Chat keeps failing and can't reconnect

Bug Reports
1

I’m getting failures after a while of using chat. This has been happening all day, and every time it happens I lose the chat context, which is a pain.

Here’s the most recent error message:

Connection failed. If the problem persists, please check your internet connection or VPN
No response from model [unknown]
(Request ID: e0d44542-43c8-4438-abe6-ae078657e9ee)

Clicking “try again” just re-displays the message after a tiny pause. If try again would work, this wouldn’t be an issue, but it seems to be unable to form a request that will succeed once it’s failed once.

2

Adding: This happens almost every time I reach the 25 request tool limit and click to continue.

3

What happens when you start a new chat? As this bug has been around for 5+ months and doesn’t seem to be going away anytime soon.

4

It’s often a rate limit imposed by the provider, you can check the debug part (seen in with Anthropic models a lot)

Switching to another model or waiting 10min “fixes” it

5

Hey, sorry about this! Will investigate your connection error tommorow (away from my machine), but want to figure this one out!

6

I am actually using Claude mostly, so maybe that’s the issue. I wish it would indicate what the problem is so I would know how to address it. The problem is that the agent will write something, and then fix the linting errors in what it wrote. And then fix the linting errors in that. And then undo everything and try a different approach, and then fix the linting errors in that, etc. and eventually it’s made dozens of calls, and I’ve only interacted with it a few times. So if I’m hitting the rate limits it’s because the model is ■■■■■■■■ up and correcting itself.

Edit: That is a weird word to censor.

7

Cursor handles all the rate limits and similar for you. If we think you are misusing Cursor, you’d see an explicit error for that case.

This is definitely a bug, but it’s 10pm for me right now, so I can’t check that request ID out yet!

(also, I’ll remove that word for the default censor list tommorow too)

8

It’s still failing fairly reliably every time the “25 tool calls” message appears.