Bran — Plataforma de Orquestração Multi-Agent
ioancw/bran
Orquestra agentes especializados com chat, API, web e scheduling
Ver no GitHubPlataforma de fleet-orchestration construída sobre o Claude Agent SDK que permite coordenar múltiplos agentes especializados com seus próprios prompts, ferramentas e permissões. Oferece interfaces REPL interativa, CLI, API Python, servidor HTTP autenticado e agendamento tipo cron — ideal para automatizar workflows complexos e delegação inteligente de tarefas.
Instalação
npx convez-ai ioancw/bran1 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
- Automação de workflows multi-etapa com delegação inteligente entre agentes
- Pesquisa web recorrente e sumarização com agendamento cron
- Construção de servidores de agentes com autenticação para integração em sistemas existentes
- Prototipagem rápida de sistemas multi-agent em Python
Veredito

Bran é uma solução madura e bem-arquitetada para orquestração de agentes Claude em produção. A combinação de múltiplas interfaces (REPL, CLI, API, scheduler) e persistência nativa via SQLite a torna pronta para pipelines corporativos — recomendada para equipes que já dominam Python e o Agent SDK.
README
A tradução para o português deste README ainda está na fila.
bran
A fleet-orchestration agent platform built on the Claude Agent SDK — think Claude Code, but general-purpose. One orchestrator agent you can chat with, a roster of specialised sub-agents (each with their own system prompt, tools, skills, and permissions), and several surfaces to drive them: an interactive REPL, a one-shot CLI, a Python library API, a web UI, and an authenticated HTTP server with cron-style scheduling.
What's in the box
| Surface | How you use it |
|---|---|
bran chat | Drop into an interactive conversation with the orchestrator. It can delegate to sub-agents and use skills. |
bran run <agent> --task "..." | Run any agent one-shot. Returns when done, prints structured output. Cron-friendly. |
from bran import run_agent | Fire an agent run from any Python script. |
bran serve | FastAPI server (localhost + bearer-token auth) exposing /agents/{name}/run, /runs, /schedules. |
bran schedule add ... | APScheduler cron triggers that fire agent runs on a recurring schedule. |
All runs are persisted to SQLite (under ./.bran/ by default), with Agent SDK session IDs captured so you can resume conversations.
Quick start
# 1. Install (creates a venv and installs the package in editable mode)
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -e .
# 2. Configure
copy .env.example .env
# then edit .env and set ANTHROPIC_API_KEY + BRAN_API_TOKEN
# 3. Talk to the orchestrator
bran chat
# 4. Run a single agent one-shot
bran run research --task "Summarise the latest news on Mars helicopter flights."
# 5. Start the HTTP server
bran serve
# then in another shell:
curl -H "Authorization: Bearer $env:BRAN_API_TOKEN" `
-H "Content-Type: application/json" `
-d '{"task": "What is the capital of France?"}' `
http://127.0.0.1:8765/agents/research/run
# 6. Schedule a recurring run
bran schedule add daily-news research "Brief me on today's top AI news" --cron "0 8 * * *"
bran serve # the scheduler runs inside the server process
Agents
Agents are defined in src/bran/agents.py as Agent objects, plus an optional filesystem definition in .claude/agents/<name>.md. Out of the box:
| Name | Role | Tools |
|---|---|---|
orchestrator | Default conversational agent. Delegates to others and can spawn background runs. | Read, Glob, Grep, WebSearch, WebFetch, Agent, spawn_agent |
research | Web research + summarisation. | WebSearch, WebFetch, Read, Write |
Add a new agent by editing agents.py or dropping a markdown file in .claude/agents/.
Skills
Skills live in .claude/skills/<name>/SKILL.md and are auto-loaded by the SDK. An example web-research skill ships with the project.
Slash commands
.claude/commands/*.md files become slash commands in the REPL. An example /digest command is included.
Layout
bran/
├── .claude/
│ ├── agents/ # filesystem-defined subagents
│ ├── skills/ # auto-loaded skills
│ └── commands/ # slash commands
├── src/bran/
│ ├── config.py # env + paths
│ ├── agents.py # Agent registry (translates to ClaudeAgentOptions)
│ ├── persistence.py # SQLite store for runs + schedules
│ ├── runner.py # core run loop wrapping the SDK
│ ├── repl.py # interactive chat (ClaudeSDKClient)
│ ├── tools/ # in-process MCP tools
│ ├── cli.py # Typer CLI
│ ├── api.py # FastAPI server
│ └── scheduler.py # APScheduler integration
└── scripts/
└── example_programmatic.py
License
MIT.