Models and Providers API

Provider APIs configure the model backends used inside AI Studio by workflows, knowledge bases, and agents. They do not perform OpenAI-compatible inference. For application inference with a Genesis API key, use the Model API.

Provider types

The product contract separates three provider families:

Family	Root	Typical role
LLM	/newmoi/llm	Generation, reasoning, extraction, and agent chat
Embedding	/newmoi/embedding	Build vectors for knowledge and retrieval
Parser	/newmoi/parser	Convert supported documents into processable content

Each enabled family can expose backends, backend detail, endpoints, endpoint status, and router configuration. LLM and embedding providers also expose model discovery and probe operations.

Backends and endpoints

For LLM backends:

Method and route	Purpose
GET /newmoi/llm/backends	List configured backends
POST /newmoi/llm/backends	Create a backend
GET /newmoi/llm/backends/:id	Read a backend
PUT /newmoi/llm/backends/:id	Update it
DELETE /newmoi/llm/backends/:id	Delete it
POST .../:id/endpoints	Create an endpoint
GET .../:id/endpoints	List backend endpoints
PUT .../:endpoint_id/status	Change endpoint status
POST .../:id/probe-models	Probe models through that backend

Embedding and parser use the same resource pattern under their respective root paths. Fields differ by provider type: for example, parser configuration can describe supported MIME types, while model providers can expose model IDs and credential references. Use the deployed schema and never copy fields between provider kinds merely because their routes look alike.

Probe before enabling an endpoint. A successful probe verifies the submitted connection at that moment; it does not guarantee future availability, model quota, compatible vector dimensions, or permission for every workspace user.

Routers and available models

GET and PUT /newmoi/{family}/router read and update routing configuration. LLM and embedding model lists are available through /newmoi/models and /newmoi/embedding/models; probe routes are also available at the family or backend level.

Router strategies and their fields are deployment-defined. Update a router only with backend and endpoint IDs returned by the same workspace. After an update, read the router again and perform a representative probe before moving production traffic.

Initial setup and secrets

POST /newmoi/llm/setup provides the product setup operation for an LLM backend; it can orchestrate more than a single low-level create. Prefer it when the deployment documents the setup flow.

Provider API keys and credentials are secrets. Send them only in fields explicitly marked for credentials, do not return them to browsers, and do not expect list/detail responses to reveal them. Rotating a provider key can affect agents, workflows, and knowledge indexing at once; inventory those consumers and retain rollback information before changing it.

Deleting a backend or disabling its last healthy endpoint can interrupt active work. Check endpoint health and downstream bindings, then confirm the impact. For failures inside a workflow, correlate the provider probe with the workflow execution ID rather than repeatedly rerunning the job.