Problem
The LLM Proxy model router exposes an OpenAI-compatible /embeddings endpoint, but embedding requests currently route through the OpenAI embeddings adapter.
Knowledge Base embedding code already has provider-specific embedding clients for some non-OpenAI providers, but that capability is not available through the general LLM Proxy/model-router embeddings endpoint.
Definition of Done
Add embeddings support through LLM Proxy/model-router for non-OpenAI providers where the provider offers compatible embedding APIs.
- Demo video showing:
- Embeddings requests working, for each provider added, going through the "provider specific proxy handlers" (ie. non model-router endpoints)
- Embeddings requests working, for each provider added, going through model-router (ie. openai transformed request/response schemas)
More notes
- Route
POST /v1/model-router/{llm-proxy-id}/embeddings by provider-qualified embedding model IDs, not only OpenAI models.
- Add provider-specific request/response adapters where needed while preserving the OpenAI-compatible external response shape.
- Include available embedding models from supported providers in
GET /v1/model-router/{llm-proxy-id}/models when they can be used by /embeddings.
- Document which providers are supported once implemented.
Problem
The LLM Proxy model router exposes an OpenAI-compatible
/embeddingsendpoint, but embedding requests currently route through the OpenAI embeddings adapter.Knowledge Base embedding code already has provider-specific embedding clients for some non-OpenAI providers, but that capability is not available through the general LLM Proxy/model-router embeddings endpoint.
Definition of Done
Add embeddings support through LLM Proxy/model-router for non-OpenAI providers where the provider offers compatible embedding APIs.
More notes
POST /v1/model-router/{llm-proxy-id}/embeddingsby provider-qualified embedding model IDs, not only OpenAI models.GET /v1/model-router/{llm-proxy-id}/modelswhen they can be used by/embeddings.