Skip to content
OdyssAI

Getting started

From a fresh Apple Silicon Mac to a first inference call in about ten minutes. Every path installs by dragging an app to Applications — no Python, no git clone, no terminal unless you want one.

Choose your path

Section titled “Choose your path”
PathHardwareApp you installTime
One Mac → TelemakMac Studio or MacBook ProTelemak.app~5 min
Multiple Macs → cluster2–5 Mac Studios on TB5OdyssAI Configurator~30 min
Cloud-only → chatAny MacOdyssAI Configurator~5 min

Most users are on Path A. Path B is for frontier models across a cluster. Path C is the cheapest way to try the chat client without owning the hardware.

Path A — One Mac

Section titled “Path A — One Mac”

One machine? You want Telemak — a native menu-bar app, no Docker, no orchestrator. Full walkthrough in Telemak getting started; the short version:

  1. Install — download Telemak-<version>.dmg from the releases, drag Telemak.app to Applications, open it. (First launch: Gatekeeper → System Settings → Privacy & Security → Open Anyway.) A menu-bar icon appears; the daemon serves on http://localhost:8003.
  2. Load a model — menu-bar icon → Load model → pick one that fits your RAM (the dialog tells you which fit). A 35 B MoE fits a 64 GB Mac.
  3. Smoke it:
    Terminal window
    curl http://localhost:8003/v1/models
    curl http://localhost:8003/v1/chat/completions \
      -H "Content-Type: application/json" \
      -d '{"model":"<your-model>","messages":[{"role":"user","content":"Hello, sea."}]}'

That’s a working engine. To add a chat window with history and memory, install Companion (Path C) and point it at http://localhost:8003.

Path B — Multi-Mac cluster

Section titled “Path B — Multi-Mac cluster”

Several Mac Studios? You install the whole server side with the OdyssAI Configurator — one app, one wizard. Full procedure in Install the stack; the shape of it:

  1. Serveur — run the Configurator on a Mac mini, profile Serveur. You get OdyssAI-X (:8000) + Companion (:3100).
  2. Engines — profile Engine for each Mac Studio; it provisions the MLX runtime over SSH.
  3. Topology — the Build step probes your Thunderbolt 5 wiring and writes the RDMA cluster for you.
  4. First inference:
    Terminal window
    curl http://<server-ip>:8000/admin/clusters          # your cluster appears
    curl -X POST http://<server-ip>:8000/admin/default/load \
      -H 'Content-Type: application/json' \
      -d '{"model":"mlx-community/Qwen3-7B-MLX-8bit"}'
    curl -X POST http://<server-ip>:8000/v1/chat/completions \
      -H 'Content-Type: application/json' \
      -d '{"model":"default","messages":[{"role":"user","content":"Hello, sea."}]}'

Point Companion at http://<server-ip>:8000 and you have the full stack: distributed engine, unified catalog, chat UI, memory, projects.

Path C — Cloud-only

Section titled “Path C — Cloud-only”

Want the chat client without local models? The Configurator’s Serveur profile installs Companion on its own; you just add a cloud key.

  1. Run the Configurator (profile Serveur) on any Mac — it brings up Companion on :3100.
  2. Open http://localhost:3100, create your admin account.
  3. Settings → Infrastructure → Engine → Add Cloud Provider — paste your OpenRouter / Anthropic / OpenAI key.

That’s the whole setup. Chat, history, memory, projects, all on a cloud model. Add Telemak or an OdyssAI-X cluster later as a second engine — Companion keeps a separate catalog per engine and routes per message.

What next

Section titled “What next”