The argument in one line.
AutoCloud multiplies AI coding output by running parallel Claude agents in isolated git branches with automatic merge conflict resolution and self-review loops, allowing developers to queue dozens of tasks and get them completed overnight without manual intervention.
Read if. Skip if.
READ IF YOU ARE…
- A developer with Claude API access who manages 5+ concurrent feature requests or bug fixes and wants to parallelize work across multiple git branches without manual orchestration.
- A solo founder or small team lead running production code who needs AI-assisted development but lacks time to review, test, and merge each Claude suggestion individually.
- An intermediate developer who understands git workflows and wants to see how multi-agent AI systems can handle task decomposition, conflict resolution, and self-review without human intervention.
SKIP IF…
- You don't have a Claude API subscription or Claude Code access — the tool wraps Claude as its core engine and won't work without it.
- You're working in a non-git codebase or a version control system other than git — Auto Claude relies on git worktrees and branch isolation.
- You work on codebases where you need real-time human approval on every AI decision — this tool is designed to minimize interruptions and only surface tasks after the AI marks them done.
The full version, fast.
Auto Claude is a free, open-source Electron and Python orchestrator that wraps a Claude Code subscription in a Kanban board, running parallel agents in isolated git worktrees so you can stack ten or more tasks at once without merge collisions. Each task is auto-classified by complexity, planned with subtasks, coded in a sandboxed branch, then self-reviewed by an AI layer before it surfaces for human staging, with a dedicated merge-conflict resolver patching overlaps as files evolve. A graph plus semantic RAG memory layer makes context cheaper the longer you use it, and integrations cover GitHub issues, changelog and release generation, multi-account rate-limit swapping, a roadmap generator, and up to twelve renamable terminals for hands-on work alongside the autonomous queue.
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 – 00:38
01 · Cold Open + Introduction
10x the work done — promise stated, creator introduced, product named. No warm-up.
00:39 – 01:03
02 · Project Setup
File picker -> .autocloud folder initialized. One-click onboarding.
01:04 – 02:10
03 · Kanban Board - Creating Tasks
Planning -> In Progress -> AI Review -> Human Review -> Done. Creates bug-fix task by pasting a screenshot. Shows model, thinking level, and human review gate controls.
02:11 – 06:08
04 · Task Complexity Classification & AI Review
System auto-classifies task as simple (90% confidence). Introduces worktrees (git sandboxes per task) and the merge conflict AI layer. Live log panel shows tool calls.
06:09 – 07:35
05 · Agent Terminals
Up to 12 simultaneous Claude Code terminals, renameable. Tasks can be created from terminal view. Session restore.
07:36 – 08:36
06 · Insights & Roadmap
Insights = persistent project-aware chat. Roadmap = AI-generated feature priority breakdown. Planned Canny integration.
08:37 – 11:12
07 · Context & Memory System
Project Index auto-parses codebase (Electron + Python detected). Graph memory + semantic RAG accumulates session insights — claims to become cheaper than raw Claude Code over time.
11:13 – 12:48
08 · Changelog & GitHub Integration
Changelog Generator pulls from completed tasks or Git history since a tag. One-click GitHub Release creation with emoji support. v2.2.0 generated in ~30s.
12:49 – 16:03
09 · Advanced Settings & Multiple Claude Accounts
Supports multiple Claude Max accounts with auto-switching on rate limits. GitHub Issues integration incoming.
16:04 – 18:58
10 · Install Walkthrough
Download zip -> open in Cursor -> install Node.js + Python + Docker Desktop -> pnpm install + pnpm run start. Live macOS install demo.
18:59 – 19:28
11 · Conclusion & CTA
Discord community plug, subscribe ask. Clean end.
Lines worth screenshotting.
- Auto Claude is a free, open-source Electron + Python app that wraps Claude Code in a task-planning, worktree-sandboxed, self-reviewing multi-agent orchestration system.
- Tasks are automatically classified by complexity (simple, medium, complex) at 90%+ confidence, which determines how much test coverage and spec documentation the agent generates.
- Each task runs in an isolated git worktree (a sandboxed branch) so multiple tasks can modify the same files in parallel without blocking each other.
- An AI merge conflict resolver handles the programmatic and semantic conflicts that arise when parallel worktree tasks are merged back into the main branch.
- Auto Claude's self-review loop means the agent critiques its own output before flagging a task as ready for human review — reducing the number of half-finished tasks you have to evaluate.
- A human review gate before coding is an optional setting for complex tasks where the developer wants to validate the plan before any code is written.
- The agents terminal supports up to 12 named Claude Code instances visible simultaneously — renaming terminals by task description preserves mental context across long multi-session days.
- Session restoring means closing and reopening Auto Claude returns all terminal sessions to their previous state without losing work in progress.
- A Kanban board (not just a task list) gives a visual overview of every agent's current status — planning, coding, reviewing, or ready for human check.
- The insights panel is a persistent Claude instance with project access for architecture questions, research, and exploration that runs separately from the task agents.
- Simple tasks (bug fixes with a screenshot description) get a quick spec and a single targeted test — the system does not over-engineer low-complexity work.
- Auto Claude is designed for all skill levels: vibe coders use the Kanban UI, senior developers use the worktree and terminal control for granular oversight.
Steal the architecture, not the app.
JoeFlow / Sessions playbook
The worktree-per-task pattern is the unlock — it is what makes running 12 parallel agents safe rather than chaotic.
- Worktree = sandbox per task. Each agent branch is isolated — parallel tasks cannot step on each other. Consider wiring JoeFlow Sessions rows to git worktrees.
- Pre-classify before burning context. Auto Claude's complexity classifier (simple/medium/complex) gates how much token spend each task gets. Build this into JoeFlow batch dispatch.
- The human review gate is the product. Users don't want to babysit agents — they want to approve finished work.
- Auto Claude is free + open source. Win on Windows-first polish, JoeFlow-native integration, and long-term stability.
- "I get a lot of tasks done while I sleep" — this is the positioning sentence. If Joe ships a Sessions-powered batch mode, that line belongs on the landing page.
Terms worth knowing.
- Auto Claude
- A free, open-source desktop application that orchestrates multiple Claude Code agents in parallel, managing tasks through a Kanban board and running each task in an isolated git worktree.
- Claude Code
- Anthropic's command-line AI coding agent that Auto Claude wraps to enable parallel, multi-task autonomous development workflows.
- Kanban board
- A visual task management system that organizes work items into columns — typically To Do, In Progress, and Done — used here as the interface for queueing and tracking AI coding tasks.
- git worktree
- A Git feature that creates an additional working copy of a repository in a separate folder, allowing multiple branches to be checked out and worked on simultaneously without conflicts.
- multi-agent loop
- An automated workflow where multiple AI agents run in parallel, each handling a separate task, with a coordination layer managing dependencies, merges, and human review checkpoints.
- vibe coding
- Using an AI to write or build software through loose natural-language descriptions, with no requirement for the user to have coding knowledge or write code manually.
- merge conflict
- A situation in Git where two branches have made incompatible changes to the same code, requiring resolution before the branches can be combined — Auto Claude attempts to resolve these automatically.
- Electron
- A framework for building cross-platform desktop applications using web technologies (HTML, CSS, JavaScript), used here as the foundation for Auto Claude's GUI.
- task complexity classification
- Auto Claude's internal process of evaluating a task's scope and difficulty to automatically select the appropriate AI model and reasoning depth before starting work.
Things they pointed at.
Lines you could clip.
00:04
“10 times the work done on your projects with the planning and the quality coding and testing that you should demand from your AI coding system.”
05:01
“A work tree is basically a sandbox or environment where the coding is happening in one place and it won't touch any of the other files.”
09:31
“The more you use Autocloud, the smarter it becomes at actually retrieving context at a smaller token usage. So it will become cheaper to actually use Autocloud over cloud code when you use it over time.”
13:44
“I get a lot of tasks done while I sleep.”
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.
analogystory
00:00Ready for AI coding on steroids? Today, we're focusing Autocloud where you can get 10 times the work done on your projects with the planning and the quality coding and testing that you should demand from your AI coding system.
00:14Now AutoCloud is designed to work for all skill levels from vibe coders without any coding experience to senior developers. Hi. My name is Andre Michelson, and I use AI to code production level applications every day.
00:28And today, I'm showcasing Autocloud, a open source and free project for using your cloud code subscription to get more work done.
00:39So when you're opening Autocloud and want to connect it to a existing project or create a new one, you just come to the top left corner here, add project, open existing folder, and when you open it, it's gonna prompt you to initialize Autocloud.
00:55That's gonna create a dot Autocloud folder inside your project. You just initialize and then the project is ready for you to use.
01:04And we're gonna start with showcasing the Kanban board here, which is the system for long running agents that just runs in the background. Now how it works is that you start engaging the system by just creating a task.
01:17And we're gonna start with one simple task where I found a little bug here in actual inside of AutoCode where we have a double crosshair. So we're just gonna take a screenshot of that and just paste it into the description and we see it gets added as a reference image to the task. And we're just gonna say here, and I'll use dictation.
01:39Please see the image in terms of the double cross for closing out the project files that's overlapping with the model for closing the create task model. So you can see now when we have some options here for selecting the model and the thinking level if you want more control.
01:57If you don't want any do any manual things, we have tried to keep things at a good level. We are gonna do a little bit of a work here where the system itself figures out what model and what thinking level it's gonna use.
02:11But as you can see now, we also have the section for requiring a human review before coding. If you want to work on more complex task where you want more control.
02:22But the system will work autonomously, and also we can drag and drop part project files here into the description. But right now, it's not working correctly.
02:32So we're gonna create a second task for showcasing how the system actually identifies how a task is either simple or it's a medium sized task or it's a very complex task.
02:44So you can see now it's written the task title itself. We don't have to do that if we don't want. And we can now just start this task and it will go into a planning phase where it starts to identify all the context it needs.
02:59It finds out what it needs to create as subtasks and also the complexity level of the task itself. So while that has started, we can start to see some logs.
03:09We will also create a new task here for the issue that we can't drag and drop files. When using the project files view, we cannot drag and drop the file into the description.
03:21There's only the section for reference files further down in the task section, but I think we should just mention below the description that the user can both paste screenshots and drag and drop files from the project file explorer.
03:37So now, in many cases, this is a description that's like we could give a lot more context into the task in order for it to be a much better task. And the system we're trying to like, Autocloud is gonna try and identify everything and also have a self review process.
03:54So it's gonna create a plan. We'll go here and see the logs have started. You can see now it's in planning phase and it has 27 entries.
04:04And if you want to review and understand what the AI is actually doing, you can see the logs here with all the tools, all the writings, and everything. And it has identified here, we can see the complexity level for the fixing overlapping close buttons.
04:19It has identified as a simple task with a confidence of 90%, and that dictates how everything is gonna work. It's just gonna go and have a quick spec and create a simple test for fixing or checking that it's actually fixed once it's done.
04:34So it's not gonna produce a lot of unnecessary tests, a lot of different coding, just so it saves upon the context itself here. So we don't have to pay more for a simple task integration.
04:48And if you're wondering, we are working on the same files here and the same systems, but how AutoCloud is working is that if we have a full overview over all the work trees that are for this project itself. So that means a work tree is basically a sandbox or in a environment where the coding is happening in one place and it won't touch any of the other files so you can safely work on the same as for instance, in this case, we're working on the same files on two different tasks.
05:17And then if you're a developer, you're gonna be like, hey, we're gonna have merge conflicts. But inside Autocloud, we have this merge conflict AI layer, which is doing a little bit of programmatic things, but also having an AI that does the refactoring and handling merge conflicts if there's any in these two tasks.
05:36So we're gonna see once it's done, we're gonna see that it's actually a merge conflict and we can just quickly with one button solve it and everything gets done correctly. So as this is working, you can already see that this simple task is already going over to an AI review. So the AI is gonna review its own work, and that's the power of AutoClaw that you can just have this running in the background, and it's only ready for you once the AI has deemed it ready for you to actually test and utilize straight into the project.
06:09So we also have this agents terminal, and this is a place where you can spawn up to 12 terminals and easily invoke Claude in all of the instances.
06:18And there's also some smart functionalities we're working on because myself, I like to have this type of terminal overview that I can rename terminals to what there's they are about, So I can hold my my thoughts when I navigate around, especially when you start to get more terminals, like when you start to get six and nine and even 12.
06:40Depends how much like multitasking you want to do. But for me, when doing like tasks, I want to rename the terminals.
06:48And also there's a smart functionality here in terms of that. You can add tasks just straight from the terminal view and you can also reference them basically like this and it will rename the terminal into the task description, and it will also give the description straight into the Cloud Code terminal.
07:07So you quickly can start new tasks if you have planned out things, but want to not utilize AI system with everything happening in the background, you want to be more hands on. This is the system that allows you to do that. We also have some session restoring so that you can close down the application and spawn it up again, and it will save the session so you can continue if anything happens.
07:28And we'll continue to iterate on this based upon how many people use the agent terminals in their day to day. We also have insights, which is basically just a chat with history, which is a cloud called instance that have access to your project, and you can spawn chat history here. Myself, I like to use this for more, like, just investigating things that I'm wondering about, maybe having a sparring partner, something like that is nice to have.
07:56And also the roadmap functionality is a very good thing for you actually getting good traction with your project. So here you can see we have different views for creating a roadmap and we're also gonna be integrating third party software like Canny in order to get feedback from users from those softwares and just incorporating it straight into the Autocloud system.
08:19So here you can see the AI has tried to identify your target audience, and then it's breaking down features that you should implement into your project. This is just gonna become better and better as the more you use it because Autocloud has its own memory layer.
08:37So we can go here to context, and the first thing you'll see is the project index. So it's basically what we can programmatically figure out what the project is about.
08:46It has identified we are in the auto cloud repo here, and it is in in identify the electron front end and the Python back end with the API routes and the dependencies just so that things like this, the actual project infrastructure is quickly given as context to the user.
09:05And something we're working very heavily on is the memory system. It's a graph memory system also incorporated with semantic rag.
09:15If you don't know what that means, it's basically that we have different methods for letting the AI know about your project. And as you're doing and you're working with tasks here, the more you use it, the smarter the system becomes.
09:28And for people who has a little bit more knowledge about context windows and token usage, the more you use Autocloud, the smarter it becomes at actually retrieving context at a smaller token usage. So it will become cheaper to actually use Autocloud over cloud code when you use it over time.
09:48So the roadmap functionality definitely is something to check out and see what it suggests that you should do for your project. But also we have the ideation tab here where we just can get some quick tasks if you're not sure about what you should do.
10:02And also if you have something like, I just want to see, for instance security and code improvements at a general step, you can generate that and Autocloud will figure out based upon your project, investigate where how do we have some security issues.
10:20We also have performance, so we can do performance enhancement. But also like code improvements is generally is trying to locate a low hanging fruits, quick wins, where we can quickly incorporate something that has good value for the users of the project.
10:37So let's go just go back here and see. You can see now that the two tasks that we have done is done with AI review. We can see here it hasn't identified actually that there is any merge conflicts ready for this.
10:50Let's just recheck it. In these tasks, you can go in and see the sub tasks, what it has done. You can also check the full logs in terms of what it did for validation, what it did for coding, so you have full overview over what the AI system has actually done for each of these.
11:08We're just gonna let this run because it finds five different things with each category. And while we do that, we'll talk about the change log functionality inside AutoCloud, and it's something I really like. It has the possibility to create change logs and also it's integrated with GitHub, so you can create a change log based upon either all the tasks you have done inside AutoCloud.
11:30But what we use most is to get history because I like to work with both the task Kanban board, but also like I showcased more manually. So I'm manually doing commits to GitHub, and I'm manually doing work with AI.
11:45That makes it so that when you go into Git history here, you can for instance say since last version, say, and we are version 2.1 right now, and it would actually load all the commits that's happening since that version, and you can just continue. And we will say two point o version, and this is a GitHub release, and we want some emojis.
12:08And it's done just like that, it's gonna generate a change log, And it's quite quick. We don't use opens for this functionality right now, and you'll see it identifies the new features, improvements, bug fixes, and also all the changes.
12:23So we can quickly just save this to the change log. And once it does that, we can just click create release, and that goes straight up until GitHub. So that's a very fast way to get good change logs and do GitHub releases.
12:37If you want something like that for your project, I do recommend using the change log for actually keeping track of what's happening inside your project just so you have control over time.
12:50So we talked about the context as well. It's something we really are working hard on and there's integrations with GitHub issues. So GitHub issues will come straight into Autocloud, and you can create tasks, and it can start to identify what's wrong and fix it automatically in the background.
13:08There's a lot of other things that's coming. There's a lot of good things in the settings. For instance, if you are a very heavy power user like me, you can utilize the integrations tab here.
13:21We have multiple cloud accounts. Because what AutoCloud actually allows you to do is to create a lot of tasks and just run them in the background, so you get a lot of features done quickly.
13:34But that also implies that you're getting things done not only quicker, but for instance, I get a lot of tasks done while I sleep. So basically, all my Cloud Code usage has doubled, so I need two Cloud Code Max plans.
13:49And we have created a system that you can incorporate both of them into the AutoCloud system and easily swap between them, and we will incorporate some type of like auto switching, so it knows about what the rate limits for one account and it swaps when you need to swap, basically. So now we can go ahead and just get the task into our project.
14:10We open the task which is ready now for human review, and we just press stage changes. And that makes it so that if there's any merge conflicts, it's gonna solve that by itself.
14:23And we can go here into the IDE now and we see that the changes have come into our repo and we can repute review them if we want And we now also can just go into the you can see here the changes staged successfully. So if there is any merge conflicts, and this is a very good, which I appreciate, which is the merge conflict AI layer that fixes merge conflicts automatically.
14:49So you can work on a lot of tasks and also work manually in other branches, and the system will follow along as files evolve. But we can now go and see if this is successful.
15:00This was the fix overlapping buttons and that was that if we go here and we see it browse the files, you can see now the task was successful. It implemented it and we can go ahead and merge this with our main branch.
15:15So what's coming next? We have a great focus on the framework itself that runs behind the hood. We also have the Python CLI that you can utilize if you don't want to use the front end electron application.
15:26Now we're also working on incorporating the bMAD method for context engineering, AI coding, validation, and we're just gonna put it on steroids by putting it inside AutoCloud where we get, like, the long running agents, QA looping, rag memory system, and much more.
15:44We also welcome everybody to our free Discord community. The link is in the description to talk about features, bugs, and you can also network with like minded AI enthusiasts, developers, entrepreneurs, marketers, and also a lot of people there, and then generally a nice community to feel at home.
16:01So stay tuned for big productivity update. So I just wanted to showcase how you can install AutoCloud if you're wondering about how to do that.
16:10We have a very descriptive quick install description if you want to follow along that. If not, here's how you do it quickly. You download the zip file here on GitHub and the link is down in the description.
16:24Then you go and you create either a new folder or you unzip the autocloud folder which is which you downloaded.
16:33And you just go to your favorite IDE. Here, we're using cursor just because it's simple to use with AI especially if you're a beginner and we just open up the folder that we just unzipped. Now we have the Autocloud project open here.
16:53And the first thing you're gonna do is that if you is that if you haven't installed the NodeJS Python and Docker desktop, just follow the instructions or just what you also can do is basically just copy and paste this into something like cursor and just say help me install this and it will help you do that based upon your system and that's actually a very quick way for you to get started.
17:24Once everything is installed, you're just gonna do the first step here which is to do these commands inside the terminal here.
17:35So if you don't have UV, you can basically just tell Opus to or or the agent you have inside cursor or any other a AI agent to just help me use UV, which is basically just for managing the packages here in a environment. Or you can use this you can try this also if you're if the first command here is failing.
18:01That will basically spin up a virtual environment and install everything, so now it's ready to be used.
18:10The next step is to actually start the memory layer, and here we need Docker Compose in order to do that. So you will just go and do this command in its own database. It will create the database and spin up what's necessary for the memory to work.
18:27And the last thing is to actually just install and launch the application. So we're gonna do p n p m install that installs everything, And then just command copy this command once it's done here.
18:39So now it will just build everything and it will start the application as soon as it's done. And it's taking quite a short amount of time. And now we can see we're getting a new application up and going here.
18:54So there you have it. It's up and going.
18:59And if you have any questions, I highly suggest you join our Discord community. I will be very interactive there in terms of helping you if you have any issues or bugs. And if you have liked the video, be sure to subscribe and like it and help other people find this amazing system that we are working on so we can get going with our projects at lightning speed while keep keep still keeping that quality that we want from our AI coding.
The Hook
The bait, then the rug-pull.
The title and the first spoken line are the same sentence — a clean pattern-interrupt that wastes zero frames before stating the promise. André Mikalsen opens with a question, answers it with a product name, and is inside the demo by the 40-second mark.
Named ideas worth stealing.
05:01concept
Worktree-per-task sandboxing
Each task runs in its own git worktree (isolated branch). Parallel tasks cannot clobber each other. Merge conflict AI layer resolves diffs when tasks complete.
Steal forJoeFlow Sessions — each agent row could run in its own worktree to prevent collisions
04:08model
Task complexity classifier
- Simple
- Medium
- Complex
Auto Claude classifies each task before coding begins. Simple tasks get a quick spec + one test. Complex tasks get full spec + multiple subtasks + deeper review. Controls token spend automatically.
Steal forAny Claude Code orchestrator — pre-classify before burning context
01:04model
Planning to Done pipeline
- Planning
- In Progress
- AI Review
- Human Review
- Done
The Kanban columns represent real agent states. Tasks only surface for human review after the AI has reviewed its own work. Human time is reserved for final acceptance, not QA.
Steal forJoeFlow Sessions panel architecture — the Human Review gate is the key product differentiator
09:24concept
Graph memory + semantic RAG cost curve
As Auto Claude accumulates session memory, it retrieves relevant context with fewer tokens, making it cheaper per task than raw Claude Code over time. Compounding efficiency.
Steal forJoeFlow product positioning — sessions that get smarter over time
How they asked for the click.
VERBAL ASK
18:59next-video
“Join our Discord community. If you have liked the video, be sure to subscribe and like it.”
Soft and brief — Discord first, then subscribe. No product pitch, no upsell. Matches the free/open-source positioning.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
