Replicate - BoostGPT Documentation

Any model hosted on Replicate, called by id. The same escape hatch as fal.ai, over a different catalogue.

Connecting

In the dashboard, open Connectors, find Replicate and press Connect. The key you enter is checked against Replicate before anything is saved, so a key that is refused never becomes a connection that fails later.

What you will need

Field	What it is
API Token	Your Replicate API token (starts with r8_)

Getting the credential

Create a token at replicate.com/account/api-tokens — it starts with r8_
Paste it into the connector

Notes

Asynchronous. run_model returns a prediction id. Poll get_prediction until the status is succeeded.
Finished media is copied onto BoostGPT storage, so the link keeps working.
Different models take different inputs. input.prompt and input.image cover most image and video models; a model needing something more unusual may not be reachable through these parameters.

What an agent can do with it

Tool	What it does
`run_model`	Start a model on Replicate. Pass the model’s owner and name — for example owner ‘black-forest-labs’ and name ‘flux-schnell’. Model inputs go in the `input` fields. Asynchronous: returns a prediction id to poll with get_prediction.
`get_prediction`	Poll a Replicate prediction. Status is ‘starting’, ‘processing’, ‘succeeded’, ‘failed’ or ‘canceled’. Once it has succeeded the output is rehosted on a durable URL.
`cancel_prediction`	Cancel a Replicate prediction that is still running.

Useful resources

Replicate HTTP API