Agent Panel: stopped working for me, no response/feedback after typing message (video included)

[rayburgemeestre]

Summary

First of all, awesome editor. After trying it out switched to it for work and personal projects! :)

This is perhaps something specific to just me for some reason, because I would have expected an existing bug report, I'll still share what I see, perhaps it's useful anyway... in my case the agent panel stopped responding.

I have tried: restarting the editor signout + signin switching model (even though I only have the anthropic models) tried also max mode, etc.

Inline AI still works control+enter, and typing something there.. so in my case it seems specific to the agent panel.

Description

Steps to trigger the problem:

1. Sign in
2. Start chat
3. Send some message

Actual Behavior:

Nothing happens, nothing printed to console either (I have a video + log uploaded)

Expected Behavior: some response from the model (it worked fine yesterday for me)

zed-output.txt

video: https://cppse.nl/tmp/simplescreenrecorder-2025-05-17_12.06.07.mkv

Zed Version and System Specs

Zed: v0.186.9 (Zed) OS: Linux X11 neon 24.04 Memory: 62.4 GiB Architecture: x86_64 GPU: Intel(R) Arc(tm) Graphics (MTL) || Intel open-source Mesa driver || Mesa 24.2.8-1ubuntu1~24.04.1

[senko]

Same here, identical behavior: Agent panel ignores whatever I type (any model, tried a few), inline AI also works. Nothing in Zed log.

This also worked for me until yesterday or the day before (not sure about exact timing when it started).

Zed 0.186.9 on Ubuntu 24.10

[senko]

Upgrade to a newer version (0.186.11) seems to have fixed it for me (maybe was already fixed for .10, y'all are doing updates faster than I can upgrade :)

[thomascjohnson]

I am having the same issue on a M4 Mac on 0.187.5. I downgraded to 0.186.11 and it works again. I also tried with a variety of different versions of Ollama, but that was before I realized the inline AI works – so probably not related to Ollama.

Ok, strangely enough a fresh install of 0.187.5 from the releases page works. So it seems to be something with the self update?

[rayburgemeestre]

Well I couldn't get it to work, even tried purging ~/.local/zed* and ~/.local/share/zed*, but I had to reinstall my laptop anyway and went with Ubuntu 25.04. Funny enough I restored my original ~/.local dir etc., and everything just worked. So that might kind of rule out that there is some problem there with the data

[edwinolson]

I've been having a similar problem with zed+agent. When I enter into the agent panel and hit return, my input disappears but nothing else happens*. (* actually, I've noticed that sometimes I will return to zed hours/days later and there will be a response in the agent history). This seems to happen regardless of LLM model.

This has been consistent for the last several versions-- in fact, I've never had the agent panel work right. (Inline completion works fine).

I've tried reinstalling, I tried nuking my .config/zed/settings.json. No dice...

MacOS m3max, sequoia 15.5, latest zed

[rayburgemeestre]

Totally forgot, so I didn't report it, but I experienced exactly the same ! Sometimes I'd see a response half hour later as well. Quite odd. I was trying stuff like 'test', 'asdf' , 'kljgk', a couple of times, and I'd suddenly see a response to test, out of the blue, and if I waited long enough, the asdf one as well.

[edwinolson]

Is it possible that the agent panel is scanning files in the current working directory (and maybe its children?)

if I start zed in my home directory or in ~/empty/, agent works.

if I start zed in my big projects directory, agent doesn't seem to work. (But recall from above that I do sometimes see responses after a LONG time-- so maybe it's "working" but the latency is incredibly high because of some scanning it's doing?) My big projects directory has about 800k files in it.

[edwinolson]

OK, some digging later--- I've instrumented the zed source code with many log messages, and the long delay appears to be related to zed doing a bunch of git work.

agent::mesage_editor starts a git checkpoint task... then:

git::repository does a ton of weird stuff that I don't understand, staging changes and calling commit-tree, manipulating the excludes, and then it calls "git add --all". This appears to be where the delay seems to come from, as git then scans my entire project directory with 800k files in it.

As a neophyte, I'm not sure why zed is doing so much git work at this point (would love to learn!), but it'd be great to eliminate or streamline it so that the agent is more responsive. (Is it creating a git commit that can be referenced as a identifier for future tool use? )

[maxdeviant]

It sounds like the slowness is likely from creating a checkpoint when sending a message (which we use Git for).

We've introduced a fix in https://github.com/zed-industries/zed/pull/31352.

[danilo-leal]

Closing this one as per the following PR:

• https://github.com/zed-industries/zed/pull/31352.

If the issue persists, please open a new one—thanks for all the detail & information everyone!

[hamishc]

The above PR doesn't seem to have fixed this issue for me. On opening zed in a large private repo that doesn't contain binary files, I'm seeing an no responses to any messages in the agent, without anything reported in the logs. I believe creating the checkpoint for this repo may just be taking a very long time, but I don't know how to investigate.

[thomasvuong]

I am having the same issue on a M4 Mac on 0.187.5. I downgraded to 0.186.11 and it works again. I also tried with a variety of different versions of Ollama, but that was before I realized the inline AI works – so probably not related to Ollama.

Ok, strangely enough a fresh install of 0.187.5 from the releases page works. So it seems to be something with the self update?

Could you please share this 0.186.11 version?