The inline assistant's streaming-tool path waited for the first
`rewrite_section` output, but treated any other completion event as
terminal. Responses API providers can emit `ReasoningDetails`,
`Thinking`, and other metadata before the tool call. This stopped the
stream early and marked generation as done without applying a
replacement.
EOF was not handled explicitly. Parse errors, provider errors,
incomplete tool calls, and truncated responses could also be reported as
successful completions.
Continue consuming informational events until a supported tool call
arrives. Track the rewrite tool call's streaming state, reject a second
`rewrite_section` call instead of concatenating unrelated outputs,
accept `failure_message` only after its input is complete, and reject a
`rewrite_section` arriving after a failure message (and vice versa).
Stop, EOF, malformed tool input, and stream errors now surface as
`CodegenStatus::Error` instead of silently completing. A completed tool
call followed by `Stop(MaxTokens)` or `Stop(Refusal)` is also an error:
the OpenAI Responses mapper flushes truncated tool calls as complete
before emitting the stop reason, so the stop reason must be checked even
when the tool input parsed. Completed failure messages are stored before
`Finished` is emitted, and a failure message left over from a previous
generation is now cleared when a new one starts.
Add regression coverage for metadata around streamed rewrites, Stop and
EOF without a tool call, incomplete rewrites, interrupted responses,
multiple rewrite calls, failure messages, malformed input, mixed tool
calls, and stream errors before and after rewrite output begins.
---
Closes#52714
Release Notes:
- Fixed the inline assistant doing nothing when the model emits
reasoning before its tool call.
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Smit Barmase <heysmitbarmase@gmail.com>
Closes#60013
# Objective
Right now Zed can go full screen but it does not allow you to fix the
hole screen,
by that I mean that Zed can go behind the notch so you don't have extra
useless room left.
## Solution
You can now use the `fullscreen_mode` = `simple` setting to use the new
simple full screen feature, that lives besides the normal full screen
feature. But allows you to have an option to go 100% full screen without
losing any useless space on your macbook screen.
**Note** this is mostly usefull when you have a macbook that has a notch
whitch is kinda in the way of your work flow.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
## Showcase
**Before**
<img width="5712" height="4284" alt="IMG_0339"
src="https://github.com/user-attachments/assets/9f908ffd-7cef-4999-a454-c80f72c40dc8"
/>
**After** (Note now Zed is behind your notch when using the simple full
screen feature)
<img width="5712" height="4284" alt="IMG_0360"
src="https://github.com/user-attachments/assets/5917ed4d-2a64-4464-a794-bc46fd034521"
/>
---
Release Notes:
- Added support for simple fullscreen mode using the `fullscreen_mode`
setting, set it to `simple` to try it out.
`ThreadView::regenerate` does `thread.rewind()` which de-focused the
message editor mid-flight causing the zoomed panel to disappear.
* workspace: Keep zoomed panels open when window focus is lost
Commit adds a test + tweaks a default, generic, fallback to be more
lenient: instead of focusing the workspace (and hiding all panels right
away), try to fall back to whatever panel open
* gpui: Restore focus to nearest surviving ancestor when focus is lost
Commit tries to push it down to a library level and add a way to fall
back to closest ancestor that's still visible with the focus instead.
Release Notes:
- Fixed the zoomed agent panel closing unexpectedly on rewind
## Why
`textDocument/onTypeFormatting` edits that insert or replace text at an
empty cursor use its right bias and move it past the new text. In paired
tags, pressing Enter can therefore leave the cursor on the closing tag
instead of between the tags.
## What
- Capture a left-biased pin for each empty cursor before requesting
on-type formatting.
- Skip cursor tracking unless a matching language server advertises the
trigger.
- Restore only unchanged empty cursors whose displacement is fully
covered by formatting transaction ranges, so intervening user edits are
preserved.
- Reset vertical movement state when restoring a cursor.
## Testing
- `cargo test -p editor test_on_type_formatting` (5 passed)
- `./script/clippy -p editor`
## References
- Fixes https://github.com/zed-industries/zed/issues/61574
Release Notes:
- Fixed the cursor being moved past text inserted or replaced at its
position during on-type formatting.
---------
Co-authored-by: Kirill Bulatov <kirill@zed.dev>
Closes#39860
This PR resolves relative image paths from the Markdown source file's
project path and load images through the project image store. SVG images
over remote connections remain unsupported and are left for a follow-up.
Release Notes:
- Fixed images not rendering in Markdown Preview over remote.
While building Zed with nightly rustc I've noticed it doesn't compile
because of good old pathfinder_simd. It also emits a bunch of warnings
about use of f64 literals where f32 is expected, so I've fixed them - it
should make future upgrades more straightforward.
# Objective
Prevent the thread from automatically scrolling to the end when
expanding a "Context Compacted" message at the end of the thread, as we
suspect the user would likely be trying to read it from top to bottom.
## Solution
Update how the `ThreadView` behaves when expanding a "Context Compacted"
message, namely by updating `ThreadView::toggle_compaction_expansion` in
order to anchor the scroll position, which disables the tail following
behavior, ensuring that the scroll offset is kept.
In order to better support this, a new method is introduced to
`gpui::elements::list::ListState`, `anchor_scroll_position`, which takes
the current scroll offset and scrolls to it which, in turns, disables
the following behavior.
Another option would be to simply introduce
`gpui::elements::list::ListState::stop_following` but I'm not sure
whether we'd want to expose that implementation detail, even though we
already have `gpui::elements::list::ListState::is_following_tail` .
## Testing
Tested both manually as well as introduced a new test for
`ListState::anchor_scroll_position`.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [ ] Unsafe blocks (if any) have justifying comments
- [ ] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
## Showcase
<details>
<summary>Before</summary>
https://github.com/user-attachments/assets/cf8ee10f-9c74-429d-afed-7a39ccd71c67
</details>
<details>
<summary>After</summary>
https://github.com/user-attachments/assets/78edfca9-099a-4fdb-9c4e-75f790096957
</details>
---
Release Notes:
- Improved expanding "Context Compacted" messages in the Agent Panel to
ensure that the scroll position is preserved, instead of automatically
scrolling to the end of the thread.
---------
Co-authored-by: Joseph T. Lyons <JosephTLyons@gmail.com>
Enables zooming of mermaid diagrams by supporting horizontal scroll
---
Release Notes:
- improved: Mermaid diagrams can now be zoomed and horizontally
scrolled, in both the markdown preview and the agent panel
# Objective
Fix the Agent sidebar header being two pixels lower than the title bar
on Linux.
This only affects Linux with client-side decorations: the main title bar
uses a negative one-pixel margin to overlap the client-drawn window
border, while the sidebar previously used a positive one-pixel margin on
every platform.
Fixes#53739
## Solution
Match the sidebar header spacing to the window decoration mode:
- Use `-1px` top margin for client-side decorations.
- Preserve the existing spacing for server-side decorations.
- Apply the fix to both the thread list and archive headers.
## Testing
- Manually tested on Linux. This only affects Linux because its
client-side window decorations require the adjusted title bar offset.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([[UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [ ] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
## Showcase
After this change：
[2026-08-02
18-20-19.webm](https://github.com/user-attachments/assets/1ef13ff2-2036-442b-8bcc-74ebeb8436c7)
---
Release Notes:
- Fixed misaligned Agent sidebar headers on Linux.
The agent panel's full screen button and a pane's zoom button drive the
same mechanism — both emit `PanelEvent::ZoomIn`/`ZoomOut`, and
`Workspace::zoomed` holds a single entry, so at most one of them is
active at a time.
The pane's button reflects that state: it is rendered with
`toggle_state(zoomed)`, so it is visibly active while zoomed
(`crates/workspace/src/pane.rs`). The agent panel's button only swapped
its icon, so while the panel was zoomed the button still looked like an
inactive control.
This makes the agent panel's button use the same `toggle_state` +
`selected_icon` pattern as the pane's. Behavior is unchanged; the button
now looks active while full screen is on. The element id is now stable
across both states instead of changing with the icon.
Unrelated to this change, but worth mentioning since it is the same
button: the two controls also disagree on wording. The pane calls it
"Zoom In"/"Zoom Out", the agent panel "Enable/Disable Full Screen".
Because `Workspace::zoomed` is a single slot, activating any other
panel's zoom drops the agent panel back out of full screen — which reads
as surprising under the "full screen" name and unsurprising under
"zoom". Happy to align the wording in a follow-up if you'd like, but I
left it alone here since it is a naming call rather than a fix.
Release Notes:
- Fixed the agent panel's full screen button not appearing active while
full screen was enabled.
---------
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
Splits the crate graph.
Before, the compilation graph: `editor → picker_preview → search →
project_panel → open_path_prompt → recent_projects → title_bar →
collab_ui → zed`
After, the compilation graph: `editor → picker_preview → search →
agent_ui → sidebar → zed`
With sccache disabled and project fully built, `touch
crates/editor/src/editor.rs` and `cargo build -p zed` took time
Before (0500fd9ee9): 13.19s
After: 11.85s (-10.2% speed up)
Each commit contains a separate compilation instructions, and a test in
the final commit.
The main approach is to split coupled crates and replace them with
`zed_actions` and move some shared functionality into `git_ui_core` new,
shared module.
I've also tried to add a test to prevent common pitfalls, but not sure
I'm happy with the end result — can remove it if it looks too synthetic.
Release Notes:
- N/A
# Objective
Closes#56127
Right now there is an issue where in certain contexts, specifically GPUI
div elements, horizontal scrolling w/ a trackpad works very poorly, even
when trying to scroll sideways, vertical scrolling will happen.
This happens with both a trackpad and scrollwheel. The showcase video
also shows the broken behavior.
## Solution
The way that scrolling works in the editor is really nice, scrolls are
locked to the axis they started on but with enough force can change
axis. so by taking the ongoing scroll handler from the editor and moving
it up into GPUI, we can get that same functionality for GPUI.
I somewhat "take over" the functionality of the existing
`restrict_scroll_to_axis` property, because it seems like its purpose
was basically trying to fix this problem already, just unsuccessfully.
The alternative, and would make things overall nicer to look at, would
be to tie this to `allow_concurrent_scroll`, but the effect radius tied
to that option is way bigger, so im holding off on that atleast right
now.
## Testing
Theres a good number of scroll related tests generally, so I don't think
i've perturbed any common existing behaviors. attach the functionality
to `restrict_scroll_to_axis` really limits the surface. There are also
some new tests for the ongoing_scroll behaviors.
Tested on my mac w/ both trackpad and mouse.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
## Showcase
Video show casing all the behaviors:
broken w/ trackpad
working w/ trackpad
broken w/ scroll
working w/ scroll
https://github.com/user-attachments/assets/1935cd2d-a255-4a77-81d5-daa9f72470a6
---
Release Notes:
- markdown_preview: Fix horizontal scrolling inside certain elements
# Objective
- Prevent MCP extension configuration dialogs from appearing when
switching to background workspaces.
- Closes FR-145.
## Solution
- Route configuration requests through the active application window and
workspace, while keeping uninstall settings cleanup global.
## Testing
- Added a regression test covering active and retained background
workspaces.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
---
Release Notes:
- Agent: Fixed MCP extension configuration dialogs appearing in other
workspaces.
Fixes the bug that made us remove the sandbox.
The bug in question was very dumb:
- there is sophisticated machinery for detecting whether a user-granted
writable path is swapped out for a symlink in the timing gap between
approval and sandbox creation
- there was no equivalent machinery to do the same for the (much larger)
gap between a user *persisting an approval* (either for the current
thread or permanently via settings)
- The fix is essentially to store canonical (i.e. absolute and
symlink-free at all depths) paths as the source of truth, but retain the
raw path for display purposes
- On WSL, there is extra care needed becasue of the bidirectional
mounting (i.e. `/mnt/c/...` and `\\wsl.localhost\Ubuntu\...`). In
particular, `/mnt/c/...` paths, since their inodes do not necessarily
pin NTFS file references, weaken the sandbox guarantees, and so we need
some extra UI to call this out and docs etc...
This also does not remove the feature flag, but just toggles it to
"enabled_for_all"
---
Release Notes:
- N/A or Added/Fixed/Improved ...
---------
Co-authored-by: Richard Feldman <oss@rtfeldman.com>
Co-authored-by: Jakub Konka <kubkon@jakubkonka.com>
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
Co-authored-by: zed-zippy[bot] <234243425+zed-zippy[bot]@users.noreply.github.com>
Closes FR-139
Most focus-lost issues that close the agent panel in Zen mode stem from
panels returning a transient child editor's focus handle from
`Focusable::focus_handle`. The Zen-mode check closes the zoomed panel
when the focused element is not a descendant of the panel's focus
handle, so any focus movement inside the panel that landed outside that
child editor looked like the panel losing focus.
This PR separates the two roles that handle was serving:
- `Focusable::focus_handle` now always returns a stable handle tracked
at the panel's root element, used for containment checks (Zen-mode
auto-close, `toggle_panel_focus`, dock focus subscriptions).
- The new `Panel::activation_focus_handle` returns what should receive
focus when the panel is activated (e.g. a filter/commit/message editor)
and must be a focus-tree descendant of the root handle. All "focus the
panel" callsites in workspace/dock now use it.
Migrated panels: `AgentPanel` (including thread views and the toolbar
title editor), `CollabPanel`, `GitPanel`, `OutlinePanel`, and
`TerminalPanel` (previously delegated to the active pane, so focus in a
non-active pane dropped containment). `CollabPanel` activation falls
back to the panel root when signed out or collaboration is disabled,
since the filter editor isn't rendered in those states. Also fixes
duplicate element ids for tool calls with multiple content blocks.
Tests: a regression test that clicks tool-call output and focuses the
thread title editor while zoomed (both previously closed Zen mode), and
a dock-level test asserting the activation handle receives focus on
panel activation while the panel root reports containment.
Release Notes:
- Fixed zoomed panels (e.g. the agent panel in Zen mode) closing
unexpectedly when focus moved to elements inside the panel, such as tool
call output or the thread title editor.
Objective:
Allow users to configure the font families used by agent panel content
separately from the main UI and buffer fonts.
Solution:
- Add `agent_ui_font_family` for agent responses in the agent panel.
- Add `agent_buffer_font_family` for the agent panel message editor and
user messages.
- Keep context menus on the primary UI font family.
- Document the new settings and expose them in the settings UI.
Testing:
- Ran `cargo fmt`.
- Ran `cargo check -p settings_content -p theme_settings -p markdown -p
agent_ui -p settings_ui -p ui`.
- Ran `git diff --check`.
- Manually tested the agent panel font behavior in a dev build.
Release Notes:
- Added settings for configuring agent panel UI and buffer font
families.
Co-authored-by: Finn Evers <finn@zed.dev>
## Summary
- Dragging a binary file (docx, xlsx, pdf, etc.) into the agent panel's
message editor inserted a mention link, then asynchronously tried to
load its content as a text buffer. That failed with "Binary files are
not supported", which surfaced an error toast **and** deleted the
mention text that had already been inserted.
- `confirm_mention_for_file` now catches that specific failure and falls
back to `Mention::Link` (a path-only reference, same as directory
mentions already use), instead of failing the whole mention and
reverting the edit.
## Test plan
- [ ] Drag a `.docx` or `.xlsx` file into the agent panel chat input —
the file path mention should stay inserted with no error toast.
- [ ] Drag a regular text file in — content mention still works as
before (unaffected code path).
Release Notes:
- Fixed dragging a binary file (docx, xlsx, pdf, etc.) into the agent
panel deleting the inserted reference and showing a "Binary files are
not supported" error
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Cole Miller <cole@zed.dev>
# Objective
Fixes#59333.
Selection mentions created with "Add to Agent Thread" could not be
clicked to open the referenced file, while selection mentions created by
pasting were navigable.
## Solution
The ready selection-mention crease did not receive its `MentionUri` or
`Workspace`, so `MentionCrease` could not install its click handler.
Pass the workspace through the action-completion callback and provide
both values to the shared crease renderer, enabling its existing
navigation behavior.
## Testing
I tested these changes locally.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [ ] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
## Showcase
Before:
https://github.com/user-attachments/assets/8dd4a2d0-4b22-4fb3-8777-cf97b6436b91
After:
https://github.com/user-attachments/assets/cff1cdb3-804a-45d8-a9c5-fdd8b35b6f8f
---
Release Notes:
- Fixed selection mentions added with "Add to Agent Thread" not opening
their referenced files.
Self-Review Checklist:
- [ *] I've reviewed my own diff for quality, security, and reliability
- [ *] Unsafe blocks (if any) have justifying comments
- [ *] The content is consistent with the [UI/UX
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
- [ *] Tests cover the new/changed behavior
- [ *] Performance impact has been considered and is acceptable
What:
- Closing the active terminal from the sidebar activated the remaining
terminal.But closing a terminal with Ctrl-D could leave the agent panel
in an
uninitialized `Zed Agent` state instead of activating the remaining
terminal.
<img width="804" height="380" alt="image"
src="https://github.com/user-attachments/assets/eaa67c6e-7f3d-4a1c-b72d-9cabfaab872b"
/>
How:
- Terminal exits emit `TerminalCloseRequested`.
- The sidebar handles that request through the same close path used by
the
sidebar close button.
- Neighbor activation, draft fallback, metadata cleanup, and
linked-worktree/
archive handling stay centralized in the sidebar close flow.
Added test coverage for closing an active terminal via terminal exit and
activating the remaining terminal neighbor.
Testing:
- `cargo check -p agent_ui`
- `cargo check -p sidebar`
- Manually verified Ctrl-D terminal close behavior with multiple agent
threads/terminals
Release Notes:
- Fixed closing an agent terminal with Ctrl-D leaving the agent panel
uninitialized instead of activating the remaining terminal
# Objective
Add a new `agent.compaction_model` setting that lets users specify a
separate language model for context compaction (`/compact` and
auto-compaction), independent of the thread's active conversation model.
Compaction is just summarization — there's no reason to pay Opus prices
for it when a cheaper model does the job faster. We've also seen
reasoning models misbehave on this task (empty responses, repetition
loops at high effort), so picking a dedicated non-reasoning model for
compaction is useful.
## Solution
- New `compaction_model: Option<LanguageModelSelection>` field in
`AgentSettingsContent` and `AgentSettings`, mirroring
`thread_summary_model`.
- New `compaction_model: Option<ConfiguredModel>` slot on
`LanguageModelRegistry` with `select_/set_/compaction_model()` trio,
mirroring the existing pattern. The setter deliberately does **not**
emit a registry event in v1; callers read the slot lazily at compaction
time.
- New `Thread::compaction_model(&self, cx: &App)` helper that returns
the configured model or falls back to `self.model()`. Two call sites —
`Thread::compact` and `perform_compaction_if_needed` — now go through
this helper instead of reading `self.model()` directly.
- `build_compaction_telemetry` accepts the compaction model explicitly
so the `model` field reflects the model that actually streamed the
request. `max_tokens` still derives from `thread.model()` (threshold
semantics are unchanged).
- Documentation updated at `docs/src/ai/agent-settings.md` (new
user-facing setting).
**Example:**
```json
{
"agent": {
"default_model": {
"provider": "anthropic",
"model": "claude-opus-4-6"
},
"compaction_model": {
"provider": "anthropic",
"model": "claude-sonnet-4-5"
}
}
}
```
**Resolution chain:**
```
agent.compaction_model (if set & available)
→ thread.model() (always available, current behavior)
```
**Behavior change:**
| Trigger | Before | After |
| --- | --- | --- |
| Manual `/compact` | Uses `thread.model()` | Uses
`agent.compaction_model` if set & available; else `thread.model()` |
| Auto-compaction | Uses `thread.model()` | Same as above |
| `/compact` when thread has no model | `NoModelConfiguredError` |
Succeeds if `compaction_model` resolves |
| `compaction_model` configured but provider missing / model id unknown
| n/a | Falls back to `thread.model()` and logs a one-time warning |
**Explicit non-goals:**
- No `Event::CompactionModelChanged`(no consumer; the `_cx` parameter on
`set_compaction_model` is intentionally accepted for future use).
- No GUI selector (consistent with all other feature-specific models).
- No per-profile override.
- No runtime API-error fallback — only config-time failure (provider not
registered, model id not in `provided_models`) triggers fallback. This
matches every other feature-specific model.
- No change to threshold calculation, auto-compact trigger, or
`COMPACTION_PROMPT`.
- No propagation to subagent threads.
## Testing
3 new unit tests in `crates/agent/src/thread.rs::tests`:
- `test_compaction_uses_configured_compaction_model` — manual `/compact`
routes to the configured model; thread's primary model receives no
request; telemetry reflects the configured model.
- `test_compaction_falls_back_when_compaction_model_unavailable` —
configured-but-unresolvable falls back to `thread.model()`; telemetry
reflects the fallback model.
- `test_auto_compaction_uses_compaction_model` — auto-compaction
triggered by threshold honors the same setting.
All 14 existing compaction tests still pass. Test suites in
`crates/agent_settings`, `crates/language_model`,
`crates/settings_content`, `crates/agent_ui` unchanged. `cargo clippy`
clean on the changed crates.
**How reviewers can test:**
1. Add `agent.compaction_model` to `settings.json` with a cheaper model
than the thread's primary model, run `/compact`, observe the cheaper
model receives the request.
2. Set `agent.compaction_model` to a non-existent provider/model id, run
`/compact`, observe fallback to thread model and a `log::warn!` line.
3. Trigger auto-compaction by reaching the threshold, observe it uses
`compaction_model`.
**Platforms tested:** local Linux (cargo check + cargo test on agent /
agent_settings / language_model / settings_content / agent_ui crates).
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments — no `unsafe`
blocks introduced
- [x] The content adheres to Zed's UI standards — N/A: settings-only
change, no UI touched
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable — model
resolution is an O(1) registry lookup, not on any hot path; no new work
in the streaming loop
Release Notes:
- agent: Add support for specifying which model is used for compaction
(`agent.compaction_model`)
---------
Co-authored-by: Bennet Bo Fenner <bennetbo@gmx.de>
Co-authored-by: Bennet Bo Fenner <bennet@zed.dev>
This adds a dedicated `AvailableLanguages` struct in preparation for a
more cabable language matching based on a given language config. No
functional changes, just shuffling some code around for this round.
Also made the LanguageMatcher non-cloneable in favor of wrapping it in
an Arc, since cloning is rather expensive for this and having a
reference is sufficient in all cases right now.
Release Notes:
- N/A
---------
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
# Objective
Remove `.unwrap()` calls in iterator/`Option` chains that could panic at
runtime.
## Solution
Use `and_then`/`filter_map` with `.ok()` for fallible downcasts
(`entry_view_state.rs`) and keystroke parsing (`vim/command.rs`); use a
descriptive `expect` in `suggest_autoindents` where the invariant is
intentional (`buffer.rs`).
## Testing
- `cargo test -p language buffer` (69 passed)
- `cargo test -p vim command` (26 passed)
- `cargo test -p agent_ui entry_view` (3 passed)
## Self-Review Checklist:
- [x] I have reviewed my own diff for quality, security, and reliability
- [ ] Unsafe blocks (if any) have justifying comments
- [ ] The content adheres to Zed's UI standards
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
Release Notes:
- N/A
## Summary
- Open relative file links with GitHub-style line fragments at the
referenced line
- Support line navigation from agent responses and Markdown previews
- Preserve decoded-first path resolution with literal percent-escape
fallback
## Testing
- cargo test -q -p util test_source_line_from_fragment -- --nocapture
- cargo test -q -p acp_thread test_hyperlink_percent_escapes_are_decoded
-- --nocapture
- cargo test -q -p agent_ui test_open_link -- --nocapture
- cargo test -q -p markdown_preview -- --nocapture
- cargo fmt --all -- --check
Release Notes:
- Improved relative file links with line numbers to open at the
referenced line.
Tiny tiny tweak just adjusting the thinking icon color so its consistent
with the fast mode icon. Icons that are _toggled_ on and off should be
displayed in a colorful way so its obvious in which state they're in.
---
Release Notes:
- N/A
# Objective
Fixes#61391Fixes#60876
Fix Agent Panel prompt text and controls being clipped when the panel is
narrow.
## Solution
- Allow the message editor and ACP configuration controls to shrink and
wrap.
- Use a dynamic minimum panel width based on the active Agent's prompt
text.
- Cap that dynamic width so long prompts, such as Zed Agent's, do not
prevent the panel from being resized narrower.
## Testing
- `cargo fmt --all -- --check`
- `cargo check -p agent_ui`
- `cargo test -p agent_ui --lib
conversation_view::tests::test_restored_threads_keep_available_commands`
- `cargo test -p agent_ui --lib config_options::tests`
Manual UI testing is still recommended for Zed Agent and third-party
Agent Panels at narrow widths.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
- [ ] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
## Showcase
https://github.com/user-attachments/assets/71825d64-6e5d-40fe-98a4-dcb13903e06b
---
Release Notes:
- Agent: Fixed agent panel placeholder and controls overflowing when the
panel width was narrow.
---------
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
Follow up to https://github.com/zed-industries/zed/pull/61245. That PR
introduced a bug where the thread controls would appear at seemingly
random tool boundaries as opposed to only at the end of each turn. It
caused this sort of behavior, where we'd have many instances of those
buttons showing up:
<img width="500" alt="Screenshot 2026-07-20 at 4 03 2@2x"
src="https://github.com/user-attachments/assets/6e6f716e-adb5-4532-9f89-e955398cda23"
/>
Alongside with fixing it, this PR is also making the "scroll to top"
only show up only at the very last turn. I think that makes more sense
given you'd be most likely interested to go to the top when you're at
the bottom, rather than at every turn, but I could be persuaded
otherwise; not holding that too strongly.
Release Notes:
- N/A
The diff already has a constant-time accessor to get the counts of
added/removed rows, use that instead of iterating hunks on the
foreground.
Release Notes:
- Improved performance in agent threads with many file changes.
# Objective
Fixes#61244
In the Agent Panel, after a multi-turn conversation, the action buttons
("Copy this Agent response", "Scroll to User Message", and "Scroll to
top") were only shown after the last assistant response. This was
inconvenient for users who wanted to copy or navigate from earlier
responses.
Additionally, the "Scroll to User Message" button always jumped to the
most recent user message in the entire thread, even when clicked from a
historical assistant response.
This PR addresses both issues by rendering the controls after every
non-blank assistant response, and by making "Scroll to User Message"
jump to the user prompt associated with that specific response.
## Solution
- Refactored `render_thread_controls` in
`crates/agent_ui/src/conversation_view/thread_view.rs` to accept an
explicit `entry_ix`, `copy_response_index`, `is_thread_bottom`, and
`user_message_index`, making it reusable per assistant message.
- Rendered the control bar after each non-blank `AssistantMessage`
entry, not just the last entry.
- Made button IDs unique by using tuple IDs like `("name", entry_ix)`,
matching the existing style in this file and avoiding `format!`
allocations.
- Keyed the copy button ID by the control bar location (`entry_ix`)
while still copying `response_index`'s content, so the bottom fallback
bar doesn't duplicate IDs with the per-response bar.
- Gated the generation/confirmation early return to `is_thread_bottom`,
so historical controls remain accessible while the thread is generating
or waiting for confirmation.
- Gated turn stats and thread feedback buttons to the bottom of the
thread so they only appear on the last response.
- Replaced `scroll_to_most_recent_user_prompt` with a more general
`scroll_to_user_message_index` that takes an optional target user
message index.
- For each per-response button, computed the associated user message
index as the most recent user message before that assistant response.
- For the bottom fallback (when the last entry is not an assistant
message), passed `None` so the button still scrolls to the most recent
user message.
- Updated the tooltip from "Scroll to Most Recent User Message" to
"Scroll to User Message".
- Updated existing tests and added assertions for
`scroll_to_user_message_index(Some(ix), cx)`.
## Testing
- `cargo check -p agent_ui`
- `./script/clippy -p agent_ui`
- `cargo test -p agent_ui --lib` (399 passed)
- Added assertions in `test_scroll_to_most_recent_user_prompt` to verify
`scroll_to_user_message_index(Some(0), cx)` and
`scroll_to_user_message_index(Some(2), cx)` scroll to the correct user
message indices.
- Manual testing recommended: open the Agent Panel, send multiple
messages, and verify that each assistant response has the three buttons
and that "Scroll to User Message" jumps to the corresponding user
prompt.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments (no unsafe blocks)
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
## Showcase
Before: Only the last assistant response in the Agent Panel showed the
"Copy this Agent response", "Scroll to User Message", and "Scroll to
top" buttons. Clicking "Scroll to User Message" always jumped to the
most recent user message.
After: Every non-blank assistant response displays these buttons.
Clicking "Scroll to User Message" on a historical response scrolls to
the user prompt that triggered that specific response.
https://github.com/user-attachments/assets/46e02bcc-8bed-4cf3-a08a-9ef91a206d08
---
Release Notes:
- Added "Copy this Agent response", "Scroll to User Message", and
"Scroll to top" buttons to every agent response in the Agent Panel.
- Fixed "Scroll to User Message" so it jumps to the user prompt
associated with the current response, rather than the most recent user
message.
The agent panel's context usage popover showed "1 project rules" when
exactly one project rule (e.g. `.rules`) was loaded, because the
project-rules button used an unconditional `format!("{} project rules",
...)`.
Reuse the existing `pluralize` helper (already used for diagnostics
labels in the same crate) so a single rule renders as "1 project rule"
and multiple rules render as "N project rules". The sibling "1 global
rule" label is intentionally left unchanged, since there is only ever
one global rule.
Closes#57759
Release Notes:
- Fixed the agent panel context usage popover showing "1 project rules"
instead of "1 project rule" when a single project rule is loaded.
# Objective
ACP boolean config option labels use the default text color, while
select option labels use the muted text color. This makes boolean
options such as Fast Mode look visually inconsistent with adjacent
configuration controls.
## Solution
Add a label color override to `Switch` while retaining `Color::Default`
for existing callers. Apply `Color::Muted` only when `agent_ui` renders
ACP boolean config options, leaving all other switches unchanged.
## Testing
Manually verified on macOS in the light theme.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior (verified visually)
- [x] Performance impact has been considered and is acceptable
## Showcase
### Before
<img width="449" height="37" alt="image"
src="https://github.com/user-attachments/assets/1abe3e39-b243-47b3-8a62-67a0b5a17b43"
/>
### After
<img width="445" height="39" alt="image"
src="https://github.com/user-attachments/assets/e134378c-84cb-4c72-92d1-98eecc2137e4"
/>
---
Release Notes:
- Fixed ACP boolean option labels to match other agent configuration
controls.
This is necessary to remove some `util` dependencies from crates, as
well as better sharing for our projects. This also includes the WIP
AbsPath abstraction as well as some bug fixes from internal tooling.
Release Notes:
- N/A or Added/Fixed/Improved ...
Started an old Zed version with old db entry which resulted in 0 agent
panel working, only showing me there and in logs "project path not
found".
Fixed by removing the corresponding `db` subdirectory, but decided to
make the log entries a bit more useful.
Release Notes:
- N/A
This PR introduces a component for the terminal tool call's header. It
was growing already pretty big in the terminal tool call render function
in the thread view and hard to visualize in all of its use cases. I got
motivated to do this because I wanted to make the UI for the case when
it runs outside of the sandbox simpler, but it wasn't easy to make that
show up in a true agent thread, and with it being a component, I can
just use the component preview.
Here's a quick screenshot of it, it shouldn't look too different, just a
bit more polished:
<img width="600" alt="Screenshot 2026-07-13 at 6 43@2x"
src="https://github.com/user-attachments/assets/fb222e50-2003-4cf7-9039-7b211db3c86e"
/>
Release Notes:
- N/A
This PR reworks how multi-select mode is rendered in pickers (only used
by the file finder and text finder).
The "Multi Select" entry previously lived inside the footer's Actions
menu, which was hard to discover and its toggle checkmark misaligned the
menu's keybinding column. This PR changes to an icon button at the
far-roght edge of the search editor, with a tooltip showing the
keybinding to toggle it.
Also made each list item use the actual Checkbox component instead of a
bespoke re-implementation of it. And in doing so, added a few design
improvements to the list item so that it received the checkbox while
preserving proper styles for each interaction state, as well as
displaying the keybinding to select the item or check the item.
Here's a quick video, showing these changes off:
https://github.com/user-attachments/assets/e142f7d1-87ba-4258-844b-953ed06237bd
Release Notes:
- Improved multi-select in the file finder and text finder: the toggle
now lives in the search bar with a `cmd-shift-s` keybinding, and
selection checkboxes render inside list items.
Release Notes:
- On agent notifications, the corresponding zed window will now mark
itself for OS level attention
---------
Co-authored-by: zed-zippy[bot] <234243425+zed-zippy[bot]@users.noreply.github.com>
Closes security loopholes and updates docs:
- installs seccomp filter for blocking naughty syscalls
- tightens macos seatbelt profile
- fetch tool responses that redirect are now constrained by allowed
domains list
Also adds a few "Learn More" buttons that link to the new docs.
Also fixes a bug where the agent would try to create a
`~/.config/zed/AGENTS.md` directory
Also adds unicode confusable detection to URL/path privilege escalation
prompts.
---
Release Notes:
- N/A or Added/Fixed/Improved ...
---------
Co-authored-by: zed-zippy[bot] <234243425+zed-zippy[bot]@users.noreply.github.com>
Turns out I only tested it with all tabs closed, the keybinding would
only work when all tabs were closed and you focused the agent panel. If
a file was opened in the center pane, cmd-f would open a search in that
file instead of searching in the agent panel, even if the thread view as
focused.
Closes#60686
Release Notes:
- agent: Fixed an issue where cmd-f would not work if file is open in
center pane
Removes the feature flag. The RFD is in Preview and I am confident it
will be stable without major changes by the time this hits Zed Stable.
Release Notes:
- acp: Allow ACP agents to use Elicitation capturing structure user
input.
# Objective
- Fixes#58097.
- Opening a new project and clicking `+` in the agent panel to start a
terminal thread creates two terminals instead of one. I am able to
replicate the issue on version 1.8.0 on macOS 27
## Solution
The new-thread action creates the terminal and then focuses the agent
panel. Focusing re-activates the panel (`Panel::set_active` then
`ensure_thread_initialized`) before the terminal, which is spawned
asynchronously, has registered. The panel still looks uninitialized at
that moment, so it spawns its own "initial" terminal too, and that is
the duplicate.
`spawn_terminal` now marks the spawn as in-flight
(`pending_terminal_spawn`) the moment it starts, the same way the
restore and initial-terminal paths already do, so the existing guard in
`ensure_thread_initialized` skips the redundant terminal.
This only affected new (unrestored) projects, since existing ones
restore their previous entry instead of auto-creating one. The auto-init
behavior was introduced in #57150.
## Testing
- Verified in a local dev build on macOS: opening a fresh project and
clicking `+` now creates one terminal, and clicking `+` again creates a
second, as expected. Reopening an existing project still restores a
single terminal.
- Added `test_explicit_terminal_blocks_redundant_auto_init`, which fails
without the fix.
- The change is platform-agnostic (no platform-specific code); I wasn't
able to test on Linux/Windows.
## Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content adheres to Zed's UI standards
([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
and
[icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md)
guidelines)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable
---
Release Notes:
- Fixed a duplicate terminal being created when starting an agent
terminal thread in a new project
This makes them more consistent with regards to limiting the label size.
These two tabs are similar because they frequently house a label that's
pretty big, and the agent diff tab didn't truncate it in any way.
Release Notes:
- N/A
# Objective
Allow zed's language model stack to express OpenAI Responses API custom
tools (freeform text-input tools with an optional lark/regex grammar),
so downstream consumers can offer tools like a freeform `apply_patch` to
GPT models.
## Solution
- `LanguageModelRequestTool` now carries a Function-vs-Custom input
variant; `LanguageModelCustomToolFormat` models text/grammar formats.
- `LanguageModelToolUse.input` becomes a typed
`LanguageModelToolUseInput::{Json, Text}`. Serialization is tagged so
persisted Text inputs round-trip losslessly; legacy plain JSON values
still deserialize as `Json`.
- `open_ai` gains the custom tool wire types (tool definition,
`custom_tool_call`/`custom_tool_call_output` input items with
string-or-content-part outputs, output item, and
`custom_tool_call_input` delta/done stream events). The Responses event
mapper accumulates raw text deltas into `ToolUse` events, and history
replay derives custom-vs-function tool results from the matching
`ToolUse` by id.
- All non-OpenAI providers and the Chat Completions path error
explicitly when a request contains custom tools — no silent drops or
empty-schema coercions.
Release Notes:
- N/A
