Captain Nemo - Agente de Codificação via Lark
verneagent/nemo
Agente de IA para coding direto no Lark com Claude ou Codex
Ver no GitHubDaemon agente de codificação conectado ao Lark/Feishu, alimentado pelo Claude Agent SDK ou OpenAI Codex. Executa tarefas de código em conversas de grupo, com suporte a múltiplos modelos, níveis de raciocínio configuráveis e gerenciamento de sessão.
Instalação
npx convez-ai verneagent/nemo1 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
- Executar tarefas de codificação em grupo via Lark/Feishu
- Automatizar geração e revisão de código com raciocínio configurável
- Integrar múltiplos provedores de IA (Claude e OpenAI Codex) em fluxos colaborativos
Veredito

Solução técnica robusta para integração de agentes de IA em plataformas de comunicação corporativa (Lark/Feishu). Bem estruturada, com suporte a múltiplos backends e controle fino de comportamento, mas requer configuração inicial de app customizado e dependências externas (Node para Codex).
README
A tradução para o português deste README ainda está na fila.
Captain Nemo
Lark-connected coding agent daemon. Runs either the Claude Agent SDK (default) or the OpenAI Codex SDK as the underlying coding agent.
Install
pipx install captain-nemo
Or with pip:
pip install captain-nemo
Upgrade
pipx upgrade captain-nemo
Setup
Create ~/.nemo/config.json:
{
"app_id": "cli_xxx",
"app_secret": "xxx",
"email": "your@email.com"
}
You need a Lark/Feishu custom app with IM permissions and WebSocket enabled (event subscription mode: long connection).
Usage
# Auto-discover chat from workspace tag
nemo
# Specify project directory
nemo --project-dir /path/to/project
# Specify chat directly
nemo --chat-id oc_xxx
# Sidecar mode (only respond to @mentions)
nemo --sidecar --chat-id oc_xxx
# Use a different model
nemo --model claude-sonnet-4-6
# Run with Codex instead of Claude
nemo --provider codex
nemo --provider codex --model gpt-5-codex
# Start with a reasoning effort preset
nemo --effort high
nemo --provider codex --effort medium
# Debug logging
nemo -v
Providers
| Provider | Default model | Runtime | Extra requirements |
|---|---|---|---|
claude (default) | claude-opus-4-7 | In-process Claude Agent SDK | ANTHROPIC_API_KEY or logged-in Claude credentials |
codex | gpt-5-codex | Node sidecar (codex_sidecar/run_turn.mjs) around @openai/codex-sdk | node, the codex CLI on PATH, and OPENAI_API_KEY (or CODEX_API_KEY). Install sidecar deps with npm --prefix codex_sidecar install. |
Switch at startup with --provider, or at runtime with /model <name> — nemo auto-rejects models that don't match the current provider.
Reasoning effort
--effort low|medium|high (or /effort at runtime) controls how much the underlying agent thinks before responding. The shared low/medium/high levels map per provider:
| nemo | Claude (keyword in prompt) | Codex (modelReasoningEffort) |
|---|---|---|
low | think | low |
medium | think hard | medium |
high | ultrathink | high |
off / unset | no keyword | SDK default |
/effort off clears the setting. Effort changes apply to the next turn — no SDK reconnect, no session loss.
On first run with --chat-id, nemo writes a workspace:{machine}-{folder} tag to the group description. Future runs auto-discover the chat without --chat-id.
Commands
Send these in the Lark group:
| Command | Description |
|---|---|
/model [name] | Show or switch model |
/effort [low|medium|high|off] | Show or set reasoning effort |
/clear | Reset conversation |
/cd <dir> | Change working directory |
/esc | Cancel current operation |
/ping | Status check |
/cost | Session API cost |
/norm add/remove/list | Manage group norms |
/guest add/remove/list | Manage guests |
/diag | Run diagnostics |
/help | Show help |
/exit | Stop agent, keep group |
/dissolve | Stop agent, dissolve group |
autoapprove on/off | Toggle auto-approve |
Group Config
Nemo stores persistent group config (guests, autoapprove, norms) in a pinned card titled __nemo_config__.
License
MIT