Inventory & health

GET /health

Liveness probe

Returns 200 with a small JSON body when the worker is ready.

Connection refused or non-200 means aistack is down or still starting; consumers should surface “service unreachable” with a hint to start the dev server. This endpoint never blocks on backend health — it only confirms the FastAPI worker itself is alive.

Responses

200

Successful response.

• application/json → HealthResponse

GET /v1/models

List servable models

OpenAI-compatible model list. Lists only backends that will actually serve a request right now:

• ASR entries are emitted for every provider whose ML library is importable in the running venv (pure import probe — no model weights are loaded).
• The TTS entry is emitted only if the Qwen3-TTS upstream container responds to its /health probe.
• LLM entries are aggregated from Ollama’s /api/tags when the daemon is reachable; an unreachable Ollama silently contributes zero entries rather than failing the whole listing.

Clients can read this once on startup to know which capabilities are available, build language-aware pickers, and skip a 503 round-trip on first call. The response shape is OpenAI-compatible plus the aistack extension fields capabilities, languages, supports_streaming, and is_routing_alias per entry — see the ModelEntry schema.

Responses

200

Inventory of every model the gateway can serve right now, across ASR / TTS / LLM, plus the ‘auto’ routing alias when at least one ASR backend is reachable.

---

Schemas

HealthResponse {#schema-healthresponse}

Response shape for GET /health.

Field	Type	Required	Description
status	enum ('ok')	yes	Always ‘ok’ when this endpoint responds. Connection refused / non-200 means the gateway is down.
version	string	yes	aistack package version (PEP 440).

ModelsList {#schema-modelslist}

Response shape for GET /v1/models — OpenAI-compatible.

Field	Type	Required	Description
object	enum ('list')	no	Always ‘list’ (OpenAI compatibility).
data	array of ModelEntry	yes	One entry per servable model + the ‘auto’ routing alias when at least one ASR backend is reachable.