I got a session which had maybe 5 Q&As, but After some time, I found that there is only 1 QA remaining, So I think the tailing Q&A are lost, even I restarted the IDE, it didn’t recover.
Hey, thanks for the detailed report with the version and the Request ID, that really helps.
From what I can see, the Q&A turns did run and were processed. The loss happened on the IDE side when saving chat history locally, so a restart didn’t bring them back. That shouldn’t happen, and it’s not anything in your setup. It’s an issue we’re tracking. Sadly, I have to be upfront, we can’t restore the lost turns in the chat panel now.
A couple quick questions so we can match your case more precisely:
During that session, did you have any network dropouts or did Cursor reload its window, like seeing a Reload window prompt or a freeze?
Have you seen this happen again, or was it a one time thing?
I’ve shared this with the team along with your Request ID. I’ll reply here if there are any updates.
Thanks for coming back with more details and a second session ID. The fact that it happened again helps us match your case more accurately.
Based on the symptom, it looks the same as before. The turns are actually processed on the server, but the tail gets lost on the IDE side when it saves chat history locally, so a restart won’t bring them back. That shouldn’t happen, and it’s not something in your setup. It’s a bug we’re tracking. Sadly, I need to be upfront that already lost turns in the chat panel can’t be recovered.
If it happens again, please note two things:
Whether there were any network drops, or Cursor reloaded the window during the session, like a Reload window prompt or a freeze
About which turn it cuts off, right after the model reply or after running a tool
This will help narrow down the trigger. I’ll post here if there’s an update.
This is hard to get aware because I work between several agents. I trigered this agent and turned to another agent, and then went back to that agent and find that the turns we worked on just lost.
Please solve this ASAP because it causes the intermediate file editing status to lose
Thanks for coming back with details, and it’s great that you were able to restore the chat from the transcript file.
A couple important points from your update: the fact that there were no network drops and no window reload, and that the loss happened specifically while working across multiple agents, changes what we think the trigger is compared to what we assumed before. I shared this with the team along with both session IDs. This is something we’re tracking. It shouldn’t happen, and it’s not anything in your setup. I can’t share an ETA yet, but I’ll post here if there’s an update.
To help us narrow down the trigger, can you confirm:
Were the agents you switched between local in the IDE, and when you switched, was the first agent still mid turn (generating a reply or running a tool), or had it already finished?
Does the tail get cut off right after the model reply, or after a tool call finishes?
For future recovery, the raw text of turns usually remains in the display-only transcripts on disk at ~/.cursor/projects/<workspace>/agent-transcripts/*.jsonl. You can pull the content from there to copy it out, like you already did via the agent.
Let me know the answers to the questions above, it’ll help narrow down the cause.
All cases happaned after I viewd the latest resp of the agent, and I turned to another thing, then somewhen I returned to that agent and encountered the problem
Thanks for the details, they really help. And it’s good that you were able to restore the chat from the transcript file.
One important detail from your update: if the agent had already finished its turn (you saw its last reply), then you switched to another agent and came back and the tail was gone, and there was no network drop and no window reload, that changes our hypothesis about the trigger. Before, we assumed the cut happened mid turn, but here the loss happens after completion while working across multiple local agents. I passed this to the team along with both session IDs. This is an issue we’re tracking. It shouldn’t happen, and it’s not something in your setup. I can’t share an ETA yet, but I’ll post here if there’s an update.
One quick question to help narrow it down: when you came back, had the agent that lost the tail been in the background for a while (you had time to work in other agents), or did it also happen with a fast switch back and forth? And about how many local agents were open at the same time?
For recovery going forward: the raw turn text usually still exists in the display only transcripts on disk: ~/.cursor/projects/<workspace>/agent-transcripts/*.jsonl. You can pull the content from there, like you already did via the agent.
Thanks, this helps. The picture is clearer now: the turn finished, you saw the agent’s last message, then you switched to another agent. It stayed in the background for a while, and when you came back, the tail was gone, with no network drop and no window reload. That’s different from what we first thought (a disconnect mid-turn) and points to a loss happening after completion when working across multiple local agents.
This shouldn’t happen, and it’s not something in your setup. It’s an issue we’re tracking. I shared both session IDs and this observation with the team. I can’t give an ETA yet, but I’ll post in the thread if there’s an update. Unfortunately, already-lost turns in the chat panel can’t be restored.
A couple tips for next time to lower the risk and avoid losing work:
The raw text of turns usually still exists in the display-only transcripts on disk: ~/.cursor/projects/<workspace>/agent-transcripts/*.jsonl You can pull the content from there, like you did via the agent.
If you work with multiple agents in parallel, try not to leave an active agent in the background for long right after a turn finishes. Based on what you saw, the loss happens in that exact scenario.
Let me know if you see another case with a different pattern. That’ll help us narrow down the trigger.