Agent Brain
Lukearcnet/agent-brain-oss
Gerenciador de projetos para agentes IA com dashboard móvel e checkpoints
Ver no GitHubPlataforma de gerenciamento de agentes IA com memória persistente, comunicação entre sessões e controle móvel. Funciona como um gerenciador de projeto inteligente para Claude Code, permitindo supervisão e aprovação de mudanças direto do celular com notificações push.
Instalação
npx convez-ai Lukearcnet/agent-brain-oss1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Supervisionar sessões de codificação automática do Claude Code remotamente
- Coordenar múltiplos agentes IA em um mesmo projeto com histórico compartilhado
- Aprovar mudanças críticas de código através do celular com checkpoints
Veredito

Solução robusta para quem quer escalar agentes IA de coding mantendo controle humano. Arquitetura bem pensada com persistência de estado e mobile-first, mas exige setup com Supabase e hospedagem local/remota.
README
A tradução para o português deste README ainda está na fila.
Agent Brain
Supervised agentic coding — Claude Code is your IDE, Agent Brain is your project manager.
Agent Brain provides persistent memory, cross-session communication, and mobile control for AI coding agents (Claude Code and Codex). Run agents on your projects while staying in the loop from your phone.
Features
- Persistent Memory — Sessions remember context across runs
- Cross-Session Communication — Agents can message each other
- Mobile Dashboard — Monitor and control sessions from your phone
- Checkpoints — Agents ask for approval before making big changes
- Push Notifications — Get notified when agents need your input
- Session Handoffs — Clean context transfer between sessions
- Email & Calendar — AI-powered email triage and calendar sync
- Codex Support — Works with both Claude Code and Codex sessions
- Customizable — Override views, instructions, and settings without touching core code
Quick Start
One-Liner Install
curl -fsSL https://raw.githubusercontent.com/Lukearcnet/agent-brain-oss/main/install.sh | bash
This detects Node.js, clones the repo, installs dependencies, and walks you through setup.
Manual Install
git clone https://github.com/Lukearcnet/agent-brain-oss.git agent-brain
cd agent-brain
npm run setup # Interactive setup wizard
npm start # Database migrations run automatically
See SETUP.md for detailed instructions.
Requirements
How It Works
- Install Agent Brain on your local machine (one command)
- Run the setup wizard — configures Supabase, env vars, and background service
- Open the dashboard on your phone (same WiFi or via Tailscale)
- Run AI coding sessions — they automatically connect
- Approve checkpoints from your phone when agents need decisions
Architecture
┌─────────────────┐ ┌─────────────────┐
│ Claude Code │────▶│ Agent Brain │
│ Session 1 │ │ (server) │
└─────────────────┘ │ │
│ ┌───────────┐ │
┌─────────────────┐ │ │ Memory │ │
│ Codex │────▶│ ├───────────┤ │
│ Session 2 │ │ │ Mailbox │ │
└─────────────────┘ │ ├───────────┤ │
│ │Checkpoints│ │
┌─────────────────┐ │ └───────────┘ │
│ Your Phone │────▶│ │
│ (Dashboard) │ └────────┬────────┘
└─────────────────┘ │
▼
┌─────────────────┐
│ Supabase │
│ (database) │
└─────────────────┘
Dashboard
Access at http://localhost:3030 (or from your phone on the same WiFi):
- Sessions — Active Claude Code and Codex sessions with live chat
- Memory — Project context and daily logs
- Mailbox — Cross-session messages
- Checkpoints — Pending approvals from your phone
- Email Triage — AI-classified email inbox
- Calendar — Synced calendar events
Updating
bin/ab-update
# Then restart the server — migrations apply automatically
Customization
Agent Brain is designed to be customized without conflicting with updates:
- Views: Copy any file from
views/toviews/custom/and modify it. Custom views take priority. - Instructions: Create
instructions/local.mdwith your preferences, then runnpm run generate. - Settings: Configure via the Settings UI at
/settings.
Remote Access
Same WiFi (easiest)
Your phone can reach Agent Brain at http://<your-mac-ip>:3030.
Tailscale (secure, anywhere)
brew install tailscale
tailscale up
# Access from anywhere: http://100.x.x.x:3030
Cloudflare Tunnel (shareable, with auth)
See SETUP.md for Cloudflare Tunnel setup with zero-trust authentication.
Docker
cp .env.example .env
# Edit .env with your credentials
docker compose up -d
License
AGPL-3.0 — See LICENSE for details.
Contributing
Contributions welcome! Please open an issue first to discuss changes.