gois-lite — Monitor e Agente de Recuperação QClaw
naubergois/gois-lite
Monitor de saúde com auto-recuperação via LLM para stacks QClaw/OpenClaw
Ver no GitHubMCP server que monitora saúde de processos QClaw, OpenClaw e Hermes com detecção de falhas e agente LLM para auto-recuperação. Expõe dashboard HTTP, métricas Prometheus e integração com 300+ skills de orquestração de IA. Edição leve focada em desenvolvimento.
Instalação
npx convez-ai naubergois/gois-lite1 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
- Monitorar continuidade de serviços QClaw/OpenClaw em desenvolvimento
- Auto-diagnóstico e recuperação de falhas usando LLM
- Exposição de métricas Prometheus e dashboard web para status
- Integração com Cursor IDE via MCP servers para skills QClaw
Veredito

Ferramenta robusta para DevOps de stacks QClaw, com recuperação inteligente por LLM. Requer setup complexo (DeepSeek, MongoDB, Node.js 22+) e é melhor para equipes em produção ou desenvolvimento intenso.
README
A tradução para o português deste README ainda está na fila.
gois-lite — QClaw Monitor (Lite)
Generative Orchestration Artificial Intelligence Swarm — lite edition
Health monitor, HTTP dashboard, and AI orchestration for QClaw, OpenClaw, and Hermes — with development-focused skills only (default port 9102). Full edition: gois.
| Language | Document |
|---|---|
| 🇺🇸 English | You are here |
| 🇧🇷 Português | README.pt.md |
| 🇨🇳 简体中文 | README.zh-CN.md |
Overview
gois watches QClaw (and related services) on macOS and Windows. It periodically checks process health and HTTP endpoints; after repeated failures, an LLM recovery agent diagnoses logs and attempts self-healing. It exposes an HTTP dashboard, Prometheus metrics, and deep integration with Hermes, OpenClaw, RuFlo, Kanban, WhatsApp, and 300+ QClaw skills.
flowchart TB
subgraph Clients["Clients & Channels"]
UI["Browser Dashboard :9101"]
WA["WhatsApp (wacli)"]
Cursor["Cursor IDE + MCP"]
end
subgraph Gois["gois Monitor"]
MON["Health Monitor"]
AGT["LLM Recovery Agent"]
DASH["HTTP Dashboard"]
MCP["MCP Servers"]
end
subgraph Stack["Vendor Stack (.stack/)"]
OC["OpenClaw"]
HM["Hermes Agent"]
DB[("MongoDB / Redis / SQLite")]
end
subgraph External["External"]
LLM["DeepSeek / OpenAI-compat"]
PROM["Prometheus"]
end
UI --> DASH
WA --> DASH
Cursor --> MCP
MON -->|"pgrep + HTTP + hang"| OC
MON -->|"failure threshold"| AGT
AGT --> LLM
AGT --> OC
DASH --> OC
DASH --> HM
DASH --> DB
MCP --> DASH
MON --> PROM
What it does
| Component | Role |
|---|---|
| Monitor | pgrep, HTTP health probes, optional UI hang detection, log scanning |
| Recovery agent | OpenAI-compatible LLM (default DeepSeek) with tools to read logs, restart QClaw, etc. |
| Dashboard | UI at http://127.0.0.1:9101/ — status, chat, Hermes roles, Kanban, skills |
| MCP servers | qclaw-skills, qclaw-cards, qclaw-jobs, qclaw-swarm for Cursor / IDE |
| Metrics | Prometheus endpoint (configurable in config.yaml) |
| Skills | 300+ QClaw skills — slides, AWS, WhatsApp, research, content, and more |
Health-check flow
sequenceDiagram
participant M as Monitor
participant Q as QClaw / Gateway
participant A as LLM Agent
participant L as Logs
loop Every interval_seconds
M->>Q: pgrep + HTTP GET health_url
alt Healthy
M->>M: Reset failure counter
else Unhealthy
M->>M: Increment failures
alt failures >= threshold
M->>L: Tail log_paths
M->>A: Diagnose + recover
A->>L: Read evidence
A->>Q: Restart (if needed)
end
end
end
Requirements
| Item | Version | Required |
|---|---|---|
| Python | 3.11+ | Yes |
| Git | 2.x | Yes |
| DeepSeek API key | — | Yes (LLM agent + chat) |
| Node.js | 22+ | Yes (OpenClaw vendor) |
| MongoDB | 7.x | Recommended |
| Redis | 7.x | Recommended |
| macOS / Windows | — | Native install |
| Docker + Compose | v2 | Optional (Linux / server) |
Quick start
The repo is self-contained: Hermes (vendor/hermes-agent) and OpenClaw (vendor/openclaw) ship as Git submodules. ./scripts/start.sh bootstraps everything on first run.
git clone --recurse-submodules <repo-url> gois
cd gois
./scripts/start.sh
start.sh is idempotent. It:
- Creates
./.venvand installsgois(pip install -e .[dev]) - Runs
./scripts/unify-stack.sh(submodules,.stack/, vendor deps) - Ensures
./config.yamland./.env - Starts the monitor
Dashboard: http://127.0.0.1:9101/
Fast restart (skip vendor reinstall):
./scripts/start.sh --skip-vendor
Full install guide: docs/INSTALACAO.md
Docker
cp .env.example .env # set DEEPSEEK_API_KEY
docker compose up -d --build
DeepSeek API key
Never commit .env. Copy from example:
DEEPSEEK_API_KEY=sk-your-key
DEEPSEEK_API_BASE=https://api.deepseek.com
DEEPSEEK_MODEL=deepseek-chat
Or sync from a sibling project:
.venv/bin/python scripts/sync_deepseek_env.py
Project layout
flowchart LR
subgraph Repo["Repository"]
SRC["src/gois/"]
SKL["skills/ (300+)"]
VND["vendor/"]
SCR["scripts/"]
end
subgraph Runtime[".stack/ (gitignored)"]
SH["hermes/"]
SO["openclaw/state/"]
SB["openclaw/bundle/"]
ACC["accounts/"]
CH["chat/"]
end
VND --> HM["hermes-agent"]
VND --> OC["openclaw"]
SRC --> Runtime
SCR --> Runtime
gois/
├── src/gois/ ← Python monitor + dashboard + MCP
├── skills/ ← QClaw skills (chat tools, batch slides, AWS, …)
├── vendor/
│ ├── hermes-agent/ ← Hermes (git submodule)
│ └── openclaw/ ← OpenClaw (git submodule)
├── .stack/ ← Runtime state (generated at boot)
│ ├── hermes/
│ ├── openclaw/
│ ├── accounts/
│ └── chat/
└── scripts/
├── start.sh
└── unify-stack.sh
Migrate legacy state from ~/.hermes or ~/.qclaw-oversea:
./scripts/unify-stack.sh --migrate
Environment overrides:
| Variable | Default |
|---|---|
GOIS_STACK_ROOT | ./.stack |
GOIS_HERMES_HOME / HERMES_HOME | .stack/hermes |
GOIS_OPENCLAW_SOURCE | ./vendor/openclaw |
GOIS_OPENCLAW_STATE_DIR | .stack/openclaw/state |
MCP integration (Cursor / IDE)
flowchart LR
Cursor["Cursor IDE"]
SK["qclaw-skills"]
CR["qclaw-cards"]
JB["qclaw-jobs"]
SW["qclaw-swarm"]
RF["ruflo"]
Cursor --> SK
Cursor --> CR
Cursor --> JB
Cursor --> SW
Cursor --> RF
SK -->|"slides, AWS, files"| Gois
CR -->|"Kanban, teams"| Gois
JB -->|"cron, health"| Gois
SW -->|"multi-agent"| Gois
| MCP server | Tools (examples) |
|---|---|
qclaw-skills | team_files_search, slides_batch_images, aws_overview, monitor_update |
qclaw-cards | get_cards, move_card, list_teams, get_errors |
qclaw-jobs | jobs_health, jobs_list_running, jobs_cron_list |
qclaw-swarm | swarm_create, swarm_health, run_team_swarm |
ruflo | Swarm orchestration for development |
Setup RuFlo for Cursor:
./scripts/setup-ruflo-dev.sh
Dashboard URLs
| URL | Description |
|---|---|
/ | Monitor status |
/chat | OpenClaw / QClaw chat (DeepSeek + skills) |
/chat/ruflo | RuFlo-integrated chat |
/chat?whatsapp=1 | WhatsApp session |
/skills | OpenClaw skills catalog |
/roles | Hermes role profiles |
/kanban | Team Kanban boards |
/whatsapp/status | wacli queue and listener state |
WhatsApp bridge
flowchart LR
Phone["Phone"]
WACLI["wacli"]
WH["POST /whatsapp/inbound"]
Chat["/chat session"]
Agent["DeepSeek + skills"]
Send["wacli-send.sh"]
Phone --> WACLI
WACLI --> WH
WH --> Chat
Chat --> Agent
Agent --> Send
Send --> Phone
Enable in config.yaml under whatsapp_digest. See README.pt.md for detailed Portuguese setup notes, or config.example.yaml.
macOS auto-start
./scripts/start.sh --launchd
Logs: gois.stdout.log, gois.stderr.log
Tests
.venv/bin/pytest
Security
- Never commit
.env,sk-…keys, or filledconfig.yaml - Use
config.example.yamlas template; revoke exposed keys immediately - See SECURITY.md
License
See repository; treat as internal use until a LICENSE file is added.