Meta models

Every call to a Meta model goes through one client against the Model API at https://api.meta.ai/v1. Model IDs are configuration, never literals in code (see Configuration).

Model	ID (default)	Used for
Muse Spark	muse-spark-1.3	label extraction with structured output, listing text read into fields, price guidance with search grounding, the decision cards and the uses of an item, listing drafts, search plans, project proposals through a tool-calling loop
SAM 3.1	sam-3.1	the item, the label and the connector found in a photo, one concept per call; the label crop feeds extraction
Muse Voice Transcribe	muse-voice-transcribe-1.0	voice notes kept as the owner's words, from the web and from the phone
Muse Image	muse-image-1.0	a picture of what an item could become or of a finished project, drawn only when the owner clicks and the switch is on; illustrative, never evidence

Considered and not used in this version, each with its own page saying so: Muse Wonder, Muse Glimmer, Muse Spark Open; there is one provider (Providers and routing).

Each model page covers its job in the app, the exact request shape, the settings that affect it, cost notes and failure modes. The rules decide and the model proposes, everywhere: a verdict, a price, a listing line or a project allocation is kept only where the evidence backs it.

Fixture mode

With OPL_FAKE_API=1 the client serves recorded responses from backend/tests/fixtures/ and spends nothing. Tests always run this way.