Backends

AmosBackend options:

Option	Description
baseUrl	Default https://api.amos.scira.ai. Point it at your own amos.serve to self-host.
model	Default amos-1-large. Always sent: the hosted API otherwise answers with amos-1-mini.
retry	Retries rate limits, server errors, timeouts and network failures, up to 4 attempts. false turns them off.
timeout	Milliseconds to wait for each request. Default 60 seconds. false waits indefinitely.
maxRequestBytes	Default 4,000,000. The hosted API drops larger requests, so they’re refused before sending.
userAgent	Sent as the User-Agent header.

For tests, or another model, pass any object with decide({ context, image, questions }) that returns { answers }. See TaggingBackend for the shapes.