Providers and models

Local: Ollama, LM Studio, any OpenAI-compatible server. Cloud: OpenRouter, OpenAI, Anthropic, Google Gemini, xAI, Mistral, DeepSeek, Groq, Together, Fireworks, Azure OpenAI.

Only reasoning models can be chosen. Dolphin Pod decides using the provider’s own metadata where available (OpenRouter supported parameters, Ollama capabilities, Anthropic and Gemini model capabilities) and otherwise a bundled catalog (packages/shared/catalog/models.json). Non-reasoning models are shown disabled with the reason.

The reasoning-effort control shows only what the selected model supports: levels (for example low / medium / high), a thinking-token budget, an on/off switch, or nothing.

Keys are stored in the system keychain and never written to the database.

Setup guides: Ollama, LM Studio, OpenRouter.