think they've reduced the by they were getting a lot of uh legal cases from different people so that's why like
they literally reduced this to 15 i didn't knew about that review it completely awesome we'll review it completely i'm gonna open that from discord so guys if you also want to get your project review you can also tell me we'll review that as well so i'm gonna grab this link open that proper here okay first thing first feedback and this was the feedback around last uh time also um shivam that your ui this purplish ui looks little slob I mean, I've seen by default, if you don't guide your agents, these basically creates these kind of UI.
So if you're a little concerned about the UI, then sometime it doesn't look that good. So you have three both sides. I'm opening this very first time.
That's the reason I wanted to review it really thoroughly. So find your people. So it's basically a social media.
It looks like a social media website for me. AI enthusiast. Nice.
so you're using or it's just a pwa application it's not even built on next years it's built on normal react react react library you're using for building this nice and by the way if i open this pre -react what is this this i think i've heard like it's extension of or it's basically a javascript library it's alternative to react and which is basically basically the very uh like i think it's lighter than react that's nice you actually open the network tab so looks like this is mostly client side so one suggestion would be if you can uh because you know seo wise it will not be that great because you you'll not be able to make if you're really serious about this you should use some framework like at least the landing pages all of that should be done using uh next year's where you are doing certain thing in server side also so that you can really um use that so you can go for next year's for the just for the ui communities at the heart and a interest okay this section is looks this section is looking good back in the there's not much in that so let me actually try joint repo so i'm gonna see you have a i think you must be using super base i feel like let's see it's not super base
let me actually try with my s club on and actually give access to your s club so you're not using super base looks like because i have not seen super base here okay that's great let's close this i think i should not have gone for this login let me actually do one thing let me actually uh log out from here i would really like to review your normal flow so if i click on sign up again my by default it should take me to okay this is good it's taking me here so i'm gonna clear my cookie so you must be storing that somewhere inside uh in the local storage or what are you storing i'm actually clear that i think you're storing that inside cookies by default so i'm gonna clear this up and now go to this page that should be logged out nice okay so let's actually try normal login so let's see that will give me a more idea about but by the way i can see that you're not using superbase at least for sign in that's good i like to uh use application oh it's not filling not filling okay display name okay i think i have to fill it manually is it
why it's not able to fill form filler is not able to fill that that's amazing so let me actually give shubham my name username be shubham maybe let's let's go with ask log email so let me try asklog at mailinator .com let's try this one so military is a tool guys through which you can literally um create a website sorry use it as a disposable email so actually let's see what goes here just so if i click on create account it is a register endpoint and oh you're using api inside it so all right so now if i go to response oh it looks like you're using what is this you're using which database it's basically not even they're using postgres or something else it's not a uuid it's not a mongodb id also so it's basically created as a timestamp uh okay verification email not configured so it's basically one thing that without even email verification i'm able to do so basically you're using real time okay got it so you're using real time that means you must be using some sort of real -time database it doesn't look like firebase to be honest it looks like
Your custom implementation API and maybe convex DB. I'm not sure But something is there so you can you can okay. Let's actually see what describes you.
I am I am a ADE agentic development environment Let's see. So if I click here or it's basically updating me here. I Hope this must be accepting Is accepting cookies Cool I'm into technology Let's continue So it's basically your custom api is already all custom apis you guess the database could be post postgres also, but i'm not really sure because By looking at the response.
It's not giving me but maybe It could be postgres also convex also, but super base. It's not super base I think maybe if you're using super base, but then you've created your own api layer So we have a lot of people here. Let me actually follow shivam and let's follow all of these guys couple of these guys let's continue i built it using ai agent okay that's cool so we have multiple follow requests and then by default i'm getting accepted accepted state okay so whatever you've used uh they have they are not using uh like in the database they are not using for you user id use uuid normally so let's say i'm actually uh i want to join this one we have five members in ai enthusiast nice i will join i stacked exploring looks pretty decent then count is what okay it's uh okay verify your email so it's saying that it is not sure if it is going to send me a let me actually check my mail that if i've got email from your site as clock so this is where you can just add a random email and you can just access it oh you're not so you're not sending email so email is not configured yet that's why your response was telling the email not configured that's fine if i click on recent what happens it will also come up with the email not configured i think or email verification not configured yeah i i get that guess that then you can create post i can literally create post
now one thing you know i i'll see i'll see if you have done that or not let's see ah nice so this is called guys you see just what's just happened this like is not yet completed so you see as you see what just happened so i'll show you one thing this cool about he has already done that where can i see as clock change logs um right now it's not public uh skyrex but i'll show you one thing guys you see what happened when i click like it immediately um didn't show a loader here though if i go actually one thing right if i go offline if i go offline that means my internet is not available for this application now if i click like it looks like that you're always it's actually handled nicely right so some people so this is called this functionality where you when you're alive like when your network is on and when you like okay when you like let me actually go back when you like something it immediately updates it here but then the api is also still not completed right so maybe i just quickly do a refresh right now let's see i mean this looks like ai agents and you've done a good good job here now let's see let's let's quickly see that immediately i'll refresh there is this concept called this concept called
this concept that we are seeing here we are not uh showing a loader here rather we're just updating the state on ui and then api call is happening in the back and you see that when i click on this like it is sending this call is not completed this post is not liked yet so but on the front end we are showing this like do you know what this concept is called we have this there's a name of this concept facebook uses it and every like wherever you want so immediate effect but then you your api call might take time so it would be very bad if i clicked on this and a loader comes once this api call is not finished so what's what you call this concept man i'm just forgetting its name this concept is called having like There is a name up to this concept by the way, I'm just not not getting that in my head right now No, what is it what it is called not getting that thing in but I've used it a lot of places Man there's something called front this related to front end Something like update this concept is called something and I'm not
does anyone know what is what we call this thing that when we don't stop wait for the backend call to finish rather we just immediately update the state on ui let's ask chief we call this concept this is very important concept Where is the chief? I think this is minimized.
We call this concept something. So you see on this website which is basically a social media website created by one of my viewers. I'm just reviewing the project.
You see when we click this like button it immediately updates the state but the API call goes in the background. What do we call this normally in tech? We call this something.
I'm not getting this in my head. I want to just know what we call this. there's something related to front end update something like that progressive no not progress ah optimistic update yeah yeah this is it optimistic update we call it optimistic update so we call it optimistic update optimistic ui i got it so this is called optimistic update guys so basically when you want to update on ui now it is not actual like yet happened but you just do it on ui by the way one problem i face you see this way let's see if you have a rate limiting on your apis let's see if you have a rate limiting on your apis this is called optimistic updates so optimistic updates means you do update on ui but you don't have a rate limiting which is bad you should have a rate limiting on your application so if someone can basically literally run it on a loop and that will create your server can be redossed with this so optimistic update is something which we call it and if i just comment it out here let's say but great work man i mean you've actually created a complete ui so you can see this is not optimistic update here here also you should have used optimistic update now how bad is this is looking right so when i was like doing liking it was really nice but when i posted a comment
Like if I post it over here, let's say this is a comment. See how bad is this? This is looking like this is a comment.
Right?
You know, this is bad because it is taking time to either what you should have done the same way you've done for like, you should have just added that comment right here. And in the background, it can keep going. And then I mean, literally anyone can DDoS and anyone can make your database down by DDoSing it.
And let me see if you're using Cloudflare or not. So if your server if I see your server somewhere in your request header not only response header Your server is cluttered. You're using cloudflare.
By the way, that's good So by default you'll get a protection from DDoS But then you should be a little bit cautious about it and you're building something like that and then let's say everywhere you've done it but only for comments you've not done it so this is looking very bad so it is and obviously your database is taking time to add a comment that way something is wrong something is wrong something is really uh taking a lot of time to just add a comment right so yeah i like this i like this and then notifications i'm sure that uh these notifications are not real time because i cannot see any web socket connection over here i don't see you have any web socket connection so may maybe the in the chat you must be having let's see let's let's explore your chat so let's explore so you have a let's see why is this taking so much time to load you can use front -end caching here who has uploaded this shivam okay nice let's go to you have a reels also then in the in readers guys you know what should you do so for example i'll show you one thing what happens when i go to this next reel
oh you only have one drill okay you've kind of like completely cloned this is nice i can share it i can say it to aryan i cannot search people oh i so you here also you can like i'll tell you i have to write the full name i it should suggest me start suggesting me oh it's doing that thing oh it's doing that thing all right nice nice niceness aria this basically your server is a little slow and i understand that because you must be hosted somewhere in cloudfield workers you've used cloudfield for sure for deployment that i can tell you nice nice nice website very very nice good work let's actually see if you've used web socket in chats or not let's say let's let's just see that because using web sockets with cloudflare workers is pretty pretty difficult so you oh so you've not used web socket i cannot see web socket here so you're doing polling here so if i send a message to let's say uh like a new message to maybe uh like shiva let me actually send a message to you
chat let's actually chat with shivam so shivam can you send message back to me i want to see if these messages works real time so this is active uh active oh nice overall it looks good so it's even sending this typing so that means for the messages you're using so if you can send me a message we will see if that comes real time so real time ping is going on so it has a so you know you're using these real so basically the way it is working it is basically pinging you using these these so you're doing polling actually you're not using web socket i can see that but it is almost real time i got approved uh for cloud for startup 5000 credits in my cloud startup i guess i'll cry i'll awesome good news now you can use you can use uh as clock edits full capacity kyrex you can see this call is going on here this call is going on here if you see this call is basically real time so you're using polling for real time but nice work pretty nice work polling is good i mean like web sockets are really difficult to order scale pretty decent work and then we have a community option also database is taking time but that's fine i understand that
Faizan is saying needs you improvement in UI caching. Yeah, definitely So for example, if I go back to real it'll load again though, that's bad because we've already seen that page, right? So you can front -end cache on the front -end So for example here I have to again click on explore although the same content is loaded So if I go back to here, it'll load again.
That's called Basically, you're not using proper front -end caching that can be improved using you can try this thing, right? You can try 10 sect query I'll drop that in a chat, but good project good project i like it nicely done amazing i'm sure this is hosted on cloudflare yeah i'm more excited to use chief with opus then uh got approved for off that's crazy by the way there's an also an android app for this good work good work the only thing is that social media platforms are very uh hard to scale unless you're giving a specific uh yeah this website can be scaled because you've used cloudflare workers you've deployed on cloudflare so cloudflare by default get like can it be it depends so for example there has to be a usp around it because every social media website are already there right you need to understand why will people use your social media website So you need to have some ESP around it By the way guys like we have got this But good work shivam overall very good work
okay so this is the mobile strategy for everyone who wants to know like how can we do payments using uh this let me actually make sure my workers are working on like oh they are not working on anything let me actually close this and i'm actually open that chief Right now the USP is communities and interest based feed algorithm.
Was that me?
Communities are also there in discord, right discards. We can also create groups in Facebook.
I Have one idea by the way for you I'll discuss with you separately for this application which you can Implement and then that can be a USP for you because I was also thinking to build somewhat that idea but i can give it to you so that you can use that the pricing page works locally once i noticed start in app purchase analysis okay that's fine so uh is the testing around uh manual uh editing is done and if yes then can you also start fixing the bugs of mobile application if needed you can launch it into simulator and then yeah if i open a new workspace can i ask chief to spawn workers on particular yeah definitely you can do that you can start a new work even you can tell chief to create a new workspace for you you can start a completely new project you know that's the best way of testing chief is that completely start a new project from scratch okay when will you discuss that idea and where maybe tomorrow i'll discuss i'll connect with you on discord or maybe on whatsapp i have a very nice idea guys like i think which can be little uh
nicer for this application, but you'll have to do a little bit of Stuff around optimizing it location Like you should be knowing how to use location api's Give a given a brief hint about the idea, but yeah, it can be then great Maybe you guys like within you know who the regular viewer I think Pritam is also here, right?
Is Pritam watching or not?
the manual mode it has fixed all in one pr which is green which is nice now it's basically fixing the mobile bugs yeah i want both of these pr to get merged and alongside with that that uh pricing pr to be get merged and everything to be deployed on uh web today i'll be testing payment today so i want all of these to be merged so get it reviewed if needed and then merge them give us some work to these guys now you know what i'm gonna do now And see I don't have any idea of location APS, but till morning I'll good this is a kind of Like this is the kind of What do you say Enthusiasm I want in my in my audience that they can literally Build something You know people like you guys what you can do even group together and build something and any anywhere you you face problem with infrastructure or anything i can help in that so that way community will also grow and then you guys will also get idea that how to build projects i can give you some nice idea maybe some sort of hackathon in ask love to help you guys build stuff and then scale it and then maybe earn money
That's the good I want I want these streams to be very productive for you guys like Some of you maybe you know beat them you and kyrax can connect together and then build a product and then you know use your skills to because rhythm is really good in front and right and You can you guys can really build something together Just a second guys Yeah, we guys should build a startup or something together.
Yeah, definitely. That's what I want I mean, I want you guys to actually learn something and then Build something together and that will be great. I would really be very very happy with that if that happens Hackathon is a great idea.
Let's go. Yeah, we actually I can do hackathon and then give some giveaway also to you guys Whoever has the nice idea just a second guys just a second yep so i think it has started the mobile issue also fixed and i think all of those when once will fixed then we'll be able to launch the application uh today and uh then probably for the free account only i'll put the restriction of invite code and for the paid accounts they can literally create accounts um something like that and i'll also give some of you guys that access to noticing free that we will be able to test also and we'll be able to also Get customers onboarded Let's review that one right, let's review this one what's what's going inside the payment Abhinav Shukla.
Hey, bro. Hey, hey Abhinav. What's up?
welcome to the stream man let's actually really discuss about this thing guys like uh what how do i can do stuff in mobile payments like how noticing can take payments on android in india and the us without handling over 30 percent of single google so i'd answer 30 is not what you would pay and one place where linking to the website really saves money iphone in the us is now fully allowed okay that's good are you a regular coders or why i'm a regular coder i use wipe coding but i use supervised wipe coding so uh we discuss about a lot of concepts in this channel guys like uh you know we we literally go of the architecture we we drive these agents to do and i use this tool called asklaw that i built for managing these many agents using this orchestrator that i created chief so it's not that i'm a regular wipe coder i use a lot wipe coding but then i know every single piece of code that they are these guys are these agents are writing and this tool i have built as cloud .ai we can it can really help me connect multiple machines of mine i can access all these agents from my phone and right now we are building noticing .com like a second brain kind of application and while building that we're discussing about these concepts right and that's how like we are making it
so right now we're discussing about the mobile in -app purchases because they take 15 from us straightforward so i'm just discussing what's the best way to do handle payments on mobile how noticing can take payments on iphone and uh okay in india and the us without handling handing over 30 without a single rule broken so our answer to this 30 is not why what we would pay the one place is uh linking really lives i do wipe coding that's great if you do wipe coding but i would suggest if you can if you have time and if you really want to go deeper into concepts you should also start learning about few things right not in too much detail too much technical but at least you should have idea to how does things works so that if if you want to scale it then that will help you to scale okay so myth it's 15 under 1 million a year before store takes 15 apple the small business program is there and google a 15 tier outside the us so on the google in the us 10 plus five percent billing fee and i never knew all these coding uh this but plot and codex changed my life that's great iphone us link to the web so since may 25 us apps to buttons and links to buy on web
with no entitlement and today apple takes zero zero percent on them while the code set a fee stripe cost around four to seven percent that makes sense pay billing uh is the better deal linking out uh on android cost about uh pay about as much as you play itself in the us google it still takes 10 on a linked subscription within 24 hour in india alternative billing only cuts four points add stripe and reserve pay it saves zero zero two points uh for a lot of work okay no in app way around apple so outside of outside the us and eu and japan and korea an app may not link button or nudge to web purchases that we may always do email our users about web plan and outside so basically it's like it's suggesting that in us you can do that option because there are a lot of cases against these guys these apple and google that they have to kind of like do that right have you heard about twin mind i have heard about it it is uh i've actually checked on the learning page the way they work is little little different from what so we are more like a data structure and we manage our data at the end they normally store all the things in md file sort of like
Can you check out my website project which I developed like before three months, I guess I dropped this project So it's not fully working. You can do a little review on this only on front -end and other things I can improve I'm in sure sure you can post your project on discard So it's saying that oh, yeah, so it's basically saying that For us you can do this.
I'm like, I'm not sure like so it's like a brief rules as for the Apple United States link out only 15 percent uh sbp zero percent today link i think we can do that linking are taking that link back the nine the ninth circuit lets apple require parity in size for example like between iap and the link when uh both are supreme court petition pending look at the i mean there are like supreme court petitions are pending if i posted it sir i'll just be quickly review this and then we'll review our project apple india no this cannot be work like look at this like how diplomatic these companies are like apple in for us they are sending because there is a court case going on so it's like very bad no buttons links to action uh to other purchases methods in the app i certainly metadata yes with entitlement yes 15 s bp What's that?
IAP? What's that IAP fee? What is IAP fee?
I don't know. 10 % SBP on sales within 7 days of the tap in app alternative. Since 1 October 2096, one set of terms attachment chosen options kept 12 months only reporting.
You know, when you're starting, when you're pretty new, you can literally put these links, but then they start removing your applications when you grow. Yes, enrolled. Yes.
only alternative billing is besides play okay so one rule for current setup so let's see what it is recommending to us today the all the apps sell nothing right yet a web plus plan raises rents limit in them apps multi -platform services allow using the in the app uh what has what was bought on the web provided those items are also available as an in -app purchase within the app apple applies uh these loosely to apps with real fee use and no purchase talk but the reviewer outside uh the us can cite it adding iap close classes for it's good google has no such clause its payment policy governs app payments made in the app okay in these kind of applications your database storage is local is lot uh right so what have you thought of for that like we have discussed about this when i started building working on the database schema but you i'll i can explain once again already compatible apple requires credit card bought by iap never expires rpac already never expires subscription .ts grants them expires at null
monthly plan credits are the part of subscription not the purchase i think okay let's say it has created this nice table where i can literally see what each route actually cost so web directly today six point six percent three percent and pack is also 6 .0 so it's basically the fee of stripe stripe also charge five to five to ten percent five to six percent which is also bad you can go with this path plus uh five percent of apple small business drop is a proposal approved essentially we can reduce this to that iphone iap which is basically 15 anywhere in person small business program you can control yada makes hi yada makes welcome to the stream once again i think you you've joined before as well so android us link is 16 i don't know how to deal with these guys man i mean like they are doing monopoly so we're launching platform on their website they're already charging us 100 apple is charging every year google is charging 25 dollars a single time fee but still they want more money from us this is so bad options and decision metrics so
I want to test my game if you can. It's called Flappy Bird, only on Play Store.
okay i don't have android as of now for now so i can try once i have android phone i'm actually looking for looking to buy an android phone so this is what we can do it is telling us to do so it's saying that get more from ren plus 500 credits continue on noticing .com subscribe with apple so two options we can give it and then iphone outside the us every android it's like subscribe here and restore purchases okay your plus plan anyway plus works on web iphone and android you can subscribe at noticing slash belly through email you can literally send this because it'll be good that they'll be seeing that what it takes to build no code yet store purchases both apps adapters and us iphone links it's kind of like given us all of the possible scenarios recommendation and fallback rewards a decision by the way you decided the apps sell nothing billing readme um the hybrid needs your word so hybrid approach they're saying that basically us iphone continues on noticing .com zero percent apple today beside an equal subscribe with app button every other phone one tap store kit
billing uh stays cheapest and the source truth one ledger revisit android india just use blue stack android emulator i i suck i i don't like that tool that takes a lot of ram okay the best phone for android about 500 is a google pixel 10a gotcha i was also thinking of going with Samsung, but if you're recommending that I'll just check that as well.
Where are you watching from yada makes If I be must wait till ahead That's coming to my call so I'm gonna actually ask one thing Chief are you listening to this who is listening to this as law page mobile payments and strategy?
and actually go back to this i'm actually directly ask this can you can what will happen let's say if we just do with link with uh pay on website directly on every android and ios no matter what their their term says is there a chance of getting banned or application getting removed because we are so new right now so i don't think they'll capture us what do you think i'm trying to hack it around because maybe let's see what happens right this is i've done for one of my project i'll go for cheaper around 415 samsung galaxy a57 5g thank you so much for your recommendation let's see what do what do they recommend i mean i'm okay to go with policy wise but i think they have to him's question what we have link short answer we would very likely be caught that's bad so probably at at the very first review and the damage would be much bigger than money saved being new does not hide so being new does not hide us it makes us even more visible why would we be caught every ios build is reviewed by a person who taps the app a pay on noticing button outside the u .s front google scan every release automatically for links and payment and hide it from review is worse
showing the button only on remote flag what happens when you get caught usually first the update is rejected until it is fixed no update ships oh it's like not even a bug fix there's a critical path to go in i would not like that because what would what it would it would actually got the us phone of uh iphone link is already allowed in the plan at zero percent to apple everywhere else breaking those rules 15 minus the stripe or razor about 95 a month per plan a thousand a thousand indian subscribers is 95 okay so what about like um if we just don't show anything related to payment on phone we just simply show like Your I mean your credits are now over just like that.
Check out noticing .com. How about that?
Again whisper flow sucks. What is this problem with whisper flow that it doesn't allow us to paste thing in here?
I mean what I'm trying to say is that Normally these reviewers would know would have credits, right? So yeah, what do you think?
Guess where are you guys watching from like I just wanted to know I want to understand where my viewers are coming So can you get quickly tell me I don't know about Pritam and I like the other maze Where are you watching from? There are five five people right now watching if you guys can tell me where are you watching from?
That would be great let's see what it has done now this is the second question the idea is close to a good one saying that but name but naming the website is a part of but naming the website is a part of that breaks the rule change one sentence and it becomes fully allowed so outside of the us your credits are over check out noticing .com So, Shivam is watching from Ghaziabad.
Kyrex is watching from India.
So, it's basically saying that outside the US, we can do this. No price and no button doesn't help. Pointing someone to where it really breaks the rule.
So, that's bad.
I am from Uttarakhand, Dehradun. You have used this month. then gets on december until everything still works your tables and everything so it's something we can do there this allowed everywhere something like this and then maybe they can go to the web that's the problem because people will be consuming the tokens and then where it fits us iphone the same place okay but there is one problem people who are downloading direct application they won't be able to uh pay right they have to how do we manage them the big deal guys because giving these amount of money to these guys is not really good uh like can you imagine 15 is not good that's giving extra 10 on top of a stripe how do we manage our margins a lot of technical discussion very less technical discussions they're more like legal discussions but i think sometimes we have to do this by the way let's actually check what our agents are doing that agent actually testing and fixing mobile bugs that's nice third question this is nice that i can see email them allowed strongest in app purchase without it someone who use only use the phone outside the us
iphone can pay only if they find the website on their own there is everything you may we may do them and where it stops email them make the website home allowed and it stops push notifications and it's basically email them is allowed make the web the second home is allowed Push notification. We can't send them for related to purchase.
That's bad And by the way, if let's say we have to pay this fees to Apple pay, how does it Apple pay Google pay? How does that normally works like is it like We don't have to pay to stripe then and how does that how does that money comes to my bank What is wrong with that there's so much restrictions, I mean Apple and Google both has these restrictions So they but they say that if you're creating let's say if I'm creating my platform noticing calm and if I'm having this pricing I'm having this pricing of $10 per month.
They'll straight away take $3 of that Not $3 now, they're saying that they'll take 15 % It's not 30 % because they really they were like three three dollars, right? How can I even earn money?
But they are very bad. They are like monopolizing everything. Taking 15 % of your revenue is very, very bad.
Okay. So they are saying that for in -app purchases, they do it like this. we don't need to pay for razor pay and we don't need to pay for subscribe all we need to they just do monthly payout apple it takes around 45 days to reach your bank like buy subscription in web series you don't know guys if you will know how much netflix and amazon prime has to fight just search about it with these guys so they'll basically uh i think fee is so apple store is like 15 small business program and google is like 15 and then stripe is also like two to seven percent stripe is the best way to earn money but let me see what is the hybrid approach what was the hybrid approach u .s phone via the web everyone else one tap iap rejection low best money per unit of risk in order to uh have these payment in -app purchases enabled do we need to take any additional approval around the application or like existing approval will also work this is also like important because they might require additional approval i think the best way is to show inflated prices on phone so you know how uh youtube premium works it so they show inflated prices on phone on app store and when you purchase from website it they charge less
is askcloud .ai is a website how can i access it it is answering in a very well structured and formatted way of course i'll use my api and buy a subscription you can you can down you can access it from ask uh ai pages oh sorry askcloud .ai it's not a website by the way it's you have to download it for your machine access from here you have to you can use your existing cloud and codec subscription but you have to pay for a subscription uh to ask which is around uh 8 .49 a month and if you want to get a lifetime access then you'll have to pay this amount i don't know what to do man i don't know what to do i need to consult my legal team i think for this i need to consult my lawyer or some they might help us because this is very serious thing here but i think we'll have to go this path so in -app purchases uh let's actually see what answer do we need to go through any additional process fifth question if it isn't on today the current approval doesn't cover it
it ain't on today uh the app have no code yet nothing it uses it it it's not a separate permission to apply it though it goes through normal manual review of the next app update okay this is basically there's no legal approval process we need to kind of set our payments and all of that there so for small so for large business they still take 30 percent for small business they take only 15 percent that's very bad that's very bad and by the way can we do this thing this way can we show inflated prices on apps when when the users are purchasing it from app we'll show some different prices maybe inflated prices so that we can manage our margins and on website we show redux reduced prices would that work the difference is around seven percent the stripe is charging five to six percent and they are charging like notes like i guess around nine nine percent so you're giving extra ten percent on nine ten percent to these guys sixth question answer let's see so that's how i brainstorm with these uh using this asklog pages guys this is really good and this come out of the box with ask .ai
Yes, that works. It is allowed on both stores. We set up our own in -app prices.
Neither app and Google requires them. So web price is currently $10. We keep web 9 .34.
App price this and we keep 9 .34. Oh, what is this? Web price is this.
We keep web 9 .34. App price is 10 .99.
and we keep app after 9 .34 so it's basically saying that you we we can charge 10 .99 like one dollar extra bro is the chief open source no it's not open source it's closed source you can get it from askcloud .ai it's a paid tool i think this makes more sense to me what we may say use us iphone 10 on noticing .com and 10 .99 with app apple side by side on save on noticing .com for us you can do this everywhere else so on only the store's own price i want the i want the contact pipeline and worker tab that you were using Yep that like now you can see chief is testing this application So you can get that with s clog If you if you buy like s clog you'll get that out of the box and I can actually help you Set up that as well for you So, I mean this is this is best you have a subscription how much is it it's uh eight four nine per month and if you want to buy the lifetime license then you can go for like eight four double nine it's it's available for limited time to be honest like so you can try it for one month also
this looks good some people will see the higher price and pay nothing rather than go to web watch it if the one goes so some people might think i look up as clock do you own yeah i own the as block some people will see the price higher app price and pay nothing rather than go to web watch it if the buyers are rare try equal prices on for a month and compare i mean for now i'm gonna go with this i think i'm gonna go with this this option so can you update these questions uh your call accordingly i want to go with like showing inflated prices on google pay like on phone and they always have an option to buy on website also so for example for us wherever it is allowed to pay with uh pay on noticing .com that also can be uh must be added so i want you to start like two different agents working one one working on the google app purchase and a codex agent to set up that configuration on google app store sorry apple app store and one codex agent which can do the setup on and you know all the licensing and everything can guide me on play store so i want to do this thing
how did you manage this commission thing for asklog android and ios so asklog android and ios uh we don't have to do that because asklog see asklog is a platform where they can literally come and people will directly be downloading from web first right they'll be downloading desktop mobile application are secondary mobile application are like additional functionalities so i don't much worry about that i mean they are like don't have to show anything related to payment on mobile application they are secondary right and they can only access mobile application if they already have the subscription on right they can only access then free connect their devices android and ios so that's why like it's like anyway like not a problem but here it is right so yeah so it's basically a hybrid approach uh higher prices in the app just enough to cover this thing and then both equal size web first awesome Let's send an end I Think I have not You would these on the oh my god literally He chief I accidentally closed that agent but I have not I think I'm not cute my answer so my answers are straightforward send that details to that agent because I Want to use higher prices hybrid approach?
wherever possible alongside with apple we can show that link pay on web and where and then i want these two uh play store and app store to make sure that they properly set up my app store account for in -app purchase and play store account for in -app purchase alongside with that they should also make sure they apply for that 15 small business right away in front of me so that i can see it going on so guys like this is how i use chief i use these codecs for my computer use because they are good in computer use now they are going to do basically do login stuff and then they're going to do all the setup requires and they're going to implement that in -app purchases for me and that's amazing that's what i want and by the way one thing i like about apple and google play store and this thing they try to i've seen this thing i've got a lot of feedback that they if you use their in -app purchases they actually promote your application which is good and by the way mobile testing is going on as you can see right here it is another agent is doing mobile testing this mobile bug fixes it's doing couple of mobile testing i don't know i think i still don't like the mobile ui that a lot but i'm just just trying to get the most part of done then i'll be reviewing it myself and i'll be giving feedback to mobile and app both that's how my idea is so you can see the ui doesn't look good in terms of mobile
especially the data management related but i will let that fix not a problem they can you can see everything is automated right i am just here it is testing the application actually bring it right here this is doing all the things for me how good is that But I wanted to get this thing this payment thing settled today because then I can make my application live It's already live and because of this payment only I'm not able to make it alive.
So I just wanted to get this done Now literally we can actually see if any of these agents I started it's actually not started yet So, you know what the best thing is that now he will Tell them what to use how to use and accordingly they will be basically doing stuff on my behalf They'll be accessing whatever they need they'll be creating all the stuff so I used to Have you have to do a lot of work manually initially, but now we just can outsource this to chief That's the good thing about it And then I'll be launching a new version to App Store and Play Store and then we're good just a second i'll be back Thank you.
do my do you mind sharing your um cloud setup the multi -agent setup that you've created especially the qa agent uh outsider 176 okay so outsider 176 um i'm using uh this tool called as cloud .ai that i built myself and in that i have got i got this setup out of the box so i have this chief which i talked to it orchestrate all this agent i don't have to manually do any any kind of setup i only download have to download this from as cloud .ai right from here so it's basically doing all all that stuff guys so i'm gonna let me actually quickly do a quick sign in here so then i can literally then you know these agents can work autonomously hold on and do this do this cancel this it's not what i want this one heck why the heck it is showing just a second guys i'll do a quick login so let it do the work in the background i'm gonna open this right here Now that you know in the background, it's basically doing all the setup requires I Don't even have to worry about it So here you have to go to go to this as cloud .ai slash install You download it for your Mac Linux Windows, whatever you have and then you can also access that from your phone.
That's great So then you get this this software, which is pretty cool. It has a lot of functionalities like it can literally Hide the stream. Okay, it literally telling me to hide a stream this is the checklist that i have to follow and actually open it on the browser first then actually close this here and then we have a checklist for here as well so look at this how nicely it is doing that stuff for us i'm gonna just close i like to see this as clock pages by the way on my on my browser instead of that on so yeah so when you do that you literally get this kind of setup where you have multiple workers running right and then every workers and everything is all your projects is being managed by this single chief like that i can put right here it has its own brain sort of like you can see this brain right how it is it keeps growing right it has a very nice memory when you get this out of the box and then you can literally tell it to test your application to create a work and it uses existing cloud and codec subscriptions yeah so it doesn't require anything else right
and then it can also you can connect multiple machines without doing ssh you guys doing tail skills setup just by logging with the same account you can connect multiple machines so if you have mac mini you can run 24 7 uh using this your approved strategy page already uh set the yearly price at 109 not 105 both store workers are using that only prices holding them until you confirm so it's basically saying uh 2000 credits 30 dollar web and 2499 yeah it's basically trying to access uh saying that app uh okay so screen warning hide the stream okay so it like the chief is so smart that is saying that since you have live you need to hide the stream because this is something confidential will be entering which i understand not much worried about that let's see that uh finish the apple sign in privately then say i am signed in already you can proceed on apple let me actually see if it also tried signing on my i didn't actually do that i've used cloudflare style d1 database for my tv hi i kind of guessed that it's good it's good shivam not a bad thing it's good i'm not receiving the chats on my phone now programming with current is saying bro just took out of two minutes can you tell me all the feature and show me uh show them in as clock i just want to see chief worker tab
Another taps. I'll just take the subscription right away Okay, the first thing when you start a project inside this you can literally Manage multiple multiple subscriptions here you can you can have this option where you can add multiple accounts here you Beat your cloud beat your codex and then you can literally manage to you and then you also get by default the voice detection i use uh whisper flow because i am like fond of it but you get also this by default so i'm keep improving this so it's very nice experience then if you already have codecs and cloud that's all you need to do i mean you'll start it will actually guide you this chief also will guide you this chief you can put on the right side and you can also put it on the on a dock view so this is one way and then you can just uh close this right view which normally i like to use it because then i take can take it out of like uh as log also and then you have this file manager itself so you don't have to ever leave this application you can see all you can navigate through all your files literally you can see github changes by the way when chief assigns any task let's say for example this manual identity parity it literally assign those work tree automatically and then you can literally see what's going inside that all the changes that is there so it's basically very smartly does that and then
You can you can pin your certain projects. You can literally connect so you can see I have my Mac mini connected So if you just don't log in in the same account that way you can literally connect that that machine will automatically come here and you can access all these Agents without any just with login from your phone also So and on the phone we are keep improving the experience, but this is like like like thing you get out of the box i have inside mac mini a lot of projects are going on inside my macbook pro i can see this project is going on mostly i'm working on noticing project so that's how it works and then you can anytime uh see the memory of your cheap what is it storing around the projects around yourself and you can literally search for those memory right so all of all of that memory so if not not really you need to get to this place but if you just want to debug what's going in behind the scene right and that's how it just works I hope that has given you an idea.
Programming with Karan. Right now, let's say I can go to detach view. What are you using?
Seem to be consuming tons of token. No, no, no, no, no. That's the myth.
And I think I've explained it multiple times that. See, when you use your own cloud only, then it creates a lot of sub agents, right? See, team actually uses less tokens.
And by the way, you get this out of the box also with Chief, this AskLockPages, right? And this AskLockPages, if you don't know, I can literally communicate with this and that sends back to the agent. So all you need to do, whenever you're creating something, you want to brainstorm, tell the Chief that can you start an AskLockPage and it automatically will start an AskLockPage for you.
And that's pretty amazing. So token -wise, it's even lesser because the reason why I'm saying that for every single task, it starts a new agent, right? if you have let's say even 20 dollars plan right obviously you will work on feature single single wise right individually so you can just assign and it uses less tokens that is my experience i have done a benchmark around it as well so yeah so i think now uh it's basically saying see it manages everything for me i have told apple app store worker to go ahead shubham it reloads app store connect in your chrome someone join discord oh i think they joined nice oh actually we have to review one project as well let me actually do this kind of stuff a little bit then we'll do that uh keep the stream hidden i think apple is signed in it is doing stuff yeah my stream is like actually uh already hidden you just take decisions only on your own uh i trust you and then everywhere the prices should be like i should be at the end getting the same price so you need to inflate prices everywhere even if they're purchasing additional credits and if even if they're like um on the price on the yearly subscription as well and the monthly subscription as well that's my decision waiting for uh 266 memory fix to go green then deploy
As there is a bug in web then I think that's what it is waiting for that's a recap So it has a smart feature that it sends me recap all of that. So it is hard -coded in in its system instruction And it uses something called as clock CLI under the hood which makes it possible for it to communicate with agents and agents can communicate back All of that happens that way now i'm actually mute i'm actually mute on uh discord i don't know if he's here there or not he just left so i'm on discord if you guys want to join discuss i can help you out play as everything else is needed and including the final prices one step left for you okay play but where is he managing play oh it's actually doing where is he managing play store i think play store he is managing directly where i don't know these are the basically checkpoints like where i need to work on it's basically like payment details i need to add and all of that i think most of this i can outsource this to achieve but let's do that later on first let me just gonna see what it's saying because i have to always communicate with this agent this chief because i cannot like like for example it can be a little confusing for me right so it will guide me actually play has everything it needs
including the final prices one step left for you because only you know which legal entity to use okay that's good in the google payments profile uh tab in play console okay pick the legal profile but where is it like i cannot see i cannot see the play store one can you help me out open that in my browser i'm not able to see that which which profile you're using and also the apple one um can you tell me what do i need to where do i need to click do you have majority 50 corporate individual or partnership interest in ownership or shares of another uh no i think i don't know how but i have to pick this these are very confusing terms if you see oh it actually opened it here oh my god it has opened over here on my different uh provider altogether oh payments profile i think i have to pick uh payments profile uh organization and this i can pick this one by setting up billing give the legal name i think it's actually asking me something some details That's very confusing guys like actually right now I have I don't know multiple or open I think for these kind of job when we are doing such something like this We should have like only one profile open and then doing this thing Let's see, let's see.
Let's see. I mean actually it is confusing but let's see It has a lot of things open by the way gonna write some details here I've selected the payment profile in play in that tab so you can continue now and in in in like app store I'm getting infused with the where do I need what what options we need to select can you just do that for me in front of me Let's actually go out of the for a second Not many things are open This is again guys, I'll just do this setup analysis Let me pause the stream for some time then we'll come back Let me finish this Let me actually close all of these because these are confusing me.
i was having a lot of chrome browser session open so that was confusing start the process once again i'll now i'm just actually having one on one only in the as your goodie profile so use that and then do this thing and i've now completely paused my stream right now something confidential is going on i mean uh it's setting up all my agents and everything so that's why like uh it's kind of like taking time so yeah it'll take some more time then we'll come back by the way This agent is managing completely my setting up my app store place or everything.
That's crazy, man. That's crazy It's really really crazy I put this that's doing all the setup that is required is there anyone watching right now i don't know which is having a one viewer oh that's great guys if you're watching on twitch or on youtube like i don't know if someone is watching can ping me hi on the chat i can see only one three viewers on youtube and one on twitch that's great do not other place okay so is linked to enroll into the google ask to yes or no uh questions does this the first answer is no uh we i only have this only account does anyone the developer uh no no they do not no no other developer published with same branding have accepted everything in apple can you just quickly check okay i think no one is watching guys i have to get these setups done today because i want to launch it and these are the road blockers for me to launch it's doing some license testing i don't know what it is doing whatever it is doing it is doing crazy stuff to be honest like and i hope uh if in the meanwhile the
development work must be going on right around app purchases and everything i hope so and i hope so all the work around development must be going on behind the scene because right now we are setting up these so yeah development work should keep going on because we need to ship the complete payment on both applications today so guys if you are watching right now i'm actually doing some something confidential so i'll be right back Because I'm actually setting up payments and all.
Like payment gateways and all.
Just hold on.
By the way, who is watching right now? I don't know. Somebody's watching.
Hello, hello, hello.
I think Apple, I've already accepted all the things. I just need to add the bank account, I feel like. Now in Apple.
I think in Apple, I just need to add a bank account now.
And on both. This also.
Play Store also. Let me know if everything is finished and add the bank account details. I've already grabbed the bank account details with me.
So it was really hard to set up these things initially, but now it is very easy with these guys like Chief.
It's so easy, literally.
I'm going to offline on Discord now.
Thank you.
Bang details I can add myself But I'm not sure i think this is done all the payment related information is done it asks a lot of details let me actually reduce the volume a little bit so is anyone watching can you guys say hi in that chat if anyone is watching right now if not then i might close the stream for today It's gonna close all these because you see how many things I think I'm live again You're live I'll told every work at a screen play give the server access to play api yes or no so this external links terms yes or no give the server access to play api yes accept the us term yeah i think uh both two years both yes let me actually do one thing let me actually bring this down and actually open only this thing over here so that we can actually see that view very confusing otherwise i like that guys you can literally uh set up everything from you know you can get everything all of these things set up using um this thing i'm going to remove this payment setup now
saving everything so it's basically kind of setting up the payment still so these two are working is the deployment to uh pricing uh web is done or not is it still remaining i want all deployment to be done today now the plan is now the plan is like this we'll release the app store and play store both today one version and we can also remove the invite code i think we'll only have invite code for free users for now on web and we will have for the paid users they can literally pay on web like on web noticing .com and then um they can select a plan and they they'll get direct they'll their account will be created directly because i don't want to hold the paid users to access the application yeah i've taken the google chrome screen down so that's just fine for me currently i'm only showing the uh as clock screen to users also does this um that uh pub sub that we created does that come up with the cost on google cloud so guys it kind of like set up those uh google pub sub for notification of monetization for me i'm not sure whether that come up with the cost or not i hope my mic is on yeah it's on
So that's how it's going to release my application to both App Store and Play Store with proper payment integrated. And that's what I really wanted.
The best thing is that it can keep working on the background.
effectively free google okay so effectively it is free uh i think that's enough for us if it if it grows that that will be great yeah i mean uh these these uh both app store and play store they can keep continuing setting up things like yearly plan monthly plan everything i'm okay with it and get everything done and we're gonna make it live done confirms three external links fact contact details ps pci subscribe and raise a pay uh and a dispute process through refunds that's fine that's great anything that needs to be updated on privacy policy support or anything just update that and also use hello at askcloud .ai email everywhere We'll buy an email for noticing also later on, but for now use this.
This is how it looks like setting up the whole business using chief using gas clock a payment stuff and everything right This is really cool You can see Okay, so it's gonna open the play console for me the monetization is set up Because I think it's gonna add I'd have to add these things there I mean, it's actually managing everything for me.
I can see people are still watching. Like five people are watching. That's great.
Guys, whoever watching, can you send hi? We just literally completed our app store in -app purchases setup. I was off screen for some time.
But I would really like to know who is actually watching right now. If you can just ping a hi on Twitch and... i think we have one viewer on touch as well now paid agreement yes all for ownership yes this is great have you also applied for app store that 15 % one plan that you were mentioning yes submit hmm awesome is the mobile bugs also fixed mobile application box so that we can release applications yeah just fix all the conflicts resolve all the prs and just merge it and both of these agents by the way running on like fast mode that's cool that's as important as needed i think the one only one use case i can see of codex uh and chat gpt subscription is to use computer use literally like computer use they are the best in computer use and browser use also better than cloud Thank you.
okay so looks like we're good in many ways let's see it's shubham the two play play forms need you since won't these edit worker contact details okay all right so where do i need to put contact details i think contact details are already good i can see it correct so apple sandbox texture you can create yourself and then let's just actually um you can put that in one pass Use the email account of as clock and then adding a plus Gmail account adding a plus Maybe upper test or something like that but this account actually has clock yeah in -app purchase build what what is this in -app purchase build doing is it not like completed or it's still remaining the hell is this agent doing i actually see that it's actually working that's good Okay, keep grinding.
What else we need? We have set up the payments. We need to do a production test.
We'll do that later on. We were opening some pages, I think, right? Something was pending from our side.
It's not pending. Let me just quickly check.
anything else is pending from our side yeah we will test the ui bugs but that's fine you can check that later on i would like to uh once this in -app purchases are done i would like to generate a apk i mean aab file and uh uh test flight for this application to release today itself and the web also to be released web deployment is going on or is it like it's still remaining by the way we need to still need to revamp that logo of the application doesn't look good professional by the way i want to do meanwhile one thing i would i would like to understand uh for the startups aws google gcp and even cloud and uh opening and they give some startup credits right so how can i avail those can you create a s clock page uh for all the in -depth research that i can get access like some startup credits for all of these platforms also i need a admin panel i you know we started an admin panel development where you can track every user everything all the payments in one place so what is the status of that
What is the status of this paid skip worker and why didn't you respond with admin questions of mine?
But I want to host that also inside a pod. But that pod should only be accessible through my tailskill. Some sort of that configuration I want.
I want to host that in a similar Mumbai region. All the applications should go in Mumbai region only.
Is the application deployment done? Bye -bye.
Let's actually see the status of PRs. I hope no PRs. Oh, there are four PRs still.
I think these are related to mobile. All of them.
That's good.
Is anyone watching from Twitch?
People are coming and going. People are seeing what the heck this man is doing.
By the way, guys, I just want to wrap this up and I'm going to go for sleep. I'm feeling very sleepy right now.
i just want to complete this milestone for today that payments are set up in app purchases already set up the admin pod is built until skills uh sidecar inside the pod okay that's great policy okay i think i'll have to can you also open this tail scale setup on my browser and get this done using codex agent Where is it setting up?
Is it really setting it up? Alright, then I guess I start it. Ask for MFA, then I'll just see that.
Tailskill, I've already logged in.
amazing by the way this admin panel should have a consolidated view of all the payments that i'm receiving all the user what are uh where are my tokens getting spent on what what the user usage is everything should be tracked in that and also have a have it integrated with postdoc events so i can see the users activities also in the application that's going to be a next task altogether but yeah time for now what's stopping us to keep giving tasks to these agents agents which are already done they can be closed now i don't know why i have to keep reminding uh chief regarding the agents which are already done to you know just i don't know how much time this is gonna take man This is fine.
I can wait for this Play Store guide. I can see this Play Store guide is still working on something. I don't know why, but it is working right now.
And I cannot see any, like, is the applications are already ready to be released? Because I cannot still see that. The work is completed and reviewed.
I can see the PRs also opened. Get the work keep going. I mean, I can see only Play Store setup guide is there.
Do everything man, please Keep the work going on literally I this is very important. I Want both web and mobile application to be deployed on place for today an app store people how's the stream going i mean it's going good like were you still there or like you just waked up again so we have kind of set up the in -app purchases and now i just want to distribute and launch those applications so yeah i'm just waiting for that i just wanted to complete this target of launching it to uh with the payments today and i make sure that happens so that's why it's gonna kind of like grinding so i think you're the only one who is watching and there are two more people Saying you're right Shubham work is stalling on CI again PR test kept taking our two runners Just Just remove these tests for now just release it out we can test it later on Stop these checks and on GitHub actions as of now I Woke up again.
Actually, I'm having a fever. Oh my god, so you shouldn't you should take rest man Take a rest Fun and enjoy 6771 you don't sleep at night.
No, I am like a owl who don't sleep at night i'll sleep i mean i just want to wrap up this payment stuff and then i just want to go for sleep there's literally i'm not doing much i mean these agents are doing but yeah the startup credits page is ready so i've also looked for them startup credits can you can you open this page on uh browser sometime i have to keep telling this chief that okay this page is ready this page is something i want to evaluate right now uh okay every startup credit book worth knowing and incorporated so we have like microsoft for startups mixed panel for startups I think this is something which you can do later.
We're gonna apply for all of these now. I actually want GCP. I don't know if they provide Google for startups cloud.
I think this is what I need.
I don't really need MongoDB and all.
aws credits for startups is good but it's invite only only i don't know if these can be used for i don't like aws to be used as a like you don't have office today no it's it's sunday man the saturday man guitar for startups okay open ai for startups not published this is something which i want plot for startup let me actually ask this agent right let's actually ask this from this list i want to remove everything i only want to have let's say um one which are related to cloud and uh ai credits that's all i think 11 labs is also there by the way 33 million characters can do we can get this kind of like you have to have less than monetization plan business email no bird products for miners 12 months i think 11 labs also makes sense you should apply for those as well and i when i am applying for let's say uh microsoft or anyone i don't want to get vendor logged in i just want to integrate that those to my kubernetes nodes and then then we'll use those by the cloudflare i like infra ones i like for aws activate you need you need a partner that's bad so okay
now we're left with uh this one and then we have deep gram i don't know it can be used for voice ai i can literally tell my uh okay save these notes we'll be applying these to tomorrow or the day after tomorrow i just wanted to evaluate if we have anything because i like to have use google cloud 350 vc funded you need a vc funded that's bad start plan can really help founded less than two years working mvp plans to raise email domain website domain gemini vortex is covered but gemini is also covered that's nice i really like to go with this i'm gonna apply for this for sure i think this is only related to ai studio which can be helpful well let's see let's see what's going on here okay the startup one i i have saved can you save that page maybe the day after tomorrow after tomorrow we'll use that we'll apply for these all the startup give me the status of web and application builds where are we with those i'm feeling sleepy now Okay web not live yet checks are switched off so nothing is blocking on test deployment is running on which is great Apps merged Only you can do it one is generating app purchase key into one pass a store App store that oh, yeah, I think this I'll have to do create a sandbox tester with autofill.
Okay admin console pr is opened it's good go and sleep workers keep look at this message guys like how nice message is this so on me only this uh ios is stuff is blog right so can you help me doing that on my browser i would directly like to do that so help me uh what do i need to generate and how do i save that take me to the page i like to do that and yeah i think you guys can keep working i want you to upload both ap like aab file and app store for review and also this admin console should also be deployed on my tail scale that's how that's how i work guys with chief like i literally give them ask for the overnight and then they will keep doing that stuff then guide me what do i need to do on app store right on this browser actually sorry you opened the wrong browser do it now now then they're gonna do that always use chat gpt for such things always use codecs and chat gpt for such things and it has an option let me actually name it called noticing in app purchase i have downloaded the in -app purchase key uh it is there in my downloads put that in one pass and yeah
start using that anything else needed from my site now it's going to access that key and store that in one pass so benefit of one pass is that it can also help me automating a lot of things right i could have like uh just have to do some things manually if i would not have that in mind the idea is that if i can literally get my uh this thing done no then i can just i think they'll just review over the weekends and i can have my application up and live by monday maybe i can see new key in the downloads and i think that's great anything else from your side sandbox test users codex agents can creates it if agent rules could block the password steps i mean just use computer use of codex right in order to do that if there's someone who's blocking hold on one second guys the browser tool blocks the step key and then just i think that's good i need uh okay so what's this i need admin panel on my tailscale web live and uh these applications submitted today that's my target
i've gotten really serious for this application guys i did not expect it uh to go such like this serious but now it's really really shaping good so okay the way i'm actually managing my admin panels i'm i use tailscale also for my private um network mesh so i'm gonna launch admin banner only on my tail scale okay so i think understood is running now i think deployment of this is already running and by the way we have set up our own github runner so if i go to let's say if i would say actually let's do that later on uh web is going to be live admin console on your scale mumbai parts as soon as you put as the oh i said hey okay tail scale key we need to put here both apps 1 .7 submitted for review okay this is something that the both android builds run run in parallel with the api deploy and the both are submitted once the api and in -app payments is live the codex agent handles the api apple key the sandbox tester and blocks just basically keep grinding right and today we've got 347 subscribers guys that's really cool we have gained good number of subscribers today from the stream i think very very soon we're gonna reach 500 subscribers and then it'll keep increasing i hope
it's almost five man let's see if anything needed from me do you need anything else from my side tail skill is in review is it or is it like already like merged you can merge that if it is not reviewed and you're saying that uh tail skill key but i don't see any action going on around that from codex agent i cannot see codex agent going doing anything can you just keep an eye on if the agents are really working or not okay okay okay final step let's see let's see i think from my side everything is kind of done and uh to be honest like the stream also has passed out six hours so i'm gonna go for a sleep now i hope you enjoyed this stream and i think uh working right now using computer used with one pass autofill and create the apple sandbox tester both uh keys are confirmed in the play skill key and the next i have queued the google play server key which is what blocks the play server secret uh i think everything is working pretty cool that looks nice apple app builds is compiling and uh builds it straight away after there's some same commit so it's kind of like working from you nothing right now at most 2fa or touch id tap if apple or google ask uh okay i don't think they'll ask or that's bad if they ask
just a second okay guys thank you so much for joining in i hope you liked the stream and i think today we have got like a lot of things covered we have my our own um scale set up now i think i can sleep peacefully now with the payment as now set up let's see if everything goes good and by tomorrow morning let's see if it can literally confirm and save everything on my system right well let's actually see i'm hoping that it will work overnight i'll just give it more some more tasks maybe and then yeah that's all thank you so much for tuning in bye
The Hook
The bait, then the rug-pull.
The title promises senior-engineer vibe coding, and the first story is the opposite: an orchestrator left alone overnight opened 78 pull requests. The next six hours show what supervision looks like when one agent runs the rest, from QA after that merge to a credit system and app-store billing plan the host checks by hand.
Frameworks
Named ideas worth stealing.
2:19:33concept
Supervised vibe coding
Let agents write the code, but brainstorm and approve architecture and critical flows yourself before any build starts; tell the agent explicitly when you are only discussing, not authorizing work.
Steal forany AI-assisted project that touches money or user data
2:37:07list
Chief build-review loop
Chief assigns a task to a fresh worker in its own git worktree
Opus 5.5 writes the code
A Codex agent reviews in the same worktree
The worker opens a PR; critical PRs get human review
Merge, then kill finished workers
The orchestration pattern used for every feature on the stream.
Steal forsetting up a multi-agent coding workflow with mixed models
53:03concept
Tool search instead of tool lists
Replace 30+ MCP tool definitions with one search tool that semantically matches the request to the right tool, then calls it, keeping context small.
Steal forMCP servers with large tool catalogs
2:18:22list
Credit metering rules
1 credit = 1 US cent of model cost at provider list price
Round every call up so nothing is free
Debit once per call against a ledger
Warn at 10 credits left
Pause AI at zero, no overdraft
Monthly allowance plus purchasable packs
The billing model drafted and reviewed for the Noticing app: free 50 credits a month, Plus at $10 for 500, $9 packs of 500.
Steal forusage-based pricing for any AI product
2:50:55concept
Never trust the browser
A checkout success redirect can be faked or lost; grant entitlements only on a signed webhook whose secret is stored server-side, and verify it on staging and live.
Steal forany Stripe or Razorpay integration
4:16:48list
Hybrid mobile payment strategy
US iPhone: link to web checkout (no Apple fee) beside an equal-size in-app button
Everywhere else: one-tap store in-app purchase
Price in-app slightly higher than web, e.g. $10.99 vs $10
Enroll in the 15% small business programs
Email users about web plans, never in-app nudges outside allowed regions
The decision reached after an agent researched current Apple and Google rules.
Steal forpricing a paid mobile companion app
11:12list
Memory types for agents
Semantic memory
Episodic memory
Observational memory
Working memory
The memory layers the stream series teaches and uses in both the Chief orchestrator and the Noticing app.
Steal fordesigning long-lived agent memory
CTA Breakdown
How they asked for the click.
VERBAL ASK
4:28:39product
“go to askcloud.ai slash install, you download it for your Mac, Linux, Windows”
Soft and repeated: the orchestrator is demoed all stream, then a viewer question about the setup becomes a tour, with the $8.49 monthly price mentioned only when asked.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
How a three-model review council, a score out of 10, and 18 hours of agent runtime turned a mid-looking open source app into one that beats Claude Code and Codex on design.
A detailed, ethically loose walkthrough of squeezing thousands of dollars in token inference out of a $200 Claude or Codex subscription, then running a dozen AI coding agents at once without losing track of them.
A walkthrough of a private, self-hosted system that runs every coding-agent task through the same process, so prompt and model changes can be measured instead of guessed at.