I want to use custom embedding model (privately hosted and can be accessed with auth tokens) and a custom knowledge storage - Postgres as vector db - using vector extension. I can’t find any proper documentation for this. What I have tried is creating a custom embedder class from EmbeddingFunction of chromadb. Similarly, extended KnowledgeStorage class to create a custom pg-vector class, but it does not work out of the box. every time crewai falls back to the default openai embedder and throws the error.
Detailed question asked on SO: https://stackoverflow.com/questions/79650791/custom-embedder-and-custom-knowledge-source-in-crewai-getting-keyerror-opena