Skip to content
OdyssAI

Welcome to Companion

The house the son comes home to.

Companion is the chat client of the OdyssAI stack. It pairs to an inference engine — local cluster (OdyssAI-X, Telemak, Ollama, LM Studio, vLLM), local Mac (MLX), or a cloud key (OpenRouter, Anthropic, OpenAI) — and gives you a consistent client across desktop and mobile.

This guide is the canonical reference for Companion. The topics are independent; pick the one that matches what you are trying to do. The same Markdown sources render at companion.odyssai.eu/docs.

What Companion is, in one sentence

Section titled “What Companion is, in one sentence”

A web client that pairs to an inference engine, keeps your chat history, memory, skills and projects in a single account, and exposes both a chat UI and an MCP brain endpoint that external agents can call back into.

What you can do here

Section titled “What you can do here”
  • Chat with any model you have access to — full conversation history, edit / regenerate, attachments, voice, code-block helpers, multi-model picker.
  • Organise chats into Projects that share a system prompt, memory toggles, and (optionally) a project wiki.
  • Extend the agent with Skills (agentskills.io-spec packages), MCP servers (Notion, Linear, Tavily, GitHub, Filesystem, your own), and Agents tokens that turn Companion into an MCP brain for external coding agents (Cline, Continue.dev, Claude Desktop, Cowork).
  • Remember with a per-user knowledge graph (Némo, Karpathy-style memory) and per-project memory. Curate by hand or let the agent append via companion_remember.
  • Voice in and out — push-to-talk on Space, full Talk mode for hands-free use.
  • Speak many engines — gateway mode (OdyssAI-X direct), hybrid (capabilities from engine, inference via LiteLLM), or legacy (LiteLLM only). One client, every backend.

Two minds, one stack

Section titled “Two minds, one stack”
  • Companion — the client you are reading this in. React + Hono + Postgres, packaged as a Docker image you deploy where you want.
  • OdyssAI-X — the distributed MLX inference engine. OpenAI- and Anthropic-compatible. Companion is happiest paired with it, but speaks plain OpenAI, plain Anthropic, and OpenRouter just as well through the legacy rail.

Companion does not ship its own model. It is a client. You decide what is behind the chat window.

What is NOT in here

Section titled “What is NOT in here”
  • How to build OdyssAI-X, run a Mac Studio cluster, or deploy your own Companion. Those live in the developer docs (OdyssAI-X →, Deploy →).
  • The API reference for companion_* MCP tools — see Agents tokens in this guide; full schema is at /api/mcp via tools/list.
  • Pricing and licensing — your admin contacts you directly.

How to navigate this guide

Section titled “How to navigate this guide”
  • First time? Read Getting started (next page), then Chat basics. Skip the rest until you need it.
  • Pairing an engine? Engine pairing.
  • Building a brain for your IDE? Agents tokens.
  • Tuning a specific bug? Troubleshooting.
  • Lost on a term? Glossary.

A note on the voice

Section titled “A note on the voice”

The agent that lives in Companion calls itself Némo. Némo is not a model. Némo is the orchestrator above whatever model you have routed — the persona, the memory, the relationship, the editor’s voice. The model is the substrate. Némo is what the substrate says when you address it.

You will see the name throughout this guide. If you renamed your assistant, the same patterns apply.

Read first

Section titled “Read first”