Today Hudini chat and Semantic Search share one provider connection. A separate embeddings section only appears for providers that can't create embeddings themselves. So once the shared connection is used for Semantic Search, the chat can't run on a different model or provider.
Request:
Two independent connections, always configurable: one for Hudini chat and one for Semantic Search, each with its own endpoint, key and model.
Support for self-hosted, OpenAI-compatible embeddings endpoints, not only OpenAI and Azure OpenAI.
A free choice of embedding model, with an automatic re-index when it changes.
Why: MSPs can keep client documentation inside their own infrastructure (GDPR), avoid per-token costs for indexing, and choose the best model for each task.
Related: "Support for local LLM with Hudini" and "Hudini – Support Hatz AI, or Custom Endpoints".