Add-ons

Optional integrations that extend what Companion can do. Configure them in Settings → Extensions → Add-ons. Most are off by default.

Add-on	What it enables	Slash command
LiteLLM	Legacy inference rail (cloud routing)	—
Auto Router	Semantic routing — picks the model per message	—
Obsidian	Vault sync for the memory wiki	—
Web Search (Tavily)	Live web search tool for the agent	—
Voice Live	Full-duplex voice via Gemini Live	—
Hermes Agent	Real-machine coding agent	`/hermes`
Pi	Reflective conversational TUI agent	`/pi`
ComfyUI Imager	Flux image generation	`/comfyui`

LiteLLM

The legacy inference rail — routes through a LiteLLM proxy when no engine is in gateway mode, or to reach cloud providers via an intermediary. In gateway mode with OdyssAI-X you can disable LiteLLM entirely (Settings → Inference) and route through the engine directly. See Engine pairing. It is a fallback, not the centre.

Auto Router

Inspects each incoming message and picks the best model pool before forwarding — useful in Easy mode. (This is Companion’s chat router, distinct from CoeOS, the engine-side per-skill router for agents.)

Obsidian

Syncs your Obsidian vault into the memory wiki: a vault ZIP export/import, plus a bearer token for the Companion Obsidian plugin that pushes vault changes back automatically. After sync the wiki re-indexes. See Memory.

Web Search (Tavily)

Gives the agent a live web_search tool. Requires a Tavily key. Appears in the agent’s catalog when agent mode is on.

Voice Live

Full-duplex voice via Gemini Live (requires a Gemini key) — a Voice button in the chat header opens a low-latency bidirectional surface. Separate from the always-on, browser-side push-to-talk.

Hermes Agent · Pi · ComfyUI Imager

The connector add-ons — each wires an external bridge to a slash command:

Hermes Agent → /hermes (workstation coding agent). Bridge URL + optional token.
Pi → /pi (reflective TUI). Bridge URL.
ComfyUI Imager → /comfyui (image generation). Bridge URL + token.

Full detail: Coding agents and ComfyUI.