The argument in one line.
PopeBot eliminates the setup friction that prevents most developers from building AI agents by letting Claude write infrastructure code through natural language, making multi-user, production-ready agents buildable in minutes instead of weeks.
Read if. Skip if.
READ IF YOU ARE…
- A solo founder or small team operator who receives high-volume email and wants to automate triage without hiring a developer or paying for enterprise tools.
- Someone comfortable with OAuth flows and basic API setup who needs a working agent deployed in under an hour rather than spending days configuring OpenClaw infrastructure.
- A builder exploring AI agents for the first time who wants to see a complete end-to-end example—from spec to deployment—before committing to a heavier framework.
- A technical operator running multiple projects who needs a free, self-hosted option with multi-user scoping and CLI access for scaling personal automation.
SKIP IF…
- You need agents that integrate with non-Google services or proprietary internal systems — this demo focuses exclusively on Gmail and basic HTTP endpoints.
- You're building a production system for non-technical end users who can't handle OAuth setup or Google Cloud credential management themselves.
- You've already shipped multiple Claude-based agents and are looking for advanced features like function calling chains or sophisticated reasoning strategies — this is entry-level.
The full version, fast.
PopeBot is a free Claude-Code-powered agent framework positioned as the easier alternative to OpenClaw for building AI agents that run your life around the clock. The system works by chatting a plain-English spec into an interface that scaffolds skills, system prompts, cron schedules, and OAuth-secured tool access on the back end, demonstrated by spinning up an email assistant that triages Gmail every thirty minutes, archives spam, and DMs urgent summaries through a Telegram bot. The practical lesson is to lean on the agent itself for setup instructions rather than waiting for tutorials, scope each agent narrowly so it stays fast and undistracted, and use multi-user routing plus Telegram integration to coordinate authorized humans through one bot.
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.
Create a free account →Where the time goes.
00:00 – 00:28
01 · Cold open + promise
OpenClaw is a pain. PopeBot is free and easy. Demo target: email agent that kills spam, DMs urgent ones, can send on my behalf.
00:28 – 02:35
02 · Gmail OAuth setup
Add new OAuth secret named GMAIL_OAUTH_TOKEN. Google Cloud console: existing project, enable Gmail API, create OAuth client ID, paste redirect URI back into PopeBot. Empower-the-viewer aside: 'ask the agent itself for step-by-step.'
02:35 – 03:39
03 · Write the spec
One-paragraph instruction: check email every 30 min, archive spam, DM urgent ones with sender + summary, add a way to send email, create the required skills.
03:39 – 05:25
04 · Agent builds itself (Claude Code on the back end)
PopeBot delegates to the Claude Code SDK. It plans, builds the email-assistant skill, writes system prompt + CLAUDE.md, registers an AI-triage Gmail label, sets a 30-min cron, runs an end-to-end test.
05:25 – 06:40
05 · Admin tour — cron + agent secrets + LLM providers
Cron job listed in Triggers. Agent secrets page under Event Handler. Provider settings support Anthropic, OpenAI, Google, DeepSeek and multiple coding agents.
06:40 – 07:40
06 · Live triage test
Pope asks the agent to find an email worth flagging. It pulls one from his accountant, DMs him on Telegram with a summary, then applies the AI-triage label on command.
07:40 – 09:00
07 · Multi-user + Telegram routing
PopeBot supports multiple authorized users per install. Telegram bot per install. Profile nicknames let you say 'send Steve a test message' and have it route correctly. Bot only listens to authorized chat IDs.
09:00 – 10:00
08 · Interactive mode = Claude Code in the browser
Tab into a raw Claude Code CLI in-browser. Shell, file editor, voice-to-text in the chat box. Web chat, Telegram, and CLI all share the same thread.
10:00 – 11:40
09 · Scoped agents (anti-context-bloat)
Dropdown to scope a chat to a single agent so PopeBot doesn't get distracted by skills from other agents when you've installed 20 of them. Demos the underlying agents/email-assistant/ folder, system prompt, CLAUDE.md.
11:40 – 12:40
10 · Install + community CTA
Link in description, two-step wizard installer for Win/Mac/Linux, built-in upgrade system. Pitches the AI Architects community: classroom, Tuesday/Wednesday support calls, Friday networking calls.
Lines worth screenshotting.
- PopeBot wraps Claude Code's SDK so that a plain-English spec like 'check my email every thirty minutes and DM me urgent ones' scaffolds the skill, cron, and Telegram alert automatically.
- An agent that can set up its own OAuth credentials, write its own skill files, schedule its own cron job, and test end-to-end before committing is doing what used to require a developer.
- Asking the agent itself how to set up a Google Cloud project and OAuth credentials is more reliable than following a tutorial — the agent gives step-by-step instructions calibrated to your current state.
- Multi-user support in an agent framework means different people can interact with different scoped agents through the same Telegram bot without their contexts bleeding into each other.
- Supporting every major LLM provider and every major coding agent in the same framework means the infrastructure investment compounds regardless of which model wins the next benchmark.
- An AI triage label applied to processed emails creates a visible, auditable record of agent actions inside the email client itself — not just in a log file.
- The barrier to agent adoption is not capability — it is setup complexity, and frameworks that collapse setup to OAuth plus one paragraph of English unlock the majority of potential users.
Steal the format.
Tutorial-as-positioning playbook
Pick the hottest tool in your category, claim it's harder than yours, and prove it in under 15 minutes of unedited screen-share.
- Title formula: '[Hyped Competitor] Is HARD (My [Thing] Is EASY)'. Parenthetical flip does the positioning in one breath.
- Open with the promise + the spec. State exactly what you'll build before you touch a tool — 'kill spam, DM urgent emails, let me send' — so every later beat is checked against that contract.
- Hand the boring setup back to the tool. When the demo wanders into territory you don't want to teach (Google Cloud, in this case), tell the viewer to ask the agent itself. Empowers them and de-risks your script.
- Reveal differentiators inside the demo, not in a feature list. Multi-user, Telegram routing, scoped agents, voice-to-text — each one drops in as a natural detour, not a slide.
- Earn the CTA. Twelve minutes of giveaway first, then a soft two-stage CTA (install link → community). No urgency, no scarcity, no $$.
- One screen + one face-cam + no music = totally viable for a tutorial channel. Production complexity is zero. The leverage is in the spec, the demo, and the framing — not the edit.
Terms worth knowing.
- PopeBot
- A free, open-source Claude Code-powered agent framework that provides a chat interface, multi-user support, OAuth credential management, cron scheduling, and Telegram integration for building personal or business AI agents.
- OpenClaw
- An open-source AI agent framework built around Claude Code that offers extensive configuration options, positioned as powerful but complex to set up compared to simpler alternatives.
- OAuth
- An open authorization standard that lets a user grant an application access to their account on another service (like Gmail) without sharing their password, instead issuing a secure access token.
- Google Cloud (project)
- Google's cloud platform where developers register applications, enable APIs, and obtain credentials (client IDs and secrets) that allow their apps to access Google services like Gmail.
- Cron job
- A scheduled task that runs automatically at defined time intervals (e.g., every 30 minutes), commonly used in server environments to trigger recurring processes without manual action.
- Gmail label
- A tag applied to emails in Gmail that organizes them into categories, functionally similar to folders but allowing one email to carry multiple labels simultaneously.
- Claude Code SDK
- Anthropic's software development kit that allows external applications to interact programmatically with Claude Code, enabling custom interfaces and automation wrappers to run Claude Code sessions in the background.
- Scoped agent
- A Claude Code session or sub-agent that is given access only to the context and skills relevant to a specific task or assistant, preventing it from being distracted by unrelated agents or configurations in the same system.
- Telegram bot API key
- A credential issued by Telegram's BotFather service that authorizes an application to send and receive messages through a custom Telegram bot on behalf of its owner.
- Headless mode
- Running a program or agent without an interactive user interface, allowing it to execute tasks in the background and report results asynchronously.
Things they pointed at.
12:10channelAI Architects community
Lines you could clip.
01:13
“I like to show people this because a lot of times people feel like they need somebody to show them things — but you can use these agents to actually show you how to do it.”
03:38
“All we have to do is tell our agent what we want to build.”
03:46
“I'd like to create an email assistant that checks my email every thirty minutes, automatically removes spam, DMs me when something looks urgent, and creates a way for me to send email.”
11:25
“If it was all just one big agent, it's gonna get harder and harder for the agent to be successful — that's why we have scoped agents.”
Word for word.
Read-along
Don't just watch it. Burn it in.
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.
metaphorstory
00:00Everyone's talking about AI agents like OpenClaw, but most of them are a pain to set up and that keeps people from actually using them. And that's why I built this easier system that you can use for free. And in this video, I'll prove it by creating an example assistant that monitors my email, kills all the spam, and can reply to emails on my behalf.
00:15And then I'll show you exactly where to get it and how to install it. By the end of this video, you'll be building agents to run your own life twenty four seven. Let's get into it.
00:21Alright. So here I am inside of the Popebot. I started a new chat.
00:25It feels a lot like Claude or ChatGPT. And the first thing I'm gonna do is just confirm it has access to my Gmail. And in order to give it access, I'm gonna click on this little key here and we're gonna add a new secret.
00:36Manual is for adding normal API keys but for stuff like Gmail, you have to use OAuth. So I'm gonna click OAuth and I'm just gonna type Gmail oauth token and under provider, I'm gonna go ahead and select Gmail.
00:49And now inside the Google Cloud, I need to get a client ID and a client secret. Now in order to do that, you will need to log into the Google Cloud. I've already got an existing project that I can use.
00:58But if any of this is new to you, feel free to use the system. You can start a new chat and just ask it a question. How do I set up a Google Cloud project, Gmail API, and get my OAuth client ID in secret?
01:08If you type that and just run that, it's going to give you precise step by step instructions on how to do that. And so I like to show people this because a lot of times people feel like they need somebody to show them things but you can use these agents to actually show you how to do it.
01:22So now from here, already have an existing project. I've already enabled the APIs for Gmail. If you haven't done that, you would come here and just type Gmail Gmail API and then you would just enable it.
01:35So I'm gonna come up here to APIs and services. We're gonna go to credentials.
01:40If you haven't set up an OAuth screen, you're gonna need to do that first. And again, just rely on the AI agent itself to give you step by step instructions. It's a lot more empowering that way.
01:48And I'm gonna come here. I'm gonna create credentials. We're gonna create OAuth client ID application.
01:53We're gonna use web application, YouTube demo. And now here for authorized redirect URLs, I'm gonna come back to our ad secret and I'm gonna grab this URI right here.
02:04Copy. Paste that in there just like that and we'll go ahead and create it. It's gonna give me the client ID.
02:09I'm just gonna grab that here, drop it right there and then the client secret, I'm gonna copy that right here and then we'll just go ahead and authorize and this is going to give your AI agent access to your Gmail, make sure you pick the right Gmail, allow and now the token is saved.
02:29So now I'm gonna come back to the message here and I'm just gonna run that, confirm you have access to my Gmail. It's gonna use that credential to log in to Gmail and just make sure everything's working.
02:39Alright. So now from here, all we have to do is tell our agent what we want to build. I'd like to create an email assistant that checks my email every thirty minutes, it should automatically remove any spam like sales pitches.
02:59By remove, I just mean remove it from the inbox. If any emails look urgent or important, make sure to send me a DM with the sender and summary.
03:16And then also create a way for me to send an email.
03:23Go ahead and add that to the schedule and also create the required skills.
03:32So once you have that, you can go ahead and hit send and it'll get to work. Now just keep in mind what's happening here is that we are sending messages from here but this is actually interacting with Cloud Code on the back end and it's using their SDK as a mechanism to do the work and it's gonna create the skills and it's gonna make sure everything is set up.
03:53And if you want, can always open these up to see what it's actually doing here. And so we'll just let it continue but what it's gonna do is just gonna investigate what it needs to investigate and it's gonna build the skills, it's gonna add the skills, it's gonna add that to the schedule every thirty minutes and then we should be good to go.
04:07Alright. So it's got a plan, it's building out the skill, it's creating the email assistant with its own system prompt and job prompt. This job prompt will be what is run every 30.
04:16It's creating a Gmail label AI triage so that whenever the agent does something we know who did what and it's adding the cron and it's gonna test the skill end to end before actually committing everything. So it says send work, so I'm assuming that means that if I go to my Gmail, it looks like it sent a test message right here.
04:35So that's good. And again, if you're ever curious, just feel free to investigate exactly what it's doing at any given time.
04:42You know, I'm just looking at the to dos that it created and which ones it's completed. Cool. And then it gives you a response back on what it did.
04:48So it created the new skill. It's able to list the inbox. It can get archive and label emails.
04:53It can send emails and it's using my Gmail OAuth token and it tested everything end to end. It created a specialized agent called email assistant. It created the system prompt.
05:02It created a claw dot m d and it also created the schedule, the cron job that will check the email every thirty minutes. And then down here in the admin area, if you ever wanna check-in on the cron, you can check that right here. So we have email triage is enabled.
05:17And then while we're here, did wanna show you this as well. If you also wanted to access the agent secrets from the admin section, you go to event handler and then agent secrets, you can do that here.
05:26You can also adjust all of your LLM providers whether it's Anthropic or OpenAI or Google or DeepSeek. It supports all of the major platforms and it also supports all of the different coding agents as well.
05:39There's a lot of different settings here. I'll do other videos to explain everything and then make sure you see all those videos. Make sure to like and subscribe now.
05:45But for now, let's go back to the chat and I'm just gonna run a test. Can you check my current inbox and pretend the email is important and justifies you sending me a DM summary of that email.
06:05So I just wanna make sure that it has the ability to check. This is an email from my accountant and I wanna make sure I can read that inbox and seeing I have an email from David and it just sent me an update on exactly what that is via Telegram.
06:20And then it just gave me a note here that it did not apply the AI triage label to this just as a demo. But I'm just gonna tell it to go ahead and do that as well just so I can see. And so if we come back over here, you can see that it added the AI triaged label.
06:32Alright. So now let me show you a few other things that the Popebot has that no other agent that I've seen has and that is the support for multiple users. So I'm logged in here as myself, but if I had other users, I could create users for them as well.
06:44And what's cool about that is that in the event handler here, in this section here where we set up Telegram, the Pope bot here is connected to a specific Telegram bot and it can communicate with any of the users or all of them at the same time.
06:58Okay? So when you come to the event handler, is what we are in right now, this is the main chat interface.
07:04If you come to Telegram, you can create a Telegram bot and then you just put that API key here and then you just set up the webhook by registering it right here. It's very simple to do. You just go through that process.
07:14And then once we have the bot connected, if I come to my profile here and I put in my first name, Steven, and my last name and a nickname, then in the system, if I were to say send Steve a message, like if I were to come back here and just say send Steve a test message, the system is smart enough to look at my nickname and then route that to the proper person and then you can see here it sent the test message.
07:41So if you were to come to your users and then you were to add another user, they would be able to come to their profile as well, add their first name, last name, and their nickname and then the POP bot would be able to send them a message as well. So it's a very powerful system in terms of Telegram. And then in the profile here is where you actually connect up your specific user to the Telegram chat.
08:01That way the POP bot only listens to messages coming from authorized people. Otherwise, the bot would literally be able to take messages from anyone. Now another thing I wanted to mention as well, this entire session here happened through Cloud Code.
08:15Now you might be used to using Cloud Code through the terminal and if you'd like to do it that way you still can. This is the chat mode. This works really good in mobile and it works good just for a nicer interface.
08:26But if you really wanted to go into the Cloud Code CLI where you can interact with it this way, then you just click that button there and it will open up. And if you notice here, the entire dialogue of what we did is still here. So you can continue and pick up right where you left off.
08:40And not only that, but if you wanted additional Claude code sessions, you can just click that on the tab here. If you wanted a shell, you could go right into the shell so that you can see the files and interact with it that way.
08:52And you can even access a file editor if you wanna manually modify things. So this is the interactive mode versus just the headless mode and if you close session, you come right back to here and then you can continue your conversation from here.
09:05And then if you wanna do voice to text, you can just turn that on and if you talk, it's just going to add that directly into the message box. And what's also cool too is that you have these messages over here, which is a duplicate of what's being sent to Telegram. So the messaging system and your chat platforms, it's all integrated into one seamless system.
09:22Now one other thing I wanna show you, I'm gonna come up here to new chat. When you come to create a new chat and you are in agent mode, an agent mode is what we've been discussing here. I've done other videos on code mode.
09:31You can check that out. This is if you're developing software. In this case, I'm developing an agent.
09:35When you're developing an agent, you'll see this drop down here. And essentially what this allows you to do is if I'm in the root position here, then I'm working within the Pope bot itself and all of the agents. Right now, we just added the email assistant.
09:48I had this test agent that I wanted to add before. And the purpose of that is is that if I wanted to scope down, like if if I wanted to have this chat so that it only was working with the email assistant, I could do so. And then that way it's not gonna be as distracted by all of the other agents or tools or skills that I might be creating in the lifespan of my specific install of the Popebot.
10:11Because you might end up creating 20 different agents. Each of those agents might have three or four different skills. And if it was all just one big agent, .
10:21So we have this concept of scoped agents and if I start a chat here, then it's only going to be operating from within this email assistant. It's not gonna have access to all of the other agents and skills and system prompts that might be distracting. And so you can see here it's saying send message to agent's email assistant.
10:38And so if I jump into interactive mode here just so I can show you around a little bit, I'm actually just gonna open the editor here. So we're seeing the full scope here. This is everything related to the PopeBot.
10:47But if I come in here to agents, we have our email assistant and the test agent I made before and then we've got the system prompt and the cloud prompt. So when you use this scoped agent, it scopes the agent with just this specific system prompt so that it's highly targeted towards the email assistant. Now if I were to start the conversation this way, it would work.
11:06It would work fine. It would go and it would read the email assistant and it would do everything that needs to do. All of the prompts and the the configuration are there.
11:14For instance, if I come here to the Cloud MD, notice if I scroll down, it lists the agents that are here. So if I were to say send an email, it would be able to find the agent. It would just take a little bit longer and there'd be a little bit more context for the PopeBot to figure out before it was able to get to work.
11:29Whereas if you use the scoped agents, it's it's gonna be able to figure things out a lot faster and it won't get distracted by all of the other features that you might be adding to your specific agent system as a whole. Alright. So now to install the Popot, all you have to do is go to the link in the description below then and you're just gonna scroll down to the installation instructions.
11:48There are some prerequisites that you need to have installed on your computer. It works on Windows, Mac, or Linux, but you'll need to have these installed. And then there are these two simple steps that you can go through and the wizard's gonna walk you through exactly how to get everything installed.
12:01If you want video instructions on how to do this, I do have a video on my YouTube channel and I'll link that below as well. Once you have it installed, if you do need support, go ahead and click that link and we'll be there. And there is an upgrade system as well, so as I release new upgrades, it will show you those as those become available.
12:15You just click on this and then you'll be able to upgrade to the newest version. And if you're interested in learning more about agents, make sure to jump into my community, the AI architects. We've got a full classroom.
12:23You can post a thread to get support or jump over to the calendar and I've got calls with me every Tuesday and Wednesday. And the networking calls on Friday are a great place to meet other people and just talk about your projects and let people see what you're building. So I'd love to see you inside the community.
12:37Either way, I hope you enjoyed this video and I'll see you on the next one.
The Hook
The bait, then the rug-pull.
Stephen Pope opens with a straight knife-fight: OpenClaw is hard, his free thing is easy, and he's about to prove it by wiring up an email-triage agent in twelve minutes flat. The whole video is the proof. There is no theory section — just OAuth tokens, one paragraph of natural-language spec, and a working scheduled agent by the credits.
Named ideas worth stealing.
00:59concept
Ask-the-tool-how-to-use-the-tool
When the tutorial hits an external dependency the viewer might not know (Google Cloud), Pope tells the viewer to ask the agent itself for step-by-step instructions instead of teaching it.
Steal forany tutorial whose setup leaks into territory you don't want to teach
10:33concept
Scoped agents
Conversation-level scope to one sub-agent so the orchestrator isn't loading every skill in the install. Faster decisions, less drift in 20-agent setups.
Steal forany multi-agent product where context bloat is the real failure mode
08:02concept
One spec, four interfaces
Web chat, Telegram chat, raw Claude Code CLI, and shell all share the same conversation thread and skill registry — same agent state across surfaces.
Steal forany solo-builder positioning their AI product against single-surface SaaS
How they asked for the click.
VERBAL ASK
11:40product
“Go to the link in the description, scroll down to installation, two simple steps. If you need support after install, click the link. And if you want to learn more about agents, jump into my community, the AI Architects.”
Soft, layered CTA — install link first, then community as the second offer. No pricing pressure, no urgency, just 'here's where to go'. Earns it after 12 minutes of giving away a working product.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
