Provider health & smoke testing
Validate that your LLM providers are reachable and usable before relying on them in agent conversations.
Health check
POST /pfw/v1/provider-health/{provider}
Runs the full validation flow:
- Discovery — queries the provider's API for the model list.
- Validation — checks that at least one compatible model is found.
- Status update — updates the credential's status in the UI.
Returns connected or failed with an error classification (auth_failure, network_error, no_models_found, unknown).
Smoke test
POST /pfw/v1/provider-runtime/{provider}/smoke
Goes a step further: it actually calls the LLM with a tiny one-sentence generation. Tries up to a handful of candidate models and returns the first success with usage data (tokens, cost). If every candidate fails, the last error is returned so you can debug.
Run a smoke test after adding a new provider or rotating an
API key. It catches configuration errors (wrong base URL, invalid
model name) that a plain health check might miss.
API key. It catches configuration errors (wrong base URL, invalid
model name) that a plain health check might miss.