Reference

Models

Run one model directly — no agent in between. The same models, credits and rules as the Run button on each model's page.

Describe a model

GET/api/v1/models/{id}

The id is the last part of the model page's URL: https://corollary-labs.com/models/esmfold2 is esmfold2. The answer lists the inputs the model takes — name, type, whether it is required — and the outputs it returns, which is enough to build a form.

bash

curl https://corollary-labs.com/api/v1/models/esmfold2 \
  -H "Authorization: Bearer $COROLLARY_API_KEY"

Run it

POST/api/v1/models/{id}/runs

Send the inputs as an object. Fast models answer with 200 and their outputs in the same response. Models that need a GPU started for them answer 202 with a poll_url; omit mode and the platform chooses.

bash

curl https://corollary-labs.com/api/v1/models/esmfold2/runs \
  -H "Authorization: Bearer $COROLLARY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "inputs": { "sequence": "MKTAYIAKQRQISFVKSHFSRQLEERLGLIEVQ" } }'

inputsobject: The model's input fields, by name — see Describe a model.
mode"sync" | "async": Optional. Force waiting for the result, or force a background job.

The model run object

idstring: The run, e.g. job_x7k2p9q4.
statusstring: running, succeeded, failed or cancelled.
outputsobject | null: The model's outputs, keyed by the output names the model describes.
errorstring | null: Why a failed run failed.
creditsinteger | null: What the run cost, when it answered inline. Long runs settle on your balance when they finish.
poll_urlstring | null: Where to read a running job.
page_urlstring: The model's page — where a person can run it too.

Poll a long run

GET/api/v1/model-runs/{id}

Read it every few seconds until the status is not running. Reading is also what collects the result, so polling promptly brings it back promptly.

NextRunsCreate, read, stream, answer and cancel a research run.