Agent UI: Collapsible Agent Responses

[fox-daniel]

Problem

When using Zed's Agent Panel (or the chat interface of Anthropic, OpenAI, etc) I find it hard to navigate long exchanges. I am often scrolling up and down trying to find the diamond in the ruff or to pick up a thread that I want to continue with.

Solution

Collapse previous responses from the AI Assistant when you input a prompt. Include a toggle to expand each individual response and the ability to expand or collapse all.

There are a lot of possibilities for more advanced UI, but I think the features described above would make chat navigation a lot more efficient.

It looks like the expanded_tool_calls in AcpThreadView struct within the agent_ui crate may provide a template for how to implement this. But I'm not sure. If there is interest in this I would be excited to work on a PR.

There is a short demo video linked in my github. But the implementation there is not related to how it would be done in Zed.

[fox-daniel]

Thanks to @aexvir for pointing out the link to his recent discussion 33401 about collapsing code edits. I like his proposal but would like to extend it to collapse previous AI assistant responses even if they don't involve code generation.

[aexvir]

yeah, sorry for the noise, when I first read it I thought it was the same I raised, so I wanted to suggest merging it, but after seeing the video I realized it was not the same, so I deleted the comment

[fox-daniel]

hi, alex. i hope both of these suggestions get some attention! it is crazy how primitive the UI is for this cutting edge technology. though it sounds like windsurf is better.

…

On Thu, Jul 17, 2025 at 2:43 PM Alex Viscreanu ***@***.***> wrote: yeah, sorry for the noise, when I first read it I thought it was the same I raised, so I wanted to suggest merging it, but after seeing the video I realized it was not the same, so I deleted the comment — Reply to this email directly, view it on GitHub <#34158 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMNJ4NGR77TVDSLIHBOE5R33I7VGFAVCNFSM6AAAAACBFJBGS6VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTGNZZGU3TONA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[benfolsom]

One option that wouldn't affect existing workflows much would be to have a button similar to "Scroll to most recent user prompt" which appears in the small bar that appears at the bottom of the chat. The new button could "collapse to the previous user prompt".

[tredondo]

👍🏻

Now that Zed has prioritize parallel agents, threads are getting longer and longer, and blocks (agenr responses, code, sub-agents, tools) are all expanded, which makes threads hard to scroll and combined with the lack of search or navigation to user prompts, pretty impossible to find stuff in.

[gerwitz]

I have an agent harness with a TLDR mode that, in my own client, displays short summary responses with progressive disclosure for more detail. I was hoping that the ACP priority field would enable something like this in Zed (e.g. sending text annotated 1.0 for the summary, then 0.0 for the details) but of course that does nothing.

I would love for Zed to have some mechanism to display semi-collapsing agent responses, akin to the way tool use message are.