The open-source coding agent
An AI pair programmer that plans, writes and fixes code across your whole repo. Use the best open models, locally or in the cloud.
open models
Frontier open models, your choice
Pick the model that fits the task and your hardware, such as Qwen3-Coder for heavy lifting or a small model for speed. No lock-in, no surprise bills.
Every command, file read and edit is shown as it happens, so you always know what the agent did.
codebase aware
Deep codebase understanding, multi-file edits
The agent maps your project structure, dependencies and conventions before it writes a line, then makes coordinated changes across files.
You review a clean diff and accept or reject it. Nothing lands without you.
terminal native
Works from your terminal
Run it in any shell, in any project, with any stack. It runs your tests and type checks to verify its own work.
Desktop, web and mobile clients will follow.
Built on open models
platforms
Available everywhere, soon
Desktop
macOS, Windows and Linux
Terminal
In any shell, on any machine
$ npm i -g coderbabaWeb
Runs in your browser
Mobile
Check in on tasks from your phone
roadmap
Where we are headed
- now
Model directory & research
Tracking and testing every open coding model to find the best ones for agentic work.
- next
Private alpha
Terminal agent with local model support, tested with early users.
- later
Public beta
Desktop, web and mobile clients, plus integrations with your tools.
Frequently asked questions
When will CoderBaba Agent launch?
We are building it now. Progress updates and the launch announcement will be posted on the CoderBaba blog and RSS feed.
Which models will CoderBaba Agent support?
Open-source coding models such as Qwen3-Coder, Devstral and DeepSeek-Coder-V2, running locally through tools like Ollama or on any OpenAI-compatible endpoint.
Will it run fully locally?
Yes, that is a core goal. With a local model, your code never leaves your machine.
How is it different from other AI coding agents?
It is open-source and model-agnostic. You choose the model, see every step the agent takes and review each diff before anything is applied.