Any model can name another model as its backup. The gateway tries the
backup when the model is offline, rate limited, or its upstream fails
before the client has received anything. The response reports the
requested model, so the client sees no switch. The request log records
each failed attempt and the address that served the request, and the
cost uses the backup's prices.
InfluxDB 3 creates a column on its first write. A dashboard query that
names ttft_ms or error failed until a streamed or failed request had
been exported. Write 0 and empty strings instead of omitting the field.
Add an optional exporter that writes one point per proxied request to
the InfluxDB 3 write endpoint. Points are batched in memory and flushed
by size or interval. The exporter is off until GATEWAY_INFLUX_URL is set.
The sync took meta.n_ctx_train from the model listing, which is the
training context, not the -c the server runs with. The llama.cpp provider
now asks the server for the live value: /props on a single server, the
router's /models status and /props?model=, and /upstream/<model>/props for
models llama-swap has loaded. Stopped models are never probed, because
that would start them.
Admins can also pin the context window per model from the Models page.
The override wins over the discovered value and survives syncs.
Describe the request path, the provider capability interfaces, the
catalog, auth, and logging for contributors. Reword the deploy section so
the author's registry and cluster read as an example.
Translate OpenAI image requests to OpenRouter chat completions with image
output and back, as the LiteLLM proxy does. Only models whose provider
reports image output accept the call; the catalog sync now stores the
reported output modalities in the capabilities column.
Liveness and readiness timeouts go from 1s to 5s, and a startup probe
covers OIDC discovery. Liveness only hits /healthz, so a database or
network blip no longer restarts the pod.
