API reference

Models

Discover the enabled catalogue instead of relying on a hard-coded model list.

List models

GET /v1/models requires an active gateway key. It returns all enabled catalogue rows, not a separately filtered per-key capability list. It does not consume inference tokens.

curl https://api.llmflux.dev/v1/models \
  -H "Authorization: Bearer $LLMFLUX_API_KEY"

Response fields

Illustrative shape only; these are not advertised production model IDs.

{
  "object": "list",
  "data": [
    {
      "id": "example-public-alias",
      "object": "model",
      "owned_by": "example-provider",
      "family": "other",
      "input": [
        "text",
        "vision"
      ],
      "output": [
        "text"
      ],
      "root": "example-provider/example-model"
    }
  ]
}

id: public alias when configured, otherwise canonical upstream ID.
root: canonical upstream ID, included only when an alias is configured.
owned_by and family: catalogue presentation metadata.
input and output: descriptive modalities. Some catalogue metadata is inferred from model names during synchronization; verify actual endpoint support with the provider-backed request.

Model resolution

The gateway checks these rules in order, against enabled models only:

An exact canonical ID.
An exact administrator-configured display alias.
An ID with only its first provider segment removed. For example, provider/team/model may be requested as team/model, not automatically as model.
For auto, an enabled canonical ID ending in /auto, otherwise the first enabled catalogue row (ordered by provider, family, ID).

No implicit renaming. Dots are not converted to dashes. Copy IDs exactly. Unknown or disabled models return 404 invalid_model. If several stripped IDs collide, the first enabled match wins; prefer a full ID or unique alias.

Choosing a model

The enabled catalogue can change as administrators synchronize providers or enable, disable and rename entries. Pin an exact ID for reproducible selection. auto is a catalogue fallback, not gateway-level prompt analysis or a guarantee of the strongest model. Any intelligence of an upstream /auto model belongs to that provider.

See supported models for practical selection, or chat completions for your first request.