LLM providers

Connect large-language-model providers for AI features.

Where: Project → Integrations → LLM · /projects/<project>/ai/llms/integrations

Connect an LLM provider so the project's AI features — chat, completions, agents — can use its models.

Screenshot TODO — capture comes from testing-plan area 07-llm-integration.

The list

Columns: Integration Name, Provider, Default Model, Enabled, and a per-environment Default badge. The backend marks the first integration you create as that environment's default; there is no manual "set as default" action. If several integrations are active and none is the default, the list shows a warning.

Add a provider

Add New Integration opens the gallery (…/pick) with one card per provider: OpenAI, Anthropic, Ollama, Groq, Google, Mistral, OpenRouter, Grok. Pick one to open the form:

Field	What it does
API Key	Required for every provider except Ollama (which has no key — you point at your own server instead). Stored encrypted and never returned by the API.
Endpoint	Optional custom endpoint URL; empty uses the provider's default. For Ollama this is your server address, e.g. http://localhost:11434. Hidden for Google.
Default Model	Optional — the model used when a request does not specify one, e.g. gpt-4o-mini or claude-sonnet-4-6.
Integration Name	How the integration is listed, e.g. My OpenAI Connection.

Saving runs an automatic connection test (the backend sends a probe request to the provider); the last test result is shown on the edit page. Edit reopens the same form, where you can also enable, disable, or delete the integration.

Per-provider details: Providers.

API reference

Endpoints: AI.