Dynamic schemas (settings, settings/lsp/*, debug_tasks) are now cached
after first generation and returned immediately via Task::ready() on
subsequent requests.
Cache is invalidated when `notify_schema_changed()` is called, which
happens when:
- Extensions are installed/uninstalled (affects settings schema)
- DAP registry changes (affects debug_tasks schema)
This eliminates repeated foreground thread blocking for the same dynamic
schema.
**Note:** This PR is based on #46823 and should be merged after it.
Release Notes:
- N/A
---------
Co-authored-by: Ben Kunkle <ben@zed.dev>
Release Notes:
- Opening bundled files, keymap, and local release notes now opens in
remote windows instead of opening a new local zed window
- Opening the settings files, keymap files, task files, debug files and
logs will now open within wsl windows instead of opening a new local zed
window
Although `poetry` may store virtual environments in a global cache
directory (unlike the standard project-local `.venv`), the internal
directory structure of these environments remains consistent with
standard `venv` or `uv` environments.
Since our existing activation logic for `venv` environments relies on
relative paths within the environment root, it is naturally compatible
with `poetry` environments. Once the toolchain locates the environment's
root directory, the activation scripts can be applied without further
modification.
Testing:
- Verified on Windows: Confirmed working across `PowerShell`, `Pwsh`,
and `Cmd` for global poetry environments.
- Observation: When using an in-project `.venv` directory, the toolchain
may categorize the environment as `venv`-managed instead of
`poetry`-managed. While this behavior relates to the toolchain's
discovery logic, the activation itself remains fully functional as the
underlying structure is compatible.
Release Notes:
- N/A *or* Added/Fixed/Improved ...
Context menus close when the window becomes inactive (e.g., clicking to
another window or app). Previously, menus would remain visible until the
window regained focus.
This aligns Zed's context menu behavior with native macOS apps, where
NSMenu automatically dismisses when the window loses focus. Users expect
menus to close when switching windows or apps.
Uses `observe_window_activation` to detect when the window becomes
inactive and calls `cancel` to dismiss the menu.
Release Notes:
- Fixed context menus dismiss/cancel when the window loses focus
---------
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
Closes#46142
Release Notes:
- Fixed AI commit message button in Git panel being hidden instead of
disabled when no AI provider is configured. The button now remains
visible with a tooltip explaining how to enable the feature.
Screenshot:
<img width="90%" alt="Screenshot 2026-01-10 at 18 16 12"
src="https://github.com/user-attachments/assets/6af9c845-7857-40e1-bfb5-3118f1af8360"
/>
---------
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
This PR fixes an issue when copilot takes 3s to complete. Right now, in
copilot edit prediction, we issue requests to both the Next Edit
Suggestion (NES) and the regular copilot inline completion endpoints.
Whichever come back first will be shown. However, there is a bug where
even if inline completion (which is usually much faster) comes back, we
still wait for NES (which takes about 3s).
This should address https://github.com/zed-industries/zed/issues/46389
and https://github.com/zed-industries/zed/issues/46880
Release notes:
- Improved responsiveness of Copilot inline completions.
It should be a no-op but causes massive main thread hangs on windows due
to the winapi calls this emits, restores the behavior pre
github.com/zed-industries/zed/pull/46256 without reverting the main
changes of that.
Release Notes:
- N/A *or* Added/Fixed/Improved ...
The cursor in terminal vi mode was not visually updating when using
vi motion navigation keys, despite internal position tracking working correctly.
This was caused by missing `cx.notify()` calls to trigger re-rendering.
Refactored keystroke handling by extracting common logic from `key_down`
and `send_keystroke` into a shared `process_keystroke` helper method.
Closes#46736
Release Notes:
- terminal: Fixed vi mode cursor not visually updating when navigating
with vi motion keys
---------
Co-authored-by: dino <dinojoaocosta@gmail.com>
Static schemas (tasks, snippets, jsonc, keymap, action/*, tsconfig,
package_json, inspector_style) are now computed once on first access
using LazyLock and returned immediately via Task::ready().
These schemas never change at runtime because:
- tasks, snippets, jsonc, inspector_style are derived from static Rust
types
- tsconfig, package_json are bundled JSON files (include_str!)
- keymap and action/* depend only on registered actions, which are
collected at compile/link time via the inventory crate (extensions
cannot add actions)
This eliminates foreground thread blocking for these schema requests.
**New functions added to KeymapFile:**
- `generate_json_schema_from_inventory()`: generates keymap schema
without App context
- `get_action_schema_by_name()`: looks up single action schema without
App context
Release Notes:
- N/A
---------
Co-authored-by: Ben Kunkle <ben@zed.dev>
This feature cost $15.
Up -> Tokens we're sending to the model
Down -> Tokens we've received from the model.
<img width="377" height="69" alt="Screenshot 2026-01-14 at 12 31 01 PM"
src="https://github.com/user-attachments/assets/fc15824f-de5d-466b-8cc1-329f3c1940bb"
/>
Release Notes:
- Changed the display of tokens for OpenAI models to reflect the
input/output limits.
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
This PR adds a list of recent projects to the multi-project dropdown,
avoiding the need to open up the picker and the mouse travel when using
a pointer to interact with it. When the recent projects list is bigger
than 5, we display more in a "View More" submenu.
<img width="500" height="800" alt="Screenshot 2026-01-14 at 5 15@2x"
src="https://github.com/user-attachments/assets/31f11e3e-d010-4b66-a539-f74cbc572e40"
/>
Release Notes:
- Workspace: Added the list of recent projects to the multi-project
title bar menu.
Closes https://github.com/zed-industries/zed/issues/34402
Release Notes:
- MCP servers can now be run on the remote server when using remote
development. This can be enabled by setting the `"remote": true`
property in the settings entry for the MCP server.
---------
Co-authored-by: localcc <kate@zed.dev>
Co-authored-by: Lukas Wirth <me@lukaswirth.dev>
## Summary
This PR adds the UI for displaying subagent tool calls:
### Thread view changes
- Add `expanded_subagents` state HashMap for tracking expanded cards
- Implement `render_subagent_tool_call()` for collapsed card with label
and chevron
- Add subagent detection via `is_subagent()` and `tool_name` field
- Handle `SubagentThread` content type in tool call rendering
- Add expand/collapse toggle button for subagent cards
- Style collapsed cards similar to terminal tool calls
- Support inline image rendering in content blocks
### Agent panel changes
- Add `open_external_thread_with_server()` for testing with stubbed
servers
### Test support
- Add `acp_thread/test-support` feature to agent_ui
- Add base64 dev dependency for image tests
Release Notes:
- N/A
---------
Co-authored-by: Amp <amp@ampcode.com>
This refactors the "Panel Button Clicked" telemetry event to fire from
`workspace.rs` instead of `dock.rs`.
Previously, the event was emitted in the `PanelButtons` render method's
click handler. Now it fires at the workspace level in two places:
- `toggle_dock()` - captures panel toggles via dock buttons
- `focus_panel<T>()` - captures panel focus via keyboard shortcuts
This ensures the telemetry fires once per user action regardless of
input method, and retrieves the panel name dynamically via
`persistent_name()` rather than relying on a pre-computed local
variable.
Release Notes:
- N/A
Doesn't support editing or deleting comments yet, but inputs work:
<img width="797" height="881" alt="Screenshot 2026-01-14 at 11 51 36 AM"
src="https://github.com/user-attachments/assets/746e085b-a919-475a-b804-ee0377e40a0b"
/>
<img width="698" height="178" alt="Screenshot 2026-01-14 at 11 47 51 AM"
src="https://github.com/user-attachments/assets/3edac0e8-62e5-4af5-a324-df2bd293f3ca"
/>
<img width="695" height="695" alt="Screenshot 2026-01-14 at 11 47 58 AM"
src="https://github.com/user-attachments/assets/64795ffd-62b4-48ea-a46c-7724221095d7"
/>
### Features implemented:
- **Local comment storage**: Comments are stored per-hunk in the Editor,
with chronological ordering
- **Expandable comments section**: Shows "N Comments" header that can
expand/collapse
- **Inline editing**: Click edit to transform a comment row into an
editable text field with confirm/cancel buttons
- **Delete functionality**: Remove individual comments
- **Send Review to Agent**: Toolbar button with badge showing comment
count, batch-submits all comments across all files/hunks to the Agent
panel
- **User avatars**: Shows the current user's avatar (from their Zed
account) next to comments, falls back to Person icon when not logged in
- **Visual tests**: Added tests for one comment, multiple comments
expanded, and comments collapsed
Note that this is behind a feature flag, so no release notes yet.
Release Notes:
- N/A
---------
Co-authored-by: Zed Zippy <234243425+zed-zippy[bot]@users.noreply.github.com>
Co-authored-by: David Baldwin <baldwindavid@gmail.com>
Release Notes:
- Agent: Added a menu item in the agent panel right-click menu for
copying a given agent's response.
---------
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
## Problem
Profiling showed that during agent streaming, `thread.rs:1393:23` was
appearing constantly as a hotspot. The issue was that every single token
from the model triggered:
1. `this.update(cx, ...)`
2. `handle_text_event()` (or thinking/redacted_thinking)
3. `cx.notify()`
4. UI re-render
This created significant foreground thread pressure, contributing to
~500ms delays visible in the profiler.
## Solution
Batch all immediately-available events using `now_or_never()` and
process them in a single `update()` call with one `notify()` at the end.
This approach is deterministic - it processes exactly what's available
right now, adapting naturally to network speed:
- When tokens arrive slowly, you get one at a time
- When they arrive in bursts, they batch automatically
## Changes
- Remove `cx.notify()` from `handle_text_event`,
`handle_thinking_event`, and `handle_redacted_thinking_event`
- Batch events in the streaming loop using `now_or_never()`
- Call `cx.notify()` once per batch instead of per-event
- Keep `cx.notify()` in `handle_tool_use_event` for immediate tool
feedback
Release Notes:
- Improve streaming tool call performance by batching UI updates.
Closes#45755
Rainbow brackets in HTML files do not really make sense - a tag is
always enclosed in angled brackets. With our current approach, that
means that we will highlight all brackets in HTML files with the same
color, because well, these angled brackets do not quite behave like in
other languages.
Hence, let themes color this again and just exluce HTML angled brackets
from rainbow bracket colorization, as it is no real rainbow bracket
colorization anyway.
Release Notes:
- N/A
Closes#40322
When users run multi-line terminal commands, the confirmation dialog was
only showing the first line with '- N more lines' truncated. This meant
dangerous commands like 'rm -rf' or 'sudo' operations could be hidden
from view, creating a security risk.
Now terminal commands always display in full, with scroll support for
very long commands (20+ lines) to keep the confirmation buttons visible.
Other tools continue using truncation for better UI efficiency since
they don't pose the same security concerns.
Release Notes:
- Agent: Fixed terminal command truncation in the agent panel to better
expose long commands (e.g., potentially dangerous multi-line commands
are now fully visible before execution)
---------
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
Co-authored-by: dino <dinojoaocosta@gmail.com>
This PR makes queueing a prompt the default behavior when sending a new
one while there's an on-going generation. You can still send a prompt
that will immediately interrupt the agent with the `cmd-shift-enter`
keybinding, though, which preserves the current behavior.
The main motivation for this change is to make the queueing not only
more discoverable, but more useful as well, as we're parting from the
assumption that most of the time, what you want is to queue it as
opposed to interrupting it (even though it's still possible to do either
through the keybinding I mentioned above or simply by stopping the
generation and sending a new one).
Here's a quick video:
https://github.com/user-attachments/assets/37f92433-70ef-459f-98ff-41ed80e3e43f
In the video, I show sending one prompt and then sending two others that
fall straight into the queue. Then, in the middle of the generation of
my first prompt, I use the `cmd-shift-enter` keybinding to send a prompt
immediately, interrupting the agent, effectively being sent in front of
the queue.
Release Notes:
- Agent: Made queueing prompts the default behavior when sending them
while there's an on-going generation.
- **copilot: Fix double lease panic when signing out**
- **Extract copilot_chat into a separate crate**
- **Do not use re-exports from copilot**
- **Use new SignIn API**
- **Extract copilot_ui out of copilot**
Closes#7501
Release Notes:
- Fixed Copilot providing suggestions from different Zed windows.
- Copilot edit predictions now support jumping to unresolved
diagnostics.
Feature flagged for now as we test this out with actual agents to see if
we need to provide any more feedback to the RFD before committing to the
current setup.
Release Notes:
- N/A
This makes sure that all of the work we do for caching/refreshing
session history is reused everywhere we need to access the list.
Very important for external agents so we aren't rerequesting history on
every search or recent list update. Puts the reloading of history in one
place
Release Notes:
- N/A
Fixes subword motion incorrectly jumping to next line when near end of
line. Updates boundary detection to use exclusive boundaries with
need_next_char parameter, matching regular word motion behavior.
Refactors word and subword motion to share boundary detection logic.
Closes#17780
Release Notes:
- Fixed subword motion incorrectly jumping to the next line when near
the end of a line
---------
Co-authored-by: dino <dinojoaocosta@gmail.com>
This reverts commit 94faaebfec.
The logic changed in the [original
PR](https://github.com/zed-industries/zed/pull/45969) is either
misplaced or lacks a counterpart that reacts on `gpui::Image` drop one
way or another.
The change was dropping the texture out of the global rendering atlas
when an image preview tab was disposed of, while in reality, another
instance (agent panel or another image preview tab) could require the
very same atlas entry to be rendered meanwhile.
Currently, only `RetainAllImageCache` in Zed does any kind of image
cleanup, and any other image usages will leak memory.
What makes it harder is that the atlas entry needs to live as long as a
particular `Arc<Image>` lives, and some public `gpui` API expects this
type to stay:
https://github.com/zed-industries/zed/blob/e747cfc955e8cfd9327d8d6b8d617cf1d3a6bcaa/crates/gpui/src/platform.rs#L1851-L1866
For image viewer in particular, we need to consider why
`RetainAllImageCache` is not used there; for the global, normal fix, we
need to consider ways to have `cx` and `window` and a way to react on
`Image` drop.
As an option, we could break the `gpui` API (as it [happens
periodically](https://github.com/zed-industries/zed/issues/46183)
already) and use `Entity<Image>` instead of `Arc`, then react with
`cx.on_release_in` for each such image.
Closes https://github.com/zed-industries/zed/issues/46755
Closes https://github.com/zed-industries/zed/issues/46435
Release Notes:
- Fixed panics on concurrent image handling
Release Notes:
- When copying and pasting from a multibuffer view (e.g. into the agent
panel), the line numbers from the selection corresponded to the line
numbers in the multibuffer view (incorrect), instead of the actual line
numbers from the file itself.
- I also handled the case in which the selection spans multiple excerpts
(different files in the multibuffer) by just returning None for that
case, but maybe it should instead be split into two selections?
Left is before (bugged, should be lines 8:16, instead it is 38:46),
right is after (fixed).
<div style="display:flex; gap:12px;">
<img
src="https://github.com/user-attachments/assets/b1bfce1d-8b6a-41c0-ac7e-51f7bd7b284e"
width="48%" />
<img
src="https://github.com/user-attachments/assets/2a4c33a0-a969-4a3e-9aa5-d2c2fefba3b2"
width="48%" />
</div>
