Skip to main content

Providers

From Admin → Providers, configure access to external AI services (credentials, endpoint) that managers will then use to create AI, OCR or embedding models from their own interface.

This split reflects a division of responsibility: an administrator configures and secures access (API key, endpoint), while a manager picks, among the available providers, which model to expose to their agents, pipelines or knowledge bases — without ever seeing or handling the API key itself.

Configuring a provider​

  • Name — label shown in provider selectors on the Manager side.
  • Type — a provider can check one or more types at once: LLM, Embedding, OCR. A single OpenAI account, for example, can serve as both an LLM and an embedding provider — no need to duplicate it.
  • Provider key — the technical integration used to route calls: openai, anthropic, google_genai, ollama, fireworks, openrouter, baseten, azure_openai or mistral.
  • Endpoint (optional) — custom URL, required for a self-hosted deployment (e.g. Ollama) or an OpenAI-compatible proxy.
  • API key (encrypted) — not required for Ollama.
note

The "Azure OpenAI" provider does not have a dedicated Azure integration (no specific handling of API version or deployment name): it is treated as a generic OpenAI-compatible endpoint. For an Azure OpenAI deployment, enter its endpoint URL in the field provided.

Any provider for which you set a custom endpoint is treated as OpenAI-compatible, which lets you connect any service exposing that kind of API.

Deletion​

A provider still referenced by one or more models (AI, OCR or embedding) cannot be deleted — first free up or reassign those models from Manager → Settings → AI Models.