Support · installing models

Connect a model in under a minute.

Odysseus speaks the OpenAI-compatible standard — bring any endpoint, or use the managed model that's connected the moment your workspace boots.

Step 1

Open setup

In your workspace, type /setup or go to Settings → Model endpoints.

Step 2

Add an endpoint

Paste the base URL (ending in /v1) and an API key if the provider needs one.

Step 3

Pick it & chat

The model lands in your picker. Select it and start chatting — instantly.

FAQ

How do I install a model?

Open your workspace, type /setup in chat (or open Settings → Model endpoints), click Add endpoint, paste an OpenAI-compatible base URL and (if required) an API key, and save. The model appears in your picker immediately. New workspaces also come with a managed model already connected — so you can chat the moment it boots.

What kinds of models / endpoints work?

Anything that speaks the OpenAI-compatible API (/v1/chat/completions): OpenAI, OpenRouter, Together, DeepInfra, Fireworks, or your own vLLM / Ollama server. Paste its base URL + key and it just works.

Can I get a dedicated GPU (e.g. RTX 5090)?

Dedicated-GPU hosting (including RTX 5090) is an enterprise option on request — coming soon. Today every workspace runs on fast managed inference with per-second metering and a hard spend cap. Need dedicated hardware for a client? Get in touch and we'll scope it.

Do I have to bring my own API key?

No — managed model access is connected out of the box. If you do bring your own key (BYOK), note that the model provider may bill usage separately. If your plan includes model usage, the limits are stated on your plan.

Is my data private?

Each workspace is project-isolated and privacy-forward — files, prompts, notes, and endpoints stay your side. For a self-hosted endpoint, inference runs on compute you control. Don't paste secrets into shared workspaces; rotate keys when needed.

The model isn't showing up / says offline — what now?

Re-open Settings → Model endpoints and hit refresh/probe. Check the base URL ends in /v1, the API key is valid, and the endpoint is reachable. For a self-hosted server, confirm it's serving on the expected port. Still stuck? Email support and include the endpoint host (not the key).

Chat vs Agent mode?

Use Chat for direct questions, writing, research, and planning. Use Agent for longer tasks that need tools, files, and multiple steps — then save the repeatable ones as agent recipes.

Still stuck? Email support@odysseus.ai with your workspace name and the endpoint host (never your API key).

Open your workspace →