Extremely slow agent processing

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

cursor agent is processing actions at an incredibly slow rate, i had a single md file that took 1.5 hours to complete yesterday. today this run 2daffbc6-95e1-44e2-a7cc-17f2e9757640 took a long time as well. its been going for about 45 min… ive never had an agent take more than a few minuets, even on complex tasks since i primarlly use MD files.

Steps to Reproduce

seems all agents are running a bit slow

Operating System

Windows 10/11

Version Information

current ide

For AI issues: which model did you use?

auto

For AI issues: add Request ID with privacy disabled

2daffbc6-95e1-44e2-a7cc-17f2e9757640

Does this stop you from using Cursor

No - Cursor works, but with this issue

5

Hey, I checked the logs for this Request ID. The server worked fine, it’s the client side that’s slow. In this session, a lot of steps built up and the conversation state got huge. I can see the “Large conversation state serialize” warning firing on every step, and serializing the state before each next step keeps getting more expensive. There was also about a 31-minute idle period and a stream stall around 6s that went into retry. This looks like a network drop or a long apply on the client.

What to try:

  1. For new tasks, start a fresh agent session instead of continuing a long one. Long multi-step threads slow down a lot because the state keeps growing, especially on MD files with frequent edits.
  2. Run Cursor Settings > Network > Run Diagnostics so we can check for connection issues. That 31-minute gap looks suspicious.
  3. If you’re on a corporate network, VPN, or proxy, open App Settings with Ctrl+,, search for HTTP/2, and turn on Disable HTTP/2. This sometimes helps on unstable connections.

If it’s still slow in a fresh session, send a new Request ID with Privacy Mode turned off, and I’ll take a closer look.

6

we need a better way to transfer context to new chats, we used to have a way to tag a prior chat, does that happen automatically now?

7

There’s no automatic carryover between chats. Every new chat starts fresh, plus any rules you’ve set up.

To pull in a past chat, type @, pick Past Chats, then choose the one you want. Its context will be added to the new chat.

If your tasks are long, this workflow also works well:

  1. At the end of a session, ask the agent to write a short handoff what’s done, what’s left, key files and decisions into HANDOFF.md.
  2. In the new chat, reference it with @HANDOFF.md plus any files you need. It’s cheaper and more predictable than dragging in the whole history.
  3. For long-term knowledge style, architecture decisions use .cursor/rules, they’re pulled in automatically.

This also fixes the original slowdown issue. Fresh chats don’t grow a big state, so steps run at normal speed.

8

oh i love that idea with a handoff markdown

thanks Dean!