AI Copilots - Troubleshooting

AI models are often unreliable and tricky to debug. Here are some common issues you may encounter, and how to troubleshoot them.

Common issues

Here’s a list of common issues you may encounter, and how to fix them.

The AI provider is currently overloaded or unavailable

This means that the AI provider you’re using is currently not available. You can check the status of your provider by visiting their status pages: OpenAI, Anthropic, Google. To avoid this issue, you can set up fallback models that use other providers.

The AI provider's rate limit or quota exceeded

This means that you’ve hit token limits defined by your AI provider (e.g. OpenAI, Anthropic, Google). You can reach out to your provider to increase your limits, use a model with lower limits, or consider setting up fallback models for busy times.

For this AI provider, 'additionalProperties: false' is required to be set in the properties

This means that your provider requires the additionalProperties field to be set to false in tool parameters. Fix this by adding the following line to your tool definition:

<RegisterAiTool  name="weather-tool"  tool={defineAiTool()({    description: "Get the weather for a location",    parameters: {      type: "object",      properties: {        location: { type: "string" },      },      required: ["location"],      additionalProperties: false,    },    // ...  })}/>

A model is ignoring my back-end knowledge

If an AI model doesn’t seem to be using your back-end knowledge correctly, there are a few things you can check:

Check you’re using the correct copilot ID in all APIs

Check you’re using the the copilot ID throughout your application, specifically in every use of AiChat and useSendAiMessage.

import { AiChat } from "@liveblocks/react-ui";
function Chat() {  return (    <AiChat      chatId="my-chat-id"      copilotId="co_tUYtNctLAtUIAAIZBc1Zk"    />  );}

import { useSendAiMessage } from "@liveblocks/react";
function SendMessage() {  const sendAiMessage = useSendAiMessage("my-chat-id", {    copilotId: "co_tUYtNctLAtUIAAIZBc1Zk",  });
  return (    <button onClick={() => sendAiMessage("What's new?")}>What's new?</button>  );}

Make sure you’re letting AI know when to use knowledge

Make sure you’re defining when AI should use knowledge in your copilot settings. When defining copilots from the server, this is the knowledgePrompt field. If it doesn’t seem to be working, try experimenting with your prompt to be more specific, authoritative, and use examples.

- When the user asks a question about code.+ It's required to do this whenever the user mentions React code, e.g. `useOthers`.

It’s worth checking if there’s an official guide for your specific model, which details how to talk to it. For example the official prompting guide for GPT-4.1.

The AI provider response exceeded the maximum token limit

This means that the response from the AI provider exceeded the maximum token limit.

This error is usually caused:

• When the model’s generated output would exceed the max tokens you set in the dashboard for a copilot.
• When the request would exceed the model/provider’s token limits (e.g., combined prompt + expected completion), leading the provider to return a token-limit error

Check first in your dashboard if the max tokens limit is set correctly.

Error table

Every error you may encounter is listed in the table below, along with a suggestion.