Skip to main content
Fleet Prime Agent brings its own provider credentials: you pick the provider and model per session. The providers in Settings → Providers come from Fleet’s own provider catalog, plus custom providers from models.json and providers that only have a stored credential. The model catalog comes from the Prime Agent daemon.

Add a provider

Two equivalent paths:
  • Settings → Providers in the web workspace.
  • /login in the composer.
Credentials are stored by the engine under ~/.prime/agent/ and are shared with any prime-agent install on the same machine, so signing in once covers both the web workspace and the terminal. When you save an API key in Fleet, Fleet writes it to the engine’s auth.json under ~/.prime/agent/ and keeps credential files readable only by your user. Providers that have a stored credential but no catalog entry also appear in Settings → Providers, so you can review or remove them.

OAuth providers

Fleet cannot complete OAuth sign-in for providers that use OAuth instead of a static API key. Settings → Providers shows terminal instructions instead, and POST /api/chat/providers/oauth returns 501. To sign in, start Prime Agent in a terminal against the same account as Fleet (for example prime-agent or fleet-agent agent), then run /login and choose the provider. Fleet reads the stored credential on every request, so you do not need to restart it. If Fleet uses a non-default account directory (the default is ~/.prime/agent), start Prime Agent with the same PRIME_AGENT_CODING_AGENT_DIR. See Terminal interface. The engine refreshes tokens for long-running sessions.

OpenAI-compatible endpoints

For self-hosted or OpenAI-compatible servers, the workspace can probe an endpoint’s /v1/models listing (POST /api/chat/models/discover) and register the discovered models. Custom provider definitions are kept by the server’s custom provider store. When you open Settings → Providers → Add models, Fleet refreshes the model list of each configured, discoverable provider and caches the result for 60 seconds. Each provider has a Refresh button with its name; select it to force a refresh. If a refresh fails, the button shows Refresh failed (stale) and the previous results stay. Discovery URLs that include a username or password are rejected. A custom provider baseUrl must not contain credentials. Saving one with a username or password (user:pass@) or a credential-like query parameter (such as key, token, secret, or signature) returns 400 with baseUrl must not contain credentials; use the API key field instead. GET /api/chat/providers strips such credentials from a baseUrl that is already stored before returning it to the browser. Every custom provider in models.json that is not a built-in provider needs a non-empty apiKey. If one is missing it, the runtime drops every custom provider in the file without an error. Fleet reports the problem in the diagnostics of GET /api/chat/models:
Built-in providers that only override baseUrl do not need apiKey. When you save a custom provider that uses the OpenAI Chat Completions API (openai-completions), Fleet applies gateway-compatible defaults:
  • If the base URL host is not api.openai.com, Fleet sets compat.supportsStore to false. Third-party gateways often reject the OpenAI-only store field.
  • Models whose id starts with gpt-6 or gpt-5.6 default to reasoning: true, with the off thinking level mapped to none. This lets function tools work on /v1/chat/completions.
Fleet only fills in these values when you have not set them. Any compat setting or per-model reasoning and thinkingLevelMap value you already defined is kept.

Pick a model

Choose the model in the composer’s model picker when starting a session, or switch it mid-session. The workspace sends the selected model with the next turn (POST /api/chat). The model catalog comes from the Prime Agent daemon (GET /api/chat/models), including thinking-level support per model. The daemon re-reads models.json and auth.json on each request, and Fleet caches the result for about 5 seconds, so edits you make outside Fleet show up within a few seconds. The picker lists only working models: models whose provider has credentials configured and that are enabled in Settings → LLM Models. Models you enable there appear in the picker as soon as you save. A provider counts as configured when you saved a key or signed in, or when its API key environment variable (for example ANTHROPIC_API_KEY or GROQ_API_KEY) is set in the environment you launched fleet-agent from. A models.json apiKey that names an environment variable that is not set counts as unavailable. If the picker is empty, add a provider in Settings → Providers, then enable models in Settings → LLM Models. GET /api/chat/models?scope=all returns the full catalog regardless of these filters. If a saved default model id no longer exists in a known provider’s catalog, for example after an upstream catalog change, Fleet falls back to that provider’s first available model instead of failing the turn. A model from a provider without credentials still fails.

Adding a new provider to the engine

Engine features, including providers and models, are developed in PrimeIntellect-ai/prime-agent, not in this repository. Contribute the provider upstream, then upgrade Fleet’s pinned runtime to a release that includes it. See Runtime pin and releases.
Last modified on November 8, 2026