lynox-ai/lynox

**Run your business. Not your tools.**
One source-available (ELv2) agent that learns your business — connects any API, runs your workflows, researches the web, and monitors for changes. Persistent knowledge graph, workflow capture, sub-agents, background worker. Bring your own LLM — **Anthropic Claude** and **Mistral** are natively supported and tested on every release, and **Ollama** and **Fireworks AI** are verified with a real end-to-end tool-calling run. Other OpenAI-compatible endpoints (LM Studio, vLLM, LocalAI, Groq, Together AI, OpenAI itself, LiteLLM) and Google Vertex AI are wired but experimental. Self-hosted, no vendor lock-in.
> [!IMPORTANT]
> **lynox is a CLI, not a library.** Run npx @lynox-ai/core, not npm install @lynox-ai/core. The npm page sidebar suggests npm i by default, but that only installs lynox as a dependency without running anything.
What it does
- **"Check my emails and tell me what's important"** — lynox reads your inbox, prioritizes, drafts replies in your tone. Tomorrow it remembers your style.
- **"Summarize this PDF"** — Drop a file, get structured key points. lynox saves the context for later.
- **"Research [topic] and give me the key points"** — Plans research phases, searches the web, cites sources, stores findings in the knowledge graph.
- **"Do this every Monday"** — teach lynox the workflow once, save it as a template, schedule it. Next Monday it runs on its own.
- **"What changed on competitor.com?"** — Started as a one-time check. Now runs weekly and notifies you when something shifts.
Quick start
You need Node.js 22+, Docker, and an LLM credential. The interactive installer walks you through provider choice (**Anthropic** or **Mistral** are the natively-supported options; the OpenAI-compatible **Custom** path lets you point at anything else), generates a docker-compose.yml, pulls the image, and opens the Web UI at localhost:3000. In the Web UI's LLM settings, one-click presets cover Ollama, LM Studio, vLLM, LocalAI, Groq, Together AI, and Fireworks — Ollama and Fireworks are verified end-to-end with a real tool-calling run; the rest are experimental, and each tile says which.
Prefer to edit .env yourself before the first run? Jump to manual Docker below.
Documentation
Full docs at **docs.lynox.ai** — getting started, integrations, features, API reference.
**Roadmap:** see ROADMAP.md for what's next, later, and under evaluation.
How it works
**BYOK** — You provide your LLM credential (Anthropic and Mistral are the natively-supported and tested paths; Ollama and Fireworks are verified end-to-end; other OpenAI-compatible endpoints / LiteLLM are wired but experimental; Vertex is wired but not usable in the Docker image). When self-hosted, lynox calls the LLM API directly: no proxy, no middleman, and no telemetry unless you opt in via LYNOX_BUGSINK_DSN for error reporting. Your data stays on the host you control. (Managed tier opt-in routes via the lynox control plane — see the Managed page for that flow.)
Key capabilities
- **Knowledge Graph** — Learns entities, relationships, and contradictions across your business. 100 languages, fully local.
- **Web UI** — Primary interface. Chat, knowledge browser, run history, settings, integrations. Installable as PWA. QR code login for instant phone access.
- **Activity Bar** — Every tool call streams live with its current sub-phase (e.g. "Reading API docs..." → "Extracting auth..."). No mysterious 30-second waits.
- **Background Worker** — Scheduled tasks, URL monitoring, recurring workflows.
- **Mobile Access** — Voice input (Whisper or Voxtral STT), push notifications, and mail/voice workflows. Install as a PWA for a native app feel, or use any mobile browser directly.