Copilot is constantly summarizing conversation history and it takes 5+ mins every time. It is making copilot very slow/hard to use. This seems to have been getting worse and worse with time.

[james-belva]

Select Topic Area

Product Feedback

Copilot Feature Area

VS Code

Body

Copilot in vscode is starting to get to be impossible to use. It is constantly summarizing conversation history almost every other interaction and it is taking minutes to summarize the history. Often critical points get lost in the summarization and I have to remind the agent of details that it forgot in the summarization often triggering another conversation summarization. It is getting ridiculous and I am not getting this same nonsense in Cursor.

[github-actions[bot]]

💬 Your Product Feedback Has Been Submitted 🎉

Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users.

Here's what you can expect moving forward ⏩

• Your input will be carefully reviewed and cataloged by members of our product teams. 
  • Due to the high volume of submissions, we may not always be able to provide individual responses.
  • Rest assured, your feedback will help chart our course for product improvements.
• Other users may engage with your post, sharing their own perspectives or experiences. 
• GitHub staff may reach out for further clarification or insight. 
  • We may 'Answer' your discussion if there is a current solution, workaround, or roadmap/changelog post related to the feedback.

Where to look to see what's shipping 👀

• Read the Changelog for real-time updates on the latest GitHub features, enhancements, and calls for feedback.
• Explore our Product Roadmap, which details upcoming major releases and initiatives.

What you can do in the meantime 💻

• Upvote and comment on other user feedback Discussions that resonate with you.
• Add more information at any point! Useful details include: use cases, relevant labels, desired outcomes, and any accompanying screenshots.

As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities.

Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐

[TanayBhalani]

I noticed the summarizer request timed out after 10 mins.
This could probably be one of the reasons for it to stay running.

[irvin-reyes-uvi]

the same than me the model detects the bottle neck and keep spending your money without telling u that the responses are not accurate and stayed in loop of summarising as if it is a summarize machine nor an agent coder.

[khandg]

I am experiencing the same problem recently - significantly - where i am working on a module (feature enhancement) in an existing project and conversation is very long because of planning, todos creation and research. Currently the situation is copilot is creating 2-3 files (roughly ~1500 lines of code) and then almost take 5 minutes for summarising the history. This is repeating with same frequency

Overall execution time is getting too slow and painful

model being used is Claude Sonnet 4.5

As the conversation is becoming longer, i can also notice the code quality is getting degraded and getting a lot of compilation errors and even syntax errors.

[EricJilot]

I am having the same issue

[t-fontes]

same issue, at a point it takes so much time it becomes useless.

[mempler]

ngl, this is becoming such a huge issue that i consider completely switching to cursor

[alex-macedo]

It is painful to use, and I am not seeing how to not move to AntiGravity, Cursor, or OpenCode.
It's not a helpful tool right now.

[ciacob]

Same issue here. And I remember better times, still. This started to get worse when "thinking" was introduced.

[sounkou-bioinfo]

Same issue here. And this makes the product basically useless

[ekimminau]

Im seeing summarization after almost every interaction. The Gemini models almost always then start going into a massive loop of repeating the same action 1000 times in a row. Ive seen GPT 5.x models doing the same and its happened twice on Claude 4.5. Whatever changes were made just before and just after Christmas is a massive regression.

[DAE51D]

Add my "me too" here. However I use VSCode and Copilot all day long at work and don't experience this. I am noticing it at home though on my personal account. I almost always use "Auto" or "Claude 4.5 Sonnet" (in both work/home). This was NOT an issue a week ago IIRC. I just started noticing it the past couple of days literally and I use this from everything from System Administration, to code, to home automations/dashboards, etc. I use it a LOT. I thought maybe it was some "experimental feature" I maybe enabled, but now I see others with my experience here...

For example, this is after a /clear and a simple paragraph sized prompt. It just does this over and over and over... incredibly frustrating and counter-productive.

Perplexity says, "it's not you, it's them" but IDK... this seems a pretty glaring problem and I'd think more people would be complaining loudly/louder. Are we part of a canary rollout maybe and that's why only some folks experience this terrible bug? I had it examine my settings.json file and it found nothing of note (I thought maybe I toggled some setting)

[DAE51D]

I may have stumbled upon a solution, at least it seems to have fixed mine... I had been using Copilot to make ComfyUI workflows when I was noticing all of these excessive summarizing (post above). That was on my notebook, but in an attempt to see if this was specific to that, I fired up VSCode on my mid-tower PC. Same issue (no surprise there). However what did catch my eye is this after a /clean prompt start.

So I shifted gears and asked this...

I just noticed in another chat you used 20 references?! that seems excessive for an initial prompt. Is there something in the #file:copilot-instructions.md that is causing that? #vscode #vscodeAPI

And sure enough... THAT seems to be the culprit!

Which I then let it refactor my .github/copilot-instructions.md file and I've been using VScode/Copilot for two days now without ANY issues (I waited to post this just to be sure I really was running it through the paces; also it's the last day of my subscription so I have a lot of prompt tokens to use up today since I hadn't been using this for a week or so due to the OP 😄)

[DAE51D]

OH and the #vscode and #vscodeAPI you can ignore in your prompt because I also asked about having THREE "github" MCP servers for some reason, so we pruned it down to just the extension one. If you're not aware, chat can edit things IN VSCode (.json configs, etc.) on your behalf if you ask it (and give it the tool permissions above). Which it did for my mcp.json file...

[Hold2Admin]

Is this just not being responded to by support? I don't understand how we're supposed to use the tool if the assistant summarizes conversation history so often and completely forgets everything. The summary is basically useless because it doesn't ever remember the most recent command/message/topic, and I have to fully remind it to read my prime directives again each time. If I don't, it will immediately start violating core development requirements and best practices as if it has never worked with me before. However, after I tell it to re-read the session instructions, I'm already halfway to the tiny 128k token limit.

Did that limit change with the recent app update? Now, I can see the token indicator, and it gets eaten up so fast that I barely have time to do any work before it summarizes again. It wasn't this bad over my first 4 months but has gotten nearly unusable since the update a few days ago. Was the token limit always this low? The assistant previously told me it was 1 million tokens, but I was not able to see the number myself. All I know is I hit the limit so fast now that I can barely work, and I run the risk of destroying my whole project if I don't remind the assistant of the prime directives (instructions) every time the session restarts itself.

Even having the full subscription plus playing hundreds more per month doesn't seem to get me a reasonable token limit. Can anyone explain?

[DAE51D]

@Hold2Admin did you read my solution and try it for yourself? see how many references are showing, it shouldn't be more than a handful at most. Ensure you don't have any large files open in VSCode editor window (maybe a log file or other ones that have a lot of text in them). Have Copilot analyze your copilot-instructions.md file and see if there are excessive "includes" that don't need to be included but can rather be on-demand referenced as needed.

Also ya'll might be interested in a .copilotignore file to also ignore .log, .zip (or other archives), images, etc... have copilot generate the file for you (just as you would a .gitignore file)
https://marketplace.visualstudio.com/items?itemName=Mattickx.copilotignore-vscode

[Hold2Admin]

Hey @DAE51D I wrote a long reply down below and I think I did try what you said: https://github.com/orgs/community/discussions/178513#discussioncomment-15849414.

I actually keep all files closed except when directly editing and then I close them immediately. I'm highly vigilant about trying every trick I've heard of. It might help a little, but the time between summarizations is disappointing. It's not immediate or anything, but it's difficult to get any sizable task done without spanning 3 or more sessions...sometimes as many as 10. It's so difficult to "retrain" the assistant when it is basically a newborn baby every time it summarizes the conversation.

I did exactly what you said with my copilot instructions and got them paired down to the dramatically with no file references. He's my prompt and the answer in case this helps:

I'll check out Copilot Ignore, thank you. I don't typically have any .log or .zip files but will look through it to see if anything can help.

[brianfobe]

I have the same issue, I'm more waiting then working.

[tcies]

As copilot just confirmed to me, this happens because the model you are using is running out of context capacity. What it probably does is writing a summary, then reseeding a new session with just the summary (rather than the full previous conversation).

As @DAE51D suggested, you can see whether context isn't bloated for some silly reason (e.g. parsing of very large files). Note that best practices mention that any open files are typically added to the context, so you might e.g. want to close irrelevant files: https://docs.github.com/en/copilot/get-started/best-practices#guide-copilot-towards-helpful-outputs

Alternatively, you can also switch to a model with a larger context size. You can check the model context size by clicking "Manage Models" in the model selection dropdown:

[Hold2Admin]

Hello @tcies you are right about all this, no doubt. In my case:

1. Yes, it's running out of context rapidly every session

2. Yes, it then summarizes conversation history, starts a new session inline, and loads that summary. I can't see the summary, have no idea how helpful it is nor what it "remembers" from the prior session. This happens mid-operation frequently, so the assistant will be fully educated on the repo, the goals, the schema, and the issues before the summarization. Then, immediately after summarization, it will act like it's meeting me for the first time and will claim to not even be able to see the most recent message I wrote directly before the summarization occurred. It loses virtually all context, and whatever context it retains from the summary is random crapshoot with no reliability

3. I keep no documents open for the reason you stated. Nothing is open, and I've pruned my session instructions to almost a form of illegible shorthand just to minimize the token burn every time the new session starts. Unfortunately, the assistant frequently forgets and violates these instructions only to then apologize each time and say "sorry, that's on me. That won't happen again" before then doing it almost on cue with the very next request

4. I've had to create extremely concise secondary versions of my design documentation and schema files so the assistant can "re-learn" the project after each summarization without burning too many tokens.

5. I had to create a prime directive in the session instructions IMPLORING the assistant to respond with extreme brevity so it doesn't waste tokens with long summary responses. I even demand 1-word responses often.

6. I can't switch to another model, because 128k - a tiny amount - is already the largest size. There should be 1 million tokens available. The summary + instructions reload + project documentation seeding take up at least 1/3 of the tokens each time before I even do anything. There was one point where it was taking up 2/3 of tokens, which led to my extreme measures.

[HazelLessiter]

In my case this is not accurate. I got this after a new conversation sending one message and giving the context of one small file. (Visual Studio Pro 2026)

[rsmvdl]

Sucks to use ... As everything from microslop.