Multi-agent AI workforce running on a Jetson Orin Nano Super inside NVIDIA NemoClaw/OpenShell sandboxes. Provides always-on AI assistants for email, calendar, messaging, and research -- fully local, no cloud dependency.
# 1. Edit config
vi config.yaml # Set your provider IPs, enable channels
# 2. Set secrets
cp .env.example .env
vi .env # Add API keys, OAuth tokens
# 3. Apply and start
./scripts/apply-config.sh
./scripts/start-gateway.sh- Primary inference: Qwen3-8B on Mac via LM Studio (LAN)
- Fallback inference: Gemma3:4B on Jetson via Ollama (local)
- 5 agents: Orchestrator, Email, Calendar, Messaging, Research
- Channels: Slack, Discord, WhatsApp (iMessage planned)
- Tools: Gmail, Google Calendar via MCP servers (extensible)
See docs/ARCHITECTURE.md for the full system design.
All settings live in config.yaml. See docs/CONFIGURATION.md for details.
| Document | Contents |
|---|---|
| Architecture | System design, hardware, network topology, memory budget |
| Configuration | config.yaml reference, providers, agents, channels, MCP |
| Agents | Agent design, SOUL.md writing guide, adding new agents |
| Deployment | Prerequisites, setup steps, operating procedures |
| Channels | Slack, Discord, WhatsApp, iMessage setup guides |
| MCP Servers | Gmail, Calendar setup, custom server development |
| Troubleshooting | Common issues, diagnostic commands |
| Component | Specification |
|---|---|
| Jetson | Orin Nano Super, 8 GB, MAXN SUPER mode |
| Mac | Inference server (LM Studio, Qwen3-8B) |
| Network | LAN, Jetson @ 10.0.0.7, Mac @ 10.0.0.232 |
- Phase 1: Foundation (config, agents, directory structure)
- Phase 2: Messaging channels (Slack, Discord, WhatsApp)
- Phase 3: Google integrations (Gmail, Calendar MCP)
- Phase 4: Automation (cron jobs, standing orders)
- Phase 5: Hardening (security audit, health checks)
- Phase 6: Extension (more integrations)