Models#

The Models API is a single, consistent way to call any model on the platform and to discover what each model can do. It unifies foundation models, property predictors, structure predictors, and structure/sequence generators behind one namespace.

Discovery is driven by model metadata. Each model publishes a metadata document describing the methods it supports, so clients (including our own web app) can build request forms and model cards directly from it — there is no need to hardcode per-model behaviour.

Core concepts#

Every method is addressed by two path segments, {input}/{output}:

  • input — how inputs are enumerated and the cardinality of the call, e.g. batch (N sequences → N results; folding is a batch call too, N complexes → N structures), single-site (one base sequence → all point mutants), indel, or generate (sampling). It is an open set.

  • output — the primary output produced, e.g. embeddings, logits, attn, loglikelihood, predictions, structures, or sequences. Also an open set; models may expose their own.

A model supports a method only if it declares that (input, output) pair in its metadata. Each declared method lists its outputs (name → output type). The request params a method accepts are fetched separately, per method, from .../{input}/{output}/params (an inline JSON Schema, enough to render an input form) — this keeps model metadata small even for models with many methods.

Like the rest of the platform, calls are asynchronous: a POST returns a job handle, and outputs are fetched from the jobs service once the job completes.

The endpoints include:

  • List models — GET /api/v1/models, with filters (scope, input, output, output_type) and a verbose flag for full metadata.

  • Get model metadata — GET /api/v1/models/{model_id}, the discovery surface a client renders from.

  • Get model tokens — GET /api/v1/models/{model_id}/tokens, the input/output token vocabularies (kept out of metadata to keep it small).

  • Get a method’s params — GET /api/v1/models/{model_id}/{input}/{output}/params, the method’s request JSON Schema (fetched per method, so metadata stays lean).

  • Run a method — POST /api/v1/models/{model_id}/{input}/{output}, which returns a job handle.

Endpoints#

On this page