I don't want to use trieve. we have custom built for the purpose of vapi a pdf extractor that creates things perfectly as we need it. it is a major pain and extra expense to use trieve. Why is Vapi so behind and seemingly struggling to provide a proper recallable vector database knowledgebase where we don't need to worry about vapi or other intermediate tools messing up our content and making it less retrievable. I hope it is possible, but i have run in circles trying to figure it out. I guess i could run all RAG in flowise then just use vapi for voice but that seems like a lot of extra work... HELP?!

You can’t directly plug your own vector DB into Vapi’s native knowledgebase (it only supports Trieve or Google vectors). To use your own DB, run retrieval yourself: Let Vapi send the user’s query to your webhook. Your system queries your vector DB, generates an answer with your LLM. Return the final text to Vapi for voice output. So Vapi just handles voice; you handle all RAG logic externally.

Vapi no longer supports direct file-based knowledge retrieval instead, it requires you to use a Query Tool backed by your own vector store, making Trieve optional but custom RAG setups essential so your best option is to keep using your custom PDF extractor and connect it to Vapi as a function tool, letting vapi handle voice only..

thank you