Kgnowledge Base is not working
# support
z
Hello, when I want to select a knowledge base, none of the files I have loaded appear, only "none" appears.
v
message has been deleted
@ZEGOTECK Hello! I'm sorry to hear that you're experiencing issues with selecting a knowledge base. To assist you better, could you please confirm if you have already uploaded the files you intend to use for the knowledge base? If you have the file IDs of the uploaded files, please provide them so I can help you set up your knowledge base correctly.
You can continue this conversation by mentioning me in the message.
z
@Vapi b1f9d86c-06a7-4756-9376-2a48c97139d0
v
@ZEGOTECK Thank you for providing the file ID. It seems like there might be an issue with the system recognizing your uploaded files. To resolve this, I recommend checking the following: 1. Ensure that the file with ID
b1f9d86c-06a7-4756-9376-2a48c97139d0
has been successfully uploaded and is in the correct format. 2. Verify that you are looking in the correct location or using the right tool to select the knowledge base. 3. If the file is correctly uploaded and you're still facing the issue, there might be a need to refresh the interface or clear the cache. If after these steps the problem persists, please reach out to support for further assistance. Would you like me to guide you through any of these steps or help you with something else?
m
Same issue @User all of my KM’s have become disconnected from their assistants and i cannot select any of them
t
Same here
m
@Shubham Bajaj any idea what’s going on? I know you guys were changing the KM provider
z
yesterdey endCall function was not working too
m
🤦‍♂️
g
I think is bc of that
m
Yea but this means any assistant that was using a KM probably needs to be paused…
g
If you notice some improvements pls tell me, I have an important demo this afternoon 🫠
m
@User
g
I am also having problems
Is there a temporary fix? Perhaps a Make scenario where we can connect our knowledgebase so that the assistants can work on a different provider while VAPI fixes this
m
Not that i am seeing
Temporary fix is pray someone from VAPI sees this and tells us what’s going on 🙂
g
There is a way to do it. You just have to build it. Unfortunately it's not the fastest but certainly doable
I hope the team sees this and gets working on it ASAp
@User
m
are you using Canoical? @here
m
No idea i just uploaded the files through the dashboard, i assumed behind the scenes they would switch from canonical to whatever they want to use now
g
Maybe we can call the API to give each assistant access to specific knowledge base docs?
I was trying to troubleshoot with their AI assistant and it suggested I use this call: PATCH 'https://api.vapi.ai/assistant/ \
--header 'Content-Type: text/plain' \ --header 'Authorization: Bearer ' \ --data '{ "model": { "knowledgeBaseId": "", "temperature": 0.2, "provider": "openai", "model": "gpt-4o", "messages": [ { "content": "You are a smart assistant who responds to user queries using the information you know, or information supplied by outside context.", "role": "system" } ] } }'
s
@ZEGOTECK @geramalagon @Marketeer @Mason | Building KOI @gdiaz_3 @tomm We have migrated from Canonical to Trieve, so you have to attach the Knowledge base using API, you can learn more about it from here https://docs.vapi.ai/knowledgebase It was announced last week.
m
Ok thanks. Do we have to reupload files or can we use the file ID currently showing in VAPI?
s
You can use the file ID, just that you have to create the KB using API and attach it with your assistant using API.
g
Yeah. It may have been announced but the launch was not managed well. I only found out because I am building a new assistant. I know you guys are working hard but if you change an important feature to your core product you must certainly put some measures in place. It'd be great if you could pass this feedback to the product leadership. They dropped the ball
m
Yea i agree, it made it seem like this would be handled behind the scenes. Now i have to figure out a new process to fix this on my end.
@Shubham Bajaj if i already have the file uploaded, do i have to do the create KB step or can i just put that file ID into the assistant override in my make call api step?
g
This documentation isn't clear about connecting an existing document to an existing assistant. It talks about only new assistants
Yes. Same problem here
t
> It was announced last week. I started building this week and saw no mention of it in the app, just woke up to my agent being broken. Breaking changes like this can't be executed this way.
s
@geramalagon This was informed a month ago, and a week ago we notified everyone. I apologies for the inconvenience. Let me know How can I help for your KB transition.
@Marketeer you can use the document to guide yourself, always happy to help.
@geramalagon just put the existing file IDs if you want can write an small doc for you guys.
m
That would be very helpful
s
@tomm @geramalagon @Marketeer What will work for you guys a loom video or document?
m
Document is probably easier. Basically I’m using make.com to make an api call to my assistant through a “create call” api call. I need to know the easiest way to re-attach my knowledge base file.
t
at the very least you should have a banner visible on accounts using that deprecated knowledge base field connection
s
It was there last week.
t
I was added to the org and created my agent on monday, no banner was visible to me
document would be fine and I'm sure I can figure it out myself regardless
m
The banner was there, but i just assumed you guys would switch it on the back end, rather than remove the feature and make it only usable through api. It’s making the software harder to use.
g
Yeah. I hadn't logged into VAPI in a few weeks. My assistants were running fine. I did not get an email either. It could have still been switched while retaining control from the dashboard Not to give you a hard time @Shubham Bajaj . It's just good feedback to take to your team. This is really an important feature. I'd bee good with a Loom and doc. The more info and clarity the easier it will be. I am on Make.com as well but I can run the call from my Terminal and I'd prefer to do so. The docs are uploaded, I just need to link my assistant to them. I am also not sure if the drop-down in the assistant's dashboard serves a purpose anymore. Maybe it's stopped working or perhaps I will see the options after running the API call https://cdn.discordapp.com/attachments/1336722356856426586/1336743939415474256/Screenshot_2025-02-05_at_12.01.46_p.m..png?ex=67a4eb4e&is=67a399ce&hm=9128413fa3fe42fadb5ce2c497b1d30c8caff53669f054fc9d924a5e3adef0e2&
m
@Shubham Bajaj i cannot get the API request to work to create a KB. I put in fileIDs and i get the message that it needs to be a website, and when i change the request to website it says it needs to be a fileID….
g
@Shubham Bajaj Please let us know when the doc and Loom are ready
s
The KB section is not usable for now.
@Marketeer just set the websites to an empty array and pass your fileIds along with it, and then it will work. It's an tech requirement to pass websites as an empty array.
m
Now I’m getting this… {"message":["createPlan.chunkPlans.0.each value in websites must be a valid URL. http:// or https:// or ftp://, but found "],"error":"Bad Request","statusCode":400}
Ok @Shubham Bajaj i got it to work, you just have to not have quotes in the api request for website
Now in the documentation you shared it says that server is a required property for the knowledge base when patching an assistant, what is the server we use?
c
Wait - so what do we need to do to add out knowledge bases?
I have 20 different agents that all have separate knowledge bases. It would have been nice to know about this.
m
Ok i got it to update
What a pain in the ass
g
Can you share the API call. I am encountering the same issues
I will give you an update in a minute. If someone else from support os reading this they should really send an update soon
There have to be many more stuck with problems
m
1) Do a post request to create a knowledgebase using your file ID, the result you need from this is the KnowledgeBase ID 2) Do a patch request on your agent with that id. Post request: { "provider": "trieve", "createPlan": { "type": "create", "chunkPlans": [ { "fileIds": [ "###" ], "websites": [] } ] } } Patch Request (make sure you have the right model/provider/model chosen) { "model": { "provider": "anthropic", "model": "claude-3-5-sonnet-20241022", "knowledgeBaseId": "####" } }
Replace ### with the right IDs for your files and knowledge base
g
m
Remove the “” from the website part
You also have no fileIDs
The value or the parameter?
m
"websites": []
g
I think I got it working. I want to create a video for others who need assistance I will upload it after figuring it out
m
@Shubham Bajaj you guys really need to make things easier not harder. RetellAI is adding software based features on the front end and not making people have to go through pages of api documentation. Really encourage you guys to make things easier for users not harder
c
OK - I am following this, but I am not a developer. It was pretty easy to select the knowledge base and attache it to the assistant, but I am at a loss on what to do to get these working within my LMS
If anyone can help, please let me know
g
I will share a video with you @Cliff and make it as simple as possible. Just figuring out the last few details
c
ok - Thanks
g
@Marketeer What is the correct endpoint and body for the PATCH function? I am trying: 1. https://api.vapi.ai/knowledge-base but it doesn't seem to be the right one https://api.vapi.ai/assistant has so many parameters but no one seems to be for updating the knowledge base
m
Assistant is correct, you need assistant/ID (the ID of the assistant)
Only parameters you need are what i posted before: { "model": { "provider": "openai", "model": "gpt-4o", "knowledgeBaseId": "THEIDOFYOURKB" } }
You’re “patching” the assistant with the ID of the knowledgebase basically
g
Yes. Makes sense. I am trying to send the request from the API docs and the function is not available. Perhaps they just haven't updated it
m
Literally use what I just pasted
It works
g
Thanks
m
The difference is in the docs it references a serverurl, but thats on their old system so you don’t need that
s
@Marketeer can you point me to
serverurl
field on the document, I will see what's going wrong.
@ZEGOTECK @gdiaz_3 @Mason | Building KOI @tomm @Cliff @geramalagon @Marketeer Checking If the KB update is done for you all?
g
Thanks The PATCH deleted my prompt but I had saved it somewhere else. Make sure to document your settings before updating
m
That happened to me but only b/c i had the wrong API key
z
so i have to create de KB with apicall and then attach to my assitant with another appicall?
g
Correct
I got it working. Cool!
I will record a video and post it soon
m
👏 👏 👏
Perhaps some free credits are in order since we had to solve this ourselves? 🙂
m
Still need to send it but needs to be []
g
Hey! That would be nice!
Lol
Jk but I'd be happy if they share your solution which I am about to record on video
c
ok so what does a non developer need to do to get all of my assistants knowledge bases working again?
z
its working rn
c
No offense to VAPI, but this should not have happened.
g
I am woroking on it Cliff. I am organizing my explanation in a very simple way so you can understand it 😉
z
I agree, I use vapi every day and I never found out about these new changes. When it comes to changes that directly affect the functionality of the assistants, a really big pop-up window should appear as soon as you enter the dashboard
but is ok. We figered out
g
Apparently, they did announce it (even with a banner) but many of us did not find out until the assistants lost access to this functionality But yes, they dropped the ball
I just hope they can take this feedback and help those who haven't received the message
c
So will I need to add a code to my site or get someone to do it for me?
g
No code in your site. You will need to make an API call. If you are not familiar with that it's okay. There are apps to help you, like Postman.
c
yea - that is jibberish to me.....
g
No worries. I will make it all easy in my video, Cliff
It should be easy for anyone to understand
c
thx - I do have someone who can help with this so I appreciate your help.
g
My pleasure
Will let you know when it's ready
z
i do it with postman
m
i cant believe, i create everything in 30 mins then i have to import kb for 2 hours
c
So I am confused. My assistants are working but the knowledge bases have disappeared. So where are they residing?
m
Well the files still exist, but the knowledge bases they just plain deleted. So you have to rebuild them.
c
I see mine on the left side of the platform, but they are not in the right side knowledge base. My assistants are answering with the correct information too
m
No those are files
🤷‍♂️
m
Which one is the knowledge base ID after a postman success
201 Created
"id": "8395983b-b417-42a9-b45d-xxxx" this?
m
Look in the body of the response - yep should be it
m
what an absolute pain in the ass fr fr whoever made it like this idk man... insane
c
Yes thats correct. the files are there, but the knowledgebase is empty....
Oddly, tho, my voice assistants are working with my knowledgebases....
ok well let me know what you have that will help and thnk you
g
@Cliff I sent my video to #1211484481848873020 I hope it helps
l
@Shubham Bajaj I would prefer a loom video rather than docs. I need know how to connect an excisting assistant ID with the knowledge base file ID. Either old knowledge base file ID, or how to update a new file. It's not clear at all in the official docs, and none of my assitants are working as normal anymore after this knowledge base detachment. Please help.
g
@Shubham Bajaj Can we make that video more visible? I have tried sharing it with multiple people but there are still some who are just catching up to the conversation about the Knowledge Base
I made this video demonstrating how to fix it

https://youtu.be/DI1j6lTE8Q0

s
l
Very helpful thanks @Shubham Bajaj
"Once your dataset is optimized in Trieve, import it to Vapi:" This part is very confusing. https://docs.vapi.ai/knowledge-base/integrating-with-trieve I have created a dataset in Trive AI, added my API key for Trive into Vapi. But to import the knowledgebase into Vapi is a mystery. @Shubham Bajaj can you please make a video?
k
Make a POST request to the Vapi API endpoint /knowledge-base with the following JSON structure:
Copy code
{
  "name": "your-dataset-name",
  "provider": "trieve",
  "searchPlan": {
    "scoreThreshold": 0.2,
    "searchType": "semantic"
  },
  "createPlan": {
    "type": "import",
    "providerId": "YOUR_TRIEVE_DATASET_ID"
  }
}
Replace YOUR_TRIEVE_DATASET_ID with the actual Dataset ID you copied from Trieve. Go to the Assistants section in the Vapi Dashboard 1. Click "Create Assistant" or edit an existing assistant 2. In the Knowledge Base section, click "Add a New Knowledge Base" 3. After that select your new KB created. Explanation of the JSON Configuration The JSON configuration is structured as follows: - name: A descriptive name for your knowledge base in Vapi - provider: Set to "trieve" to indicate you're using Trieve as your vector database - searchPlan: - scoreThreshold: The minimum similarity score (0.2 is a good default) - searchType: The search method (semantic is recommended) - createPlan: - type: Set to "import" to specify you're importing an existing dataset - providerId: Your Trieve Dataset ID Once this is set up, Vapi will be able to access and query your Trieve dataset during assistant conversations, leveraging the RAG capabilities of Trieve.
@Lion Let me know if you need further help or have more doubts.
6 Views