Hello,
I have been building a chat bot. Basically, it sends user queries to GPT APIs (e.g., gpt-4o), and displays the response to users.
At the moment, I have not yet used Assistants API.
Now, we would like to add some knowledge of our own tools to this chat bot, such that when users ask questions about our tools, the chat bot could give some answers.
Does anyone know how to do this?
Thank you
Tie
Hello!
One easy way to do this would be to simply add some knowledge to the system prompt, describing the tools to the chatbot so that it can elaborate on them if asked by the user. This might use more tokens but it’s the quick and easy way to do it.
If the chatbot has a LOT of tools and explaining to do, a better way would be through the Assistants API, where you can upload entire files and have the assistant retrieve text from it dynamically, then rephrase it to the user depending on their needs. The Assistant API is much slower than the Chat Completions API though, so that’s another thing to keep in mind.
Hello I am doing this and im an expert in the field, I can dm you!
Thank you for your help.
The knowledge is about 2000 - 5000 words, so it’s not appropriate to add it to the system prompt.
Utilize context retrieval capabilities of assistants like file search (simple plug and play) . If you would like to bring your own retrievers and chunking strategies you may use a tool which gets called each time the assistant invokes based on context. You have to be more specific in your tool description and have something like ‘You are not pre trained to answer questions regarding _____ ; use ___ tool to retrieve related context.’ in your system prompt. Adjust top-p and temp values based on your use case. Hope this helps - Cheers!
This is exactly what i want. But the problem is HOW?