Adds a new `agent.model_parameters` setting that allows the user to
specify a custom temperature for a provider AND/OR model:
```json5
"model_parameters": [
// To set parameters for all requests to OpenAI models:
{
"provider": "openai",
"temperature": 0.5
},
// To set parameters for all requests in general:
{
"temperature": 0
},
// To set parameters for a specific provider and model:
{
"provider": "zed.dev",
"model": "claude-3-7-sonnet-latest",
"temperature": 1.0
}
],
```
Release Notes:
- agent: Allow customizing temperature by provider/model
---------
Co-authored-by: Max Brunsfeld <maxbrunsfeld@gmail.com>
Co-authored-by: Marshall Bowers <git@maxdeviant.com>
This PR makes it so we only show the trial upsell in the thread view.
Fixes https://github.com/zed-industries/zed/issues/30037.
Release Notes:
- Agent Beta: Changed the trial upsell to only be visible in the thread
view.
- Languages now define their preferred debuggers in `config.toml`.
- `LanguageRegistry` now exposes language config even for languages that
are not yet loaded. This necessitated extension registry changes (we now
deserialize config.toml of all language entries when loading new
extension index), but it should be backwards compatible with the old
format. /cc @maxdeviant
Release Notes:
- N/A
---------
Co-authored-by: Anthony Eid <hello@anthonyeid.me>
Co-authored-by: Remco Smits <djsmits12@gmail.com>
Co-authored-by: Anthony <anthony@zed.dev>
This PR makes it so we don't render the trial upsell when not using the
Zed provider.
Release Notes:
- Agent Beta: Changed Zed Pro trial upsell to only be displayed when
using a model through the Zed provider.
This PR makes it so we don't render the usage callouts when not using
the Zed provider.
Release Notes:
- Agent Beta: Changed usage callouts to only be displayed when using a
model through the Zed provider.
Closes#29781
Tested this with llama3, gemma3 and qwen3.
This is a breaking change, which means after adding this code changes in
future version zed we will require atleast lmstudio >= 0.3.15. For
context why it's breaking changes check out the issue: #29781.
What this doesn't try to solve is:
* Tool calling, thinking text rendering. Will raise a seperate PR for
these as those are not required in this PR to make it work.
https://github.com/user-attachments/assets/945f9c73-6323-4a88-92e2-2219b760a249
Release Notes:
- lmstudio: Fixed Zed support for LMStudio >= v0.3.15 (breaking change -- older versions are no longer supported).
---------
Co-authored-by: Peter Tripp <peter@zed.dev>
Closes #ISSUE
Release Notes:
- Agent Beta: Fixed a bug causing "Restore Checkpoint" buttons in the
agent panel not to be rendered.
Co-authored-by: Antonio Scandurra <me@as-cii.com>
Closes https://github.com/zed-industries/zed/issues/10140
* On `menu::Cancel` action (`ESC`), close notifications, one by one, if
`Workspace` gets to handle this action.
More specific, focused items contexts (e.g. `Editor`) take priority.
* Allows to temporarily suppress notifications of this kind either by
clicking a corresponding button in the UI, or using
`workspace::SuppressNotification` action.
This might not work well out of the box for all notifications and might
require further improvement.
https://github.com/user-attachments/assets/0ea49ee6-cd21-464f-ba74-fc40f7a8dedf
Release Notes:
- Added a way to dismiss workspace notifications
This PR updates #29943 to fall back to loading agent panel settings from
the old `assistant` key if the `agent` key is not present. Edits to
these settings will also target `assistant` in this situation instead of
`agent` as before.
Release Notes:
- Agent Beta: Fixed a regression that caused the agent panel not to
load, or buttons in the agent panel not to work.
The API will return a Bad Request (with no error message) when tools
were used previously in the conversation but no tools are provided as
part of a new request.
Inserting a dummy tool seems to circumvent this error.
Release Notes:
- Fixed an error that could sometimes occur when editing using Copilot
Chat.
Co-authored-by: Bennet Bo Fenner <bennetbo@gmx.de>
This PR adds an enabled indicator in the Max Mode tooltip to show when
it is enabled:
<img width="409" alt="Screenshot 2025-05-06 at 9 49 48 AM"
src="https://github.com/user-attachments/assets/43d3f6dd-5658-467a-9df9-606ce326426a"
/>
Release Notes:
- Agent Beta: Added an indicator in the Max Mode tooltip to show when it
is enabled.
Co-authored-by: Danilo <danilo@zed.dev>
This improves the new eval scenario by ~80% (`0.29` vs `0.525`) without
decreasing performance in the other evals.
Release Notes:
- Improved the performance of the `edit_file` tool.
Closes https://github.com/zed-industries/zed/issues/27673
Closes https://github.com/zed-industries/zed/issues/29344Closes#29863
This PR fixes an issue where Zed was showing no language and `4:1` as a
line/column value on startup, as described in the linked issues. You can
actually see in the first issue that the user also experiences the same
issue as described in the second one, as his line/column value is
noticably also `4:1`.
https://github.com/user-attachments/assets/bb60e387-f4b8-4e05-80b3-4dadf1a01262
This issue arises because on assistant panel load, a new context is
created and its editor focused. However, the editor is not visible
despite having focus. The content for the editor for a new context is
`\n\n\n` and the cursor is inserted directly after that - this is where
the line:column position `4:1` comes from. For the assistant panel
editor, the language is intentionally hidden, this is why the language
is not shown on workspace load.
The issue is only present for as long as the user does not focus and
edit another editor, then that instance is focused and everything starts
to work properly again.
As this issue only arises with the old assistant panel, some staff
members were unable to reproduce in the linked issues. Once you set
`export ZED_DISABLE_STAFF=1` in your environment, you should also be
able to reproduce this issue consistently.
---
This PR fixes the issue by not creating a new context on assistant panel
load. This should not cause any regressions; every other code path I
checked creates a new context if no context is yet present.
Additionally, this also seems somewhat more reasonable, as users which
have the assistant panel disabled will never need a new context anyway,
so no context should be created.
In the following video, you can see this fixes the issue when the
assistant panel was not open the last time Zed was opened. If the panel
was open before Zed was closed, we will still properly focus the panel
and then the `4:1` will show again, which in that case is correct. The
assistant panel editor is focused and the missing language as well as
the line number then match what the user sees, experiences and expects.
https://github.com/user-attachments/assets/224a786b-52c7-4212-bccb-dff6d9db62c3
Release Notes:
- Fixed an issue where Zed would show no language and an incorrect
line/column value on startup.
Co-authored-by: Peter Tripp <peter@zed.dev>
Closes#28699
Fixes two cases in the `editor::SelectLargerSyntaxNode` action:
1. When cursor is at the end of a word, it now selects that word first
instead of selecting the whole line.
2. When cursor is at the end of a line, it now selects that line first
instead of selecting the whole code block.
Before and After:
https://github.com/user-attachments/assets/233b891e-15f1-4f10-a51f-75693323c2bd
Release Notes:
- Fixed `editor::SelectLargerSyntaxNode` to properly select nodes when
the cursor is positioned at the end of words or lines.
Closes#28792
supersedes #28854
- Adds support for Socks V4 Identification using a userid, and
Authorization using a username and password on Socks V5.
- Added tests for parsing various Socks proxy urls.
- Added a test for making sure a misconfigured socks proxy url doesn't
expose the user by connecting directly as a fallback.
Release Notes:
- Added support for identification and authorization when using a sock
proxy
The outline panel includes quotes around search terms. The rendering
makes it somewhat ambiguous whether these quotes are part of the search
term and are unnecessary, especially given other rendering
differentiation. This PR removes them.
Release Notes:
- N/A
This PR fixes an issue where scrollbar hitboxes were still inserted for
editors despite scrollbars being programmatically disabled via the
`show_scrollbars`field. This is basically the same fix as in #27467.
The thought process here is that the motivation for `show_scrollbars` is
not to just hide the scrollbars in the editor, but to fully disable
scrollbars for the associated editor. However, this is currently not the
case, as a functioning hitbox for each scrollbar is stil inserted. For
example, the behavior with the old assistant panel can be seen below:
https://github.com/user-attachments/assets/18af6338-dd28-4794-a6a6-5b9691b243f2
Whilst the scrollbar is not visible, there is still a scrollbar hitbox
inserted which triggers hover events and is fully functioning.
This PR fixes this by fully skipping the scrollbar layouting whenever
`show_scrollbars` is set to false, preventing the hitboxes from being
inserted.
https://github.com/user-attachments/assets/b6bb6dc7-902f-4383-bf03-506d0a57ec77
Release Notes:
- N/A
Release Notes:
- Fixed a bug that would cause rejecting a hunk from the agent to delete
the file if the agent had decided to rewrite that file from scratch.
There were two bugs that caused user-defined debug scenarios from being
able to run a build task.
1. DebugRequest would be deserialized to `Attach` even when `process_id`
wasn't defined in a user's configuration file. This has been fixed by
adding our own deserializer that defaults to None if there are no fields
present instead of `Attach`, and I added tests to prevent regressions.
2. Debug scenario resolve phase never got the active buffer when
spawning a debug session from the new session modal. This has been
worked around by passing in the worktree_id of a debug scenario in the
scenario picker and the active worktree_id otherwise.
Release Notes:
- N/A
When starting a selection from only carets, the action
`editor::SelectPrevious` behaved in a manner inconsistent with
`editor::SelectNext` as well as equivalent keybinds in editors such as
VSCode, by selecting substrings of whole words matching the initially
selected string on subsequent triggers.
This fix brings the `select_previous` function in line with
`select_next_internal`by calling `select_match_ranges` (previously an
internal function of `select_next_internal`) in the same way it was
previously used in the function that exhibited expected behavior.
Furthermore, the relevant test was adapted to bring it in line with the
equivalent test for the `editor::SelectNext` action
Closes#24346
Release Notes:
- Fixed inconsistent SelectPrevious behavior
