Models and providers

Ódinn separates the agent runtime from the model provider. You choose how inference is supplied:

• browser-based OAuth;
• API keys;
• device or CLI authentication where supported;
• local OpenAI-compatible servers;
• Ollama or LM Studio.

Start with the guided flow:

odinn onboard

Cloud providers

Choose OpenAI/ChatGPT during guided onboarding to use the browser sign-in flow. API-key paths are available for OpenAI, OpenRouter, and many OpenAI-compatible services.

export OPENAI_API_KEY="..."
odinn onboard --provider openai --auth api-key

Cloud providers receive the prompts, recalled context, and tool results sent to their API. Their model availability, quotas, rate limits, billing, and service behavior remain outside Ódinn’s local guarantees.

Local models

Ódinn can use an existing Ollama model:

odinn onboard --provider ollama --model <installed-model>
odinn start

Ódinn does not assume or download a particular local model. Start the model server and use the exact model name it already exposes.

LM Studio, vLLM, SGLang, and LiteLLM presets are also available for compatible local or self-hosted endpoints.

Inspect available presets

odinn config provider catalog

The catalog is the canonical installed list. A preset means Ódinn knows the adapter and configuration shape; it is not a promise that a provider account, endpoint, or model is currently available.

Change the default model

odinn config model list
odinn config model default <provider:model>

Run a real capability check after any provider or model change:

odinn onboard --verify --non-interactive

For setup details, see Providers and the provider catalog.