CoderBaba is an AI coding agent for your terminal. Describe what you want, and it plans the change, edits files across your repo, runs your tests and hands you a diff to approve.
✓ Done. Dark mode follows the system theme by default.
$
free to use
$0free to use
open source
100%open source
open-source model
Anyopen-source model
or our free cloud
Localor our free cloud
plan · edit · verify
From a sentence to working code
Tell CoderBaba what to build or fix. It reads the relevant parts of your codebase, plans the change, edits as many files as it needs and runs commands in your project.
Then it checks its own work with your tests and type checker, and keeps iterating until they pass.
coderbaba agentQwen3 Coder
Fix the expired-session bug in auth
Users get logged out after refreshing the page.
◷ Worked for 1m 04s
runnpm test -- auth✗ 3 failed
readsrc/auth/session.ts
editsrc/auth/session.ts+4−2
runnpm test -- auth✓ 24 passed
The refresh token was read before the cookie was parsed. Reordering the middleware fixes it, and all auth tests pass.
parallel tasks
Run several tasks at once
Every task gets its own git worktree on its own branch. Start a feature, a bug fix and a refactor side by side, and none of them can overwrite the others' files.
When a task is done, its branch is ready to review and merge like any other.
✓ 2 tasks running in parallel, main stays untouched
$
you stay in control
Every change is a diff you approve
CoderBaba shows every command it runs and every file it touches as it works.
Changes arrive as a clean diff. Accept them, reject them or ask for another pass. Nothing lands without you.
src/lib/users.ts+5−2
1 export async function getUser(id: string) {2- const res = await fetch(`/api/users/${id}`)3- return res.json()4+ const cached = cache.get(id)5+ if (cached) return cached6+ const user = await db.user.findUnique({ where: { id } })7+ cache.set(id, user, { ttl: 60 })8+ return user9 }
3 files changedRejectAccept all
Conductor cloud shuts down November 20, 2026
Coming from Conductor? Bring your parallel agents to CoderBaba.
Meta has acquired Conductor. Its core app keeps working until early next year, but the cloud service closes in a matter of days. CoderBaba gives you parallel coding agents from the command line: free, open source and yours to keep.
CoderBaba is a free, open-source AI coding agent that runs in your terminal. You describe what you want in plain English, such as "add rate limiting to the webhook endpoint" or "migrate this component to the new API," and CoderBaba does the work. It reads your codebase, plans the change, edits files, runs commands and verifies the result with your own tests.
It isn't an autocomplete plugin or a chat window that pastes snippets. It's a complete tool for building software from the command line, whether that's a new feature, a bug fix, a refactor, a script or an entire project from an empty folder.
Understand. CoderBaba maps the parts of your codebase that matter for the task: structure, dependencies, conventions and the files involved.
Plan. It outlines the change before writing code, so multi-file edits stay consistent.
Act. It edits files, creates new ones and runs commands in your project, showing each step as it happens.
Verify. It runs your tests, type checker and linter, then fixes what fails and tries again.
Hand off. You get a diff and a summary. Approve it, reject it or ask for changes.
Because every task runs in its own git worktree, you can start several at once. Each one has its own branch and files, so parallel work never collides.
Free to use. CoderBaba is free, whether you run models on your own machine or in our cloud.
Any open-source model. Pick the model that fits the task. Use a large agentic model such as Qwen3-Coder for complex work, a fast model for quick edits, or Devstral for a capable agent that fits on one GPU.
Local when you want it. Run a model on your own hardware and your code never leaves your machine, which matters for teams under NDAs, in regulated industries or with sensitive intellectual property.
Open source, permanently. You can read exactly how CoderBaba builds prompts, which files it reads and what it sends to a model. No company can take it away, change the terms or shut it down on you.
Quality has caught up. Open models such as Qwen2.5-Coder 32B matched GPT-4o on several code benchmarks at release. Agentic models like Qwen3-Coder and Devstral were trained specifically to fix real GitHub issues end to end.
CoderBaba runs any open-source model, and our model guide covers the major ones in depth. Four questions will narrow it down. Our full guide to choosing a model goes deeper on each.
What's the task? Multi-step agent work benefits from models trained for tool use and long tasks. Quick, focused edits can use a smaller, faster model.
Local or cloud? In our cloud, hardware isn't your problem. Locally, a rough rule is that a model at 4-bit quantization needs a bit over half a gigabyte of memory per billion parameters, plus room for context. A 7B model fits in 8 GB, and a 32B model wants about 24 GB.
How much context do you need? Repository-level work benefits from models with 128K tokens or more.
Will you ship the model's output commercially? Licenses vary. Every model page shows a clear commercial-use badge.
If you used Conductor, CoderBaba's one-task-per-worktree model will feel familiar. Our Conductor alternative guide explains how to back up your cloud workspaces before the shutdown and move your workflow over.
CoderBaba is a free, open-source AI coding agent for your terminal that runs any open-source model, locally or in our free cloud. Here is what it does.
Parameters, MoE, context windows and licenses all matter when you choose an open coding model. A practical checklist for developers and teams.
open source
Built in the open. Come build with us.
All of CoderBaba's code is on GitHub under the MIT license. Report bugs, suggest features or send a pull request, and watch tutorials and build videos on YouTube.