Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/pewdiepie-archdaemon/odysseus/llms.txt

Use this file to discover all available pages before exploring further.

Odysseus connects to any OpenAI-compatible endpoint. Providers are managed inside the app — you add a base URL and an optional API key through the Settings UI, and Odysseus probes the endpoint, discovers the available models, and starts serving them in every picker immediately. The .env file is only needed for deployment-level overrides and pre-seeding credentials before first boot; for day-to-day use, prefer adding providers in the app.

How to add a provider

1

Open Settings

Navigate to Settings → Models → Add Endpoint.
2

Enter the base URL

Paste the OpenAI-compatible base URL for your provider (see the reference table below). Do not include /chat/completions — Odysseus appends the correct path automatically.
3

Add an API key (if required)

For cloud providers paste your API key. For local providers running on your own machine no key is needed.
4

Save

Odysseus probes the endpoint, caches the model list, and makes the models available across Chat, Agent, Deep Research, and Compare immediately.

Provider reference

The Cookbook can download models and start a local vLLM or llama.cpp server for you automatically, handling port selection and GPU setup. If you want to serve a local model, try Cookbook first — it registers the endpoint in Settings on its own.
ProviderBase URLAPI Key
Ollama (native install)http://localhost:11434/v1None
Ollama (Docker — host machine)http://host.docker.internal:11434/v1None
vLLMhttp://localhost:8000/v1None (configurable port)
llama.cpp serverhttp://localhost:8080/v1None
LM Studiohttp://localhost:1234/v1None
OpenAIhttps://api.openai.com/v1OPENAI_API_KEY from platform.openai.com
OpenRouterhttps://openrouter.ai/api/v1API key from openrouter.ai
GitHub CopilotHandled via the separate Copilot integration flow

Docker and Ollama on the host

When Odysseus runs inside Docker and Ollama runs on your host machine (the most common setup on Windows and macOS), the correct endpoint URL is:
http://host.docker.internal:11434/v1
Ollama must also be listening on all interfaces, not just its own loopback. Start it with:
OLLAMA_HOST=0.0.0.0:11434 ollama serve
This connects Odysseus (inside the container) to an Ollama server running on the host machine. host.docker.internal is Docker’s built-in hostname that resolves to the host from inside any container.
host.docker.internal is only available inside Docker. For native installs use http://localhost:11434/v1 as usual.

LM Studio in Docker

LM Studio must be configured to serve on all interfaces (not just localhost) before Odysseus can reach it from inside Docker. In LM Studio’s server settings, change the bind address to 0.0.0.0, then use:
http://host.docker.internal:1234/v1

Model discovery

Odysseus scans common local model-server ports automatically, including Ollama’s 11434 and vLLM/llama.cpp defaults in the 8000–8020 range. To point it at additional hosts (remote servers, Tailscale peers, etc.) set LLM_HOSTS in .env:
# .env
LLM_HOSTS=llm-host.local,backup-llm.local
Multiple hosts are comma-separated. Odysseus always includes the primary LLM_HOST alongside any extra hosts you list.

Pre-seeding with .env

You can pre-seed the OpenAI API key in .env so it is available before first boot. This is useful for automated or container deployments:
# .env — only needed for pre-seeding; prefer adding in-app
OPENAI_API_KEY=your_openai_api_key_here
Do not commit real API keys to version control. The .env file is gitignored by default. Only use this for deployment-level configuration, not routine key management.

Custom CA certificates (LLM_CA_BUNDLE)

If your provider uses a TLS certificate that is not in the default system trust store — for example, an on-premise LLM gateway with an internal corporate CA, or GigaChat which uses a Russian Trusted Root CA — set LLM_CA_BUNDLE to a PEM file containing the missing root certificate(s):
# .env
LLM_CA_BUNDLE=/etc/odysseus/ca/extra-roots.pem
This bundle is layered on top of the system trust store; TLS verification stays enabled for all hosts. Without it, connections to those providers will fail with CERTIFICATE_VERIFY_FAILED.

GitHub Copilot

GitHub Copilot is handled through a dedicated integration flow separate from the standard endpoint form. Look for the Copilot option in Settings → Models — it walks through the authentication steps specific to Copilot’s subscription-based API.

Build docs developers (and LLMs) love