Health checks¶
Two endpoints, plus a CLI command, for monitoring and probing.
GET /health¶
Lightweight — does not hit upstream providers. Use this for load-balancer liveness probes, container health checks, etc.
Response:
{
"status": "ok",
"version": "0.1.0",
"providers": {
"openai_compat": ["openai", "gemini", "kimi", "xai", "groq", "mistral", "deepseek", "together", "perplexity"],
"anthropic": true,
"azure": false,
"bedrock": true
},
"models_registered": 38
}
anthropic, azure, bedrock are booleans indicating whether the
required env vars are set (not whether they're actually working).
GET /health/providers¶
Pings every OpenAI-compatible provider's /models endpoint. Slower (one
HTTP round-trip per provider, in parallel) but tells you which providers
are actually reachable and authenticated.
Response (abbreviated):
{
"overall": "degraded",
"providers": {
"openai": { "status": "ok", "http_status": 200, "url": "https://api.openai.com/v1/models" },
"gemini": { "status": "ok", "http_status": 200, "url": "https://generativelanguage.googleapis.com/v1beta/openai/models" },
"kimi": { "status": "ok", "http_status": 200, "url": "https://api.moonshot.ai/v1/models" },
"groq": { "status": "unconfigured", "detail": "no API key" },
"perplexity": { "status": "error", "http_status": 401, "url": "https://api.perplexity.ai/models" }
}
}
overall is ok if every configured provider returns ok, otherwise
degraded.
This is a deeper check than /health. Use it for diagnostic dashboards —
running it every few seconds against many providers can rack up rate limits.
ollama-proxy doctor¶
The same connectivity check as /health/providers, plus environment-variable
sanity checks, all from the CLI:
Returns exit code 0 if everything is healthy, non-zero on any failure.
Useful in CI/CD before deployments and when first setting up the proxy.
See the CLI reference for sample output.