I wanted to inquire about your plans for VAPI and Trieve integration. We are currently using Trieve to power the knowledge base for our assistants. Trieve is being sunset on November 1st. Do you have any plans to maintain/host Trieve yourself as it's licensed as MIT now or are there any viable alternatives that deliver results as good as Trieve does? We tried uploading our knowledge base to the builtin

query

tool, both attaching the files to the assistant directly and creating a dedicated tool with the knowledge base. Our knowledge base consists of ~140 Markdown files. The latency was very bad - ~10 seconds between the time the question ends and the assistant responds. In some cases the assistant hangs and does not provide the answer. In comparison with Trieve we were getting latencies as good as if the answers were in the main system prompt. The responses also were more accurate and sounded more natural. Can you help us find the best way forward?

Hi Pawel, With Trieve being sunset, you have a couple of options for integrating your knowledge base with VAPI. Since Trieve is now open source under the MIT license, you could consider hosting it yourself if you want to maintain continuity. Alternatively, VAPI provides robust built-in tools that you can use instead. 1. **Host Trieve Locally**: You can set up your own instance of Trieve using the MIT-licensed code. This option will require technical resources to manage and maintain the system. 2. **Use VAPI’s Built-in Tools**: Consider using VAPI's

query

tool to integrate your datasets directly without relying on Trieve. Refer to these [guidelines](https://docs.vapi.ai/knowledge-base/integrating-with-trieve) for setting up knowledge bases, and ensure that your knowledge base is optimized in terms of chunk sizes and metadata for better performance. 3. **Optimize Integration**: If you encounter latency issues, ensure your chunk sizes are between [200-1000](tel:200-1000) tokens and consider enabling features like semantic boosting and query caching (if supported) to minimize response times. For more detailed setup and best practices, review our documentation on connecting datasets to VAPI. If you decide on a different approach and need help with the setup, our support documentation can provide additional guidance. Let me know if there are other questions or specific areas where you need help.

Vapi's built-in tool is only good with static files. Cheap chatbot have been using live urls as KBs for months, sometimes more than a year now. For the trieve solution, asking our clients to buy a server for their KBs makes no sense when Vapi's competition are offering live/url KBs built-in. Kings Big, I thank you for your band-aids, but Vapi needs to fix this weeks' issues soon.

We will discuss with our team to see what is on the roadmap as far as alternative solutions for KBs