Skip to content

Telemak — cluster enrolment

A Telemak can stand alone, or it can join the fleet. Enrolment makes its models show up in the cluster catalog without giving up its independence.

A Telemak runs perfectly well on its own — one Mac, no orchestrator. But if you also run an OdyssAI-X cluster, you can enrol the Telemak so its models join the unified catalog. The OdyssAI-X orchestrator then proxies requests to it over HTTP. No RDMA, no Docker on the Telemak side.

OdyssAI-X represents a Telemak as a cluster of kind=telemak with backend http-proxy: instead of SSH-spawning runners, the orchestrator forwards /v1/chat/completions to the Telemak’s own endpoint. The Telemak keeps owning its model lifecycle — you still load and unload from its menu bar; OdyssAI-X just routes to whatever is loaded.

From the OdyssAI-X dashboard:

  1. Settings → Clusters → Add Telemak.
  2. Enter:
    • an id (e.g. telecode) — becomes the model id clients route to,
    • a label (e.g. “TeleCoder”),
    • the upstream URL (http://<telemak-ip>:8003),
    • optionally the SSH target if you want the dashboard to manage the app remotely.
  3. Save. The dashboard probes the Telemak’s capability contract and its loaded models join the catalog.

Or via the Configurator’s Solo mode at install time — same result, written into the topology for you (see Install the stack).

Once enrolled, the Telemak’s models appear in /v1/models alongside the cluster’s own, with their real per-model capabilities. Clients (Companion, an IDE agent) route to them by the Telemak’s id — e.g. model: "telecode" — exactly like any other model. The orchestrator hides the fact that it’s a single Mac behind an HTTP proxy.

StandaloneEnrolled
Setupopen the app+ one dashboard entry
Reached bydirect :8003the cluster’s :8000, by id
Use it forone Mac, one clienta mixed fleet under one catalog

The typical deployment is Telemak on a workstation and an OdyssAI-X cluster on a server, both fronted by Companion — Telemak enrolled so the chat picker shows one list.