Hey Team Got a client that has a CSV file that is dynamic and updated weekly, we need to get the prices off this. I was thinking of using Make, but then I thought, could the Agent read the URL link if it was in the knowledge base? So, the agent references the URL, which is the CSV file in my drive? Hope that makes sense Thanks

Has anyone figured out a way to fully stop audible narration? I'm still getting completely random narration of internal processes like "Storing the user's full name as..." or "Saving address as...", etc. It's not all at once, and it's not always the same thing. The issue is that it changes depending on how the conversation is going. So, sometimes it will narrate a tool and nothing else. Other times it will narrate an internal process and not the tool calls. Sometimes I'll test a call both on Voice Lab + using the phone number and it will work smoothly. I'll leave it for a few days and the narrations come back. Is this due to updates? I can't pinpoint it and I've tried a ton of prompt variations. Things I've tried so far: - Added [brackets] around internal tasks like tool calls - Explicitly stated in the ## Style Guardrails to: - Never refer to the caller as user or narrate (e.g. user is reporting an issue with...) - Never announce the use of any tool (e.g. storing the full Name as, Storing Phone as) - Never announce any internal process (e.g. starting to gather details; starting to assist) - Upgraded from GPT Mini -> GPT 4o -> GPT 4.1 Not really sure what else to do. I know others have mentioned this issue so I'm curious if there are any solid solutions in place.

NO, not that kind of hook up! haha I ended up helping 5 people getting them fully optimized prompts that work for 4o and have Assistable custom tools integrated in the prompt also. Tailored for chat/voice, whatever they were running. Feedback has been great! My question now: what else do you guys need help with? I'm in the giving mood today...so let me know and let's see if I can help you!

Let me ask you something to frame how this works: If I gave you a 1-minute timed 1-question test and allowed you to look through a textbook for the answer; would you be able to do it? No, it would take a whole day to go through the textbook. It would be slow and your brain would be scattered by the time you found it that you would forget the question. What about a 1-minute timed 1-question test with Google? Obviously, a Google search takes 1 second. THIS IS HOW KNOWLEDGE BASED WORKS - rather than looking through a giant text book for one answer that slows you down, you can get the exact answer immediately. _____________________________________________________________________________________________________ I get messages sometimes about voice AI hallucinations or not following instructions or not giving correct information - when I go to audit the assistant there is 20k characters of prompt. Most of it is FAQ, objection handling, company info etc. Save your prompt for instructions and personality - leave the nuance to kb. this is how you build reliable AI with personality. We have deployed a lot of production-grade enterprise AI with 6-8k prompts that works without skipping a beat - the secret? Knowledge base. This video below shows you A-Z how to build a proper knowledge base in voice / chat AI in Assistable with a real company (no pun intended) example: Van Staal. From scratch with no prompt, we have a voice AI answer questions exactly with 0 latency. You can opt for the cheaper option like GHL does with a tool-called knowledge base, but you will have latency and it will not be perfect because the AI has to call a tool every time to get information where our native knowledge base runs every single answer through the knowledge base. We have one of the most over-engineered knowledge base systems of all time, its at your finger tips ready to be used. ______________________________________________________________________________________________________

Hey Team I am creating a voice agent for a group of building managers. When the tenant calls, they will mention their building name. The building manager & email are in the knowledge base. I want the agent to extract that data so that we can automatically email them the transcript. I feel like I am so close but one step away. I have custom fields and tried using auto extraction, but it isn't working or will only work for the building name. I think this is because the tenant mentions the name. Can this be done, or is there a workaround? Thanks