Meta (p.chatoyant.0.13.0.doc.chatoyant.provider.Chatoyant

Module Chatoyant_provider.MetaSource

Meta Model API (Muse Spark) adapter.

Rides the OpenAI-compatible /v1/responses protocol at https://api.meta.ai/v1 and reuses the OpenAI serializers and client. Encodes the Muse Spark wire rules: reasoning cannot be disabled (reasoning_effort "none"/"off" is omitted instead of sent), only the "auto" tool_choice is forwarded, stop sequences are dropped, sampling parameters pass through, and web_search becomes the built-in server-side tool. For raw endpoint access beyond generation, use Openai.Make_client with default_base_url.

Sourceval default_base_url : string

https://api.meta.ai/v1

Sourceval models : string list

Hosted Muse Spark model ids, current generation first. The -contributor variant is the discounted tier that permits Meta to train on request data — priced and recognized, but never a preset default.

Source

val responses_request : 
  Provider.message list ->
  Provider.options ->
  Openai.responses_request

Build the Responses-API request for one generation turn, applying the Muse Spark wire rules above.

Source

module Make_provider
  (Http : Chatoyant_runtime.Effect.HTTP)
  (Config : sig ... end) : 
  sig ... end