### Closes
- Maybe https://github.com/zed-industries/zed/issues/38252 (there might
be something going on with NPX too)
- Also addresses
https://github.com/zed-industries/zed/pull/39021#issuecomment-3644818347
### Why
Some more involved MCP servers timeout often, especially on first setup.
I got tired of having to set timeouts manually per server. I also
noticed a timeout could not be set for http context servers, which
causes Context7 or GitHub's remote servers to timeout at 60s sometimes.
### Overview
MCP Timeout Configuration Feature
This PR adds additional configurable timeout settings for Model Context
Protocol servers including a global timeout and the addition of timeouts
for http servers, addressing issues where servers were timing out after
a fixed 60 seconds regardless of user needs.
**Key Features:**
- **Global timeout setting** (`context_server_timeout`) - default
timeout for all MCP servers (default: 60s, max: 10min)
- **Per-server timeout overrides** - individual servers can specify
custom timeouts via `timeout` field
- **Precedence hierarchy** - per-server timeout > global timeout >
default (60s)
- **Automatic bounds checking** - enforces 10-minute maximum to prevent
resource exhaustion
- **Support for both transports** - works with stdio and HTTP-based
context servers
- **Comprehensive test coverage** - 3 new tests validating global,
override, and stdio timeout behavior
- **Full backward compatibility** - existing configurations work
unchanged with sensible defaults
### Release Notes:
- Added the ability to configure timeouts for context server tool calls.
The new global `context_server_timeout` setting controls the default
timeout (default is 60s, max: 10min). Additionally, per-server timeouts
can be configured using the `timeout` field within servers defined in
the `"context_servers" setting
---------
Co-authored-by: Ben Kunkle <ben@zed.dev>
Embeddings have neither been used nor maintained in over a year and
there are currently no plans to use these again any time soon. Hence,
remove support for these here to more clearly indicate that these are
actually not used.
Release Notes:
- N/A
Add support for OpenAI's /responses endpoint for models that don't
support /chat/completions API. This enables compatibility with newer
model variants (`gpt-5-codex`, `gpt-5-pro`, `o3-pro`, etc) while
maintaining compatibility with existing configs
Changes:
- Add `supports_chat_completions` flag to model capabilities that
defaults to true for existing behavior
- Implement responses API client with streaming support as per [OpenAI
documentation](https://app.stainless.com/api/spec/documented/openai/openapi.documented.yml).
- Add `ResponseEventMapper` to convert responses events to completion
events for maintainer simplicity
- Update UI to allow toggling `chat_completions` capability
- Add `gpt-5-codex` model
Closes#38858
Release Notes:
- Added support for `gpt-5-codex` model
---------
Co-authored-by: Bennet Bo Fenner <bennet@zed.dev>
Closes#42217
Release Notes:
- Added/Fixed/Improved ...
added "Create Pull Request" Command when searching on the command
Palette
For more details, please refer to the issue, thank you.
---------
Co-authored-by: dino <dinojoaocosta@gmail.com>
Currently, the Python toolchain can identify `uv` and `uv-workspace`
environments (showing the correct labels in the UI), but it fails to
activate them when opening a new terminal. This is because the
activation script resolution and command generation logic were missing
for these environment kinds.
This PR adds `uv` and `uv-workspace` to the standard virtual environment
activation flow. Since `uv` creates environments with a standard `venv`
structure, we can reuse the existing `resolve_venv_activation_scripts`
logic to find and execute the appropriate `activate` scripts for
different shells.
Release Notes:
- Fixed terminal activation for `uv` and `uv-workspace` Python
environments.
## Problem
When clicking Stop button on a terminal tool call (or using terminal
kill bindings), only the shell process was killed. Child processes (like
`sleep`, `npm run`, etc.) continued running as orphans.
## Solution
On Unix, use `killpg()` instead of `sysinfo::Process::kill()` to
terminate the entire foreground process group.
`tcgetpgrp()` already returns the foreground process group ID, so
`killpg()` is the correct syscall to use here.
## Testing
1. Run a long command in terminal tool (e.g. `sleep 30`)
2. Click Stop button
3. Verify with `ps aux | grep sleep` that the process is actually killed
## Notes
- This fix is Unix-only (Linux, macOS)
- Windows behavior unchanged — may need separate investigation with Job
Objects
## Release Notes
- Fixed terminal Stop button not killing child processes on Unix (Linux,
macOS)
Closes https://github.com/4teapo/zed-luau/issues/31
Luau `.luaurc` config files contain JSON but that allows comments and
trailing commas, making them a perfect fit for JSONC.
Release Notes:
- N/A
We use `block_with_timeout` to give the reparse task a millisecond to
complete, and if it takes longer we put the work off to the background.
The reason for this is that we want tree-sitter based features to feel
snappy.
The reparse task is non-cooperative though, it has no yield points,
giving us no place to actually check for our timeout, meaning we will
always drive it to completion and block for the entire duration.
This kicks out the `block_with_timeout` in favor of using the treesitter
progress callback to handle timeouts instead.
Release Notes:
- Improved responsiveness with very language language files on edits
Closes https://github.com/zed-industries/zed/issues/44091
The issue here was that we were missing a proper default for
`ThemeSelection`. This is a disadvantage of not having the defaults in
code but actually in the `default.json`, so this here basically copies
the default from the `default.json` into code for the settings UI to
work properly.
Release Notes:
- Fixed an issue where the theme was not configurable from the settings
UI if no theme was configured prior.
Generates a training or evaluation example from a
chronologically-ordered commit. This is a port from the Python codebase
(except for the reorder_patch.rs, which was originally written in Rust
in).
Release Notes:
- N/A
Closes#45251
In https://github.com/zed-industries/zed/pull/44532 the syntax
highlighting of JS/TS was improved, but we only used `type.builtin` for
a specific case. This resulted in less consistent syntax highlighting.
Not all themes have this specific token and we don't use `type.builtin`
in other places either, so while this is a little less specific it is
more consistent.
Release Notes:
- Improved syntax highlighting consistency for builtins in JavaScript
and TypeScript
Closes#44904
Adds the Tailwind CSS LSP as suggested here:
https://github.com/zed-industries/zed/pull/39517#issuecomment-3368206678.
This allows users to use the Tailwind CSS LSP for CSS files, to support
Tailwind specific rules.
Docs will be added in a separate PR after this is released
Release Notes:
- Add Tailwind CSS Mode support, for Tailwind specific CSS rules like
`@apply`, `@layer`, and `@theme`.
When adding selections above/below with `skip_soft_wrap: true`, use
buffer column positions instead of pixel positions. This ensures
selections are placed at the same column offset in the buffer rather
than at the same visual position, which was incorrect for soft-wrapped
lines.
For example, selecting "how" in a wrapped line:
````
1. Very long line to
show [how] a wrapped
line would look
2. Very long line to
show how a wrapped
line would look
````
Now correctly adds a selection at the same buffer column in the next
line:
````
1. Very long line to
show [how] a wrapped
line would look
2. Very long line to
show [how] a wrapped
line would look
````
Closes#42137
Release Notes:
- Improved `editor: add selection above` and `editor: add selection
below` to select at the same text column when skipping soft-wrapped
lines, instead of the same visual position
Closes#39914
I was able to reproduce the issue on macOS using the same README.
The root cause was that `markdown_preview_view` attaches the image cache
to the root div, and while Div correctly applies that cache during
request_layout and paint, the prepaint phase didn’t have the same
context.
Markdown is rendered through a List, and `List::prepaint` renders its
items using `layout_as_root`. That meant images were getting loaded
during prepaint without the image cache in place. When that happened,
Img fell back to the global asset loader, which retains assets
indefinitely unless explicitly cleaned up.
So every image was loaded twice:
- once during layout/paint via RetainAllImageCache (correctly released
when the preview closed)
- once during prepaint via the global asset system (never released)
The result was doubled memory usage and a leak, since the globally
loaded images stuck around after the preview was closed.
Release Notes:
- Fixed a memory leak when previewing markdown files with images
Signed-off-by: Marco Mihai Condrache <52580954+marcocondrache@users.noreply.github.com>
This is in preparation for removing `Result<T>` from the `AsyncApp`
methods
Refactor machine goes brrrrrr
Plan and tracker for this PR:
https://gist.github.com/mikayla-maki/7dfc0d4907e76de119b5712e24665f02
This PR should be safe to merge, as I cannot observe any changes in
behavior from adding this to our application.
Release Notes:
- N/A
Follow-up to #46001
That initial fix partly addressed the issue for diffs managed by the
`GitStore`, but not for other diffs (e.g. those managed by the
`ActionLog` or `CommitView`). The underlying issue is that we switched
to using a `Buffer` to represent the diff base text, and when updating
the diff we were calling `set_text` on this buffer and not waiting for
reparsing to finish. When the base text was represented by a series of
independent `BufferSnapshot`s, this wasn't an issue because we would
parse the base text in the background as part of computing the diff
update. This PR fixes the issue by waiting on reparsing to finish after
each call to `set_text`.
Release Notes:
- N/A
This PR essentially removes the logic I introduced in
https://github.com/zed-industries/zed/pull/45361 to position workspace
modals based on the pointer click coordinates. All of this code has
become unnecessary given in
https://github.com/zed-industries/zed/pull/45924 we made the remote,
project, and branch modals use popovers when triggered with a pointer,
which already handle all the anchored positioning based on the trigger
(much better).
Release Notes:
- N/A
This PR removes the ability to retrieve a Supermaven API key from
Collab.
This was staff-gated (at least, on the server), and judging by the logs,
no one is using it.
Release Notes:
- N/A
When working on https://github.com/zed-industries/zed/pull/45969 I've
noticed that whenever I try to open a binary file
```
~/Desktop/svenska ❯ du -ha "Svenska А2B1. 07.09.2025 [u6qEIe9-COc].mkv"
456M Svenska А2B1. 07.09.2025 [u6qEIe9-COc].mkv
```
Zed allocates all its size
<img width="214" height="104" alt="image"
src="https://github.com/user-attachments/assets/e67ad522-b8a4-4e8e-9961-13030a34df1c"
/>
<img width="345" height="154" alt="image"
src="https://github.com/user-attachments/assets/ea691020-0d02-4acc-88c3-476385f309bb"
/>
only to show me this:
<img width="979" height="677" alt="image"
src="https://github.com/user-attachments/assets/ff91cc9d-eb59-4cee-b06f-5758acd1bd5d"
/>
Given that our existing code checks first 1024 bytes to decide whether
to bail on a binary file or not, this seems very wasteful — hence,
adjusted the code to read the "header" and check that first, and only
continue reading the entire file after the checks are successful.
I suspect this should also help the project search, esp. the crashes and
memory usage during that when many binary files are present?
Release Notes:
- Improved Zed's memory usage when attempting to open binary files
Closes#45046
The root of the issue is anchor resolution. When we apply adjacent
edits, they get merged into a single edit. In the scenario described in
the issue, this is what happens:
1. We create an anchor at the end of each selection (bias::right) on the
snapshot before the edits.
2. We collect the edits and apply them to the buffer.
3. Since the edits are adjacent (>=), the buffer merges them into a
single edit.
4. As a result, we apply one edit to the text buffer, creating a single
visible fragment with length = 3.
5. The buffer ends up with fragments like: [F(len = 3, visible = true),
F(len = 1, visible = false), ...]
6. After the edits, we resolve the previously created anchors to produce
zero-width selections (cursors).
7. All anchors resolve into deleted fragments, so their resolved offset
equals the cumulative visible offset, which is 3.
8. We now have 3 cursors with identical coordinates (0;3).
9. These cursors get merged into a single cursor.
I tried several approaches, but they either felt wrong or didn’t work.
In particular, I tried adjusting anchor resolution using the delta
stored in handle_input, but this doesn’t help because selections are
merged immediately after anchor resolution.
The only workable solution I found is to avoid anchors entirely for the
adjacent-edit case. Instead, we can compute the final cursor positions
directly from the edits and create the selections based on that
information.
Release Notes:
- Fixed an issue where adjacent selection insert would merge cursors
---------
Signed-off-by: Marco Mihai Condrache <52580954+marcocondrache@users.noreply.github.com>
Co-authored-by: Lukas Wirth <me@lukaswirth.dev>
Just like #43747 added the keybinding(s) for `settings.json`, added them
here for keymaps and added the tooltip to show those.
<img width="500" height="282" alt="Screenshot 2026-01-04 at 2 47@2x"
src="https://github.com/user-attachments/assets/d4dbdd6e-24eb-4b7f-baa8-6e1d9810ff94"
/>
Release Notes:
- keymap editor: Added the ability to open the `keymap.json` file with a
keybinding.
---------
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
Ever since we launched the keymap editor, we've received feedback about
how "_creating_" a new keybinding was not obvious. At first, I was
confused about this feedback because you can't "create" a keybinding,
you need to rather _assign it_ to an action... so what always made sense
to me was to start with searching for the action, which is an use case
we already support. Regardless, having an easy to reach "create" button,
which essentially still asks for action > keystroke > arguments (if
needed) > and context, feels like a win UX-wise; a bit of a redundant
flow that feels ultimately positive.
So, this PR adds a "Create Keybinding" button to the keymap editor,
which you can reach with `cmd-k`. That will open up a modal with an
action autocomplete, the keystroke recording input, the arguments input
(if needed), and the context input.
https://github.com/user-attachments/assets/86f64314-4685-47bb-bb0d-72ca4c469d1f
Release Notes:
- keymap editor: Added a keybinding creation modal to make it easier to
assign an action to a keystroke.
When `trim_completion()` creates new anchors from the current buffer
state, `completion.snapshot` was not being updated, leaving it with the
older snapshot from when the prediction was initially fetched. This
caused a panic in `interpolate_edits()` when trying to resolve anchors
with Lamport timestamps newer than what the old snapshot had observed.
The fix ensures that `completion.snapshot` is updated whenever new
anchors are created in `trim_completion()`, keeping the snapshot and
anchors consistent.
Closes#45956
Release Notes:
- Fixed a panic in Copilot edit predictions caused by anchor/snapshot
version mismatch
- Added handling for DIB format, converting to BMP after adding the
corresponding file header
Release Notes:
- fix: Unable to paste images from clipboard directly into agent panel
on Windows
---------
Co-authored-by: Lukas Wirth <me@lukaswirth.dev>
This mimics VSCode's `files.readonlyExclude` setting, to allow setting
specific path matches as readonly locations like lockfiles and generated
sources etc.
Also renders a lock icon to the right side of the path names for
readonly files now.
This does a couple more things for completion sake:
- Tabs of readonly buffers now render a file lock icon
- Multibuffer buffer headers now render a file lock icon if the excerpts
buffer is readonly
- ReadWrite multibuffers now no longer allow edits to read only buffers
contained within
Release Notes:
- Added `read_only_files` setting to allow specifying glob patterns of
files that should not be editable by default
---------
Co-authored-by: Danilo Leal <daniloleal09@gmail.com>
## Summary
Fixes#44812
The hover highlight for clickable inlay hints was not covering the last
character when the label contained multi-byte UTF-8 characters (e.g.,
`→`).
## Problem
When hovering over an inlay hint like `→ app/Livewire/UserProfile.php`,
the highlight/underline stopped one character short, leaving the final
character unhighlighted.
**Root cause:** `find_hovered_hint_part()` in `hover_popover.rs` used
`part.value.chars().count()` (character count) but `InlayOffset` wraps
`MultiBufferOffset(pub usize)` which is byte-based.
For a label like `→ app/Livewire/UserProfile.php`:
- Byte length: 32 (the `→` character is 3 bytes in UTF-8)
- Character count: 30
This mismatch caused the calculated highlight range to end 2 bytes
short.
## Changes
1. **Use byte length instead of character count** (line 112):
```rust
// Before (buggy)
let part_len = part.value.chars().count();
// After (correct)
let part_len = part.value.len();
```
2. **Fix boundary condition** (line 113): Changed `>` to `>=` for
correct `[start, end)` range semantics when hovering at part boundaries.
3. **Rename variable** for clarity: `hovered_character` →
`offset_in_hint` since it's a byte offset, not a character position.
4. **Add unit test** reproducing the exact scenario from the issue with
multi-byte UTF-8 characters.
## Testing
Added `test_find_hovered_hint_part_with_multibyte_characters` which:
- Verifies the label `→ app/Livewire/UserProfile.php` has 32 bytes but
30 characters
- Tests hovering at the last byte correctly returns the full range
- Tests boundary behavior with multiple label parts containing
multi-byte characters
Release Notes:
- Fixed inlay hint hover highlight not covering the last character when
the label contains multi-byte UTF-8 characters
