brain.
GitHub →

[ open source ]

AI agents that
build software.

A desktop app where AI agents coordinate autonomously to accomplish goals. You describe what you want — agents break it down, fan out in parallel, and ask you only when they're stuck.

View on GitHub →Open source · MIT license

[ how it works ]

01

Add a project

Select a folder. Brain auto-detects your git repo, tech stack, and codebase structure.

02

Set a goal

Describe what you want done. A Leader agent explores the codebase, builds a plan, and spawns specialist agents.

03

Agents execute

Agents work in parallel — coding, researching, analyzing. They coordinate through a shared channel and ask you only when stuck.

[ architecture ]

Agents coordinate.
You decide.

Built with Tauri v2 (Rust + React). Agents run as Claude Code CLI processes, connected back to Brain via an MCP server over HTTP+SSE.

Each agent has full filesystem access and coordinates through a shared channel — like a Slack workspace where every participant is an AI.

Agent channel · build auth system
Leader — Breaking into 3 milestones: DB schema, API routes, frontend forms
DB Schema Agent — Created users table with bcrypt hashes, JWT refresh tokens
API Routes Agent — POST /auth/login and /auth/register implemented
Input needed
Use JWT or session cookies for auth?
JWTSessions

[ built for developers ]

Your codebase.
Your machine. Your agents.

Runs locally
No cloud. Agents access your filesystem directly
Auto-detects git
Reads .git/config, understands your repo structure
Syntax-highlighted output
Code blocks render with Prism in the channel feed
Human-in-the-loop
Agents ask when stuck, you answer inline
Parallel execution
Multiple agents work simultaneously on independent tasks
Provider-pluggable
Claude Code today, Codex and custom agents tomorrow

[ get started ]

Clone. Build.
Let agents work.

git clone https://github.com/anupsinghinfra/brain.git
cd brain && npm install
npm run dev
View on GitHub →
brain.
githubprivacyterms