NEXUS-R is a local-first agent runtime that lets you execute AI-assisted workspace tasks with full control over where your data goes. It routes tasks between local models (Ollama) and cloud providers based on complexity, cost, and your privacy preferences.
Yes. NEXUS-R itself is open-source (MIT license). You bring your own API keys for cloud providers. Local models via Ollama are free.
- macOS 14+ (Apple Silicon and Intel)
- Linux (Ubuntu 22.04+, Fedora 39+, Arch)
- Windows 11+ (via WSL2 recommended)
| Provider | Type | Status |
|---|---|---|
| Ollama | Local | Full support |
| OpenAI | Cloud | Full support |
| Anthropic (Claude) | Cloud | Full support |
| OpenRouter | Cloud | Full support |
| Groq | Cloud | Full support |
| Google (Gemini) | Cloud | Full support |
| Cohere | Cloud | Full support |
| Mistral | Cloud | Full support |
| Together AI | Cloud | Full support |
| Azure OpenAI | Cloud | Coming soon |
Yes, if you configure local models via Ollama. Cloud providers require internet access.
The Cognition Router analyzes your task's complexity and routes it:
- Simple tasks (T1-T2) -> Local models (free, private)
- Complex tasks (T3-T5) -> Cloud providers (capable, metered)
You can override routing per-task or set a default profile.
Only when routing decides a cloud model is needed. You can:
- Force local-only mode
- Set a "local preference" profile
- Review the routing rationale before execution
- Add provider configuration to
nexus_r/config.py - Implement adapter in
modules/cognition_router/src/ - Add unit tests in
tests/unit/ - Update the frontend provider list
See the full guide in CONTRIBUTING.md.
Not yet -- the runtime is designed as an application. Library packaging is on the roadmap.
- Backend:
~/.nexus/logs/ - Frontend: Browser DevTools console
- Docker:
docker compose logs
# WARNING: This deletes all conversations, memory, and settings
rm -rf ~/.nexus/
# Or via Docker:
docker compose down -vGitHub Issues -- check existing issues first.