This PR expands our automatic retry behavior for certain classes of
completion errors (e.g., rate limit errors).
Previously this was only available when using burn mode.
We now auto-retry when:
- Using the Zed provider while on a token-based plan
- Using the Zed provider while on a legacy plan with burn mode enabled
- Using a non-Zed provider
Release Notes:
- Expanded automatic retry behavior for errors in the Agent. Errors
classified as "retryable" (such as rate limit errors) will now
automatically be retried when:
- Using the Zed provider while on a token-based plan
- Using the Zed provider while on a legacy plan with burn mode enabled
- Using a non-Zed provider
---------
Co-authored-by: David Kleingeld <davidsk@zed.dev>
Use path! macro for platform-specific path formatting in test
assertions, fixing hardcoded Unix-style paths that failed on Windows.
Release Notes:
- N/A
Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Make `resolve_full_path` use the appropriate separators, and return a
`String`.
As part of fixing the fallout from that type change, this also fixes a
bunch of places in the agent code that were using `std::path::Path`
operations on paths that could be non-local, by changing them to operate
instead on strings and use the project's `PathStyle`.
This clears the way a bit for making `full_path` also return a string
instead of a `PathBuf`, but I've left that for a follow-up.
Release Notes:
- N/A
Closes#38620
`Url::from_file_path` and `Url::from_directory_path` assume the path
style of the target they were compiled for, so we can't use them in
general. So, switch from `file://` to encoding the absolute path (for
mentions that have one) as a query parameter, which works no matter the
platforms. We'll still parse the old `file://` mention URIs for
compatibility with thread history.
Release Notes:
- windows: Fixed a crash when using `@mentions` in agent threads when
remoting from Windows to Linux or WSL.
Closes https://github.com/zed-industries/zed/issues/38690Closes#37353
### Background
On Windows, paths are normally separated by `\`, unlike mac and linux
where they are separated by `/`. When editing code in a project that
uses a different path style than your local system (e.g. remoting from
Windows to Linux, using WSL, and collaboration between windows and unix
users), the correct separator for a path may differ from the "native"
separator.
Previously, to work around this, Zed converted paths' separators in
numerous places. This was applied to both absolute and relative paths,
leading to incorrect conversions in some cases.
### Solution
Many code paths in Zed use paths that are *relative* to either a
worktree root or a git repository. This PR introduces a dedicated type
for these paths called `RelPath`, which stores the path in the same way
regardless of host platform, and offers `Path`-like manipulation APIs.
RelPath supports *displaying* the path using either separator, so that
we can display paths in a style that is determined at runtime based on
the current project.
The representation of absolute paths is left untouched, for now.
Absolute paths are different from relative paths because (except in
contexts where we know that the path refers to the local filesystem)
they should generally be treated as opaque strings. Currently we use a
mix of types for these paths (std::path::Path, String, SanitizedPath).
Release Notes:
- N/A
---------
Co-authored-by: Cole Miller <cole@zed.dev>
Co-authored-by: Piotr Osiewicz <24362066+osiewicz@users.noreply.github.com>
Co-authored-by: Peter Tripp <petertripp@gmail.com>
Co-authored-by: Smit Barmase <heysmitbarmase@gmail.com>
Co-authored-by: Lukas Wirth <me@lukaswirth.dev>
Closes https://github.com/zed-industries/zed/issues/38528
In the agent panel's `thread_view.rs` file, we have a `render_tool_call`
function that controls what we show in the UI for most types of tools.
However, for some of them—for example, terminal/execute and edit
tools—we have a special rendering so we can tailor the UI for their
specific needs. But... before the specific rendering function is called,
all tools still go through the `render_tool_call`.
Problem is that, in the case of the terminal tool, you couldn't see the
full command the agent wants to run when the tool is still in its
`render_tool_call` state. That's mostly because of the treatment we give
to labels while in that state. A particularly bad scenario because
well... seeing the _full_ command _before_ you choose to accept or
reject is rather important.
This PR fixes that by essentially special-casing the terminal tool
display when in the `render_tool_call` rendering state, so to speak.
There's still a slight UI misalignment I want to fix but it shouldn't
block this fix to go out.
Here's our final result:
<img width="400" height="1172" alt="Screenshot 2025-09-23 at 6 19@2x"
src="https://github.com/user-attachments/assets/71c79e45-ab66-4102-b046-950f137fa3ea"
/>
Release Notes:
- agent: Fixed terminal command not being fully displayed while in the
"waiting for confirmation" state.
Solves an issue where Google APIs refuse all requests with file-based
tool calls attached.
This seems to get triggered in the case where:
- copy_path + another file-based tool call is enabled
- default terminal is `/bin/bash` or something similar
It is unclear why this is happening, but removing the terminal commands
in those tool calls seems to have solved the issue.
Closes#37180 and #37414
Release Notes:
- agent: Fix Gemini refusing requests with certain profiles/systems.
When we refactored settings to not pass JSON blobs around, we ended up
needing
to write *a lot* of code that just merged things (like json merge used
to do).
Use a derive macro to prevent typos in this logic.
Release Notes:
- N/A
Co-Authored-By: Ben K <ben@zed.dev>
Co-Authored-By: Anthony <anthony@zed.dev>
Co-Authored-By: Mikayla <mikayla@zed.dev>
Release Notes:
- settings: Major internal changes to settings. The primary user-facing
effect is that some settings which did not make sense in project
settings files are no-longer read from there. (For example the inline
blame settings)
---------
Co-authored-by: Ben Kunkle <ben@zed.dev>
Co-authored-by: Mikayla Maki <mikayla.c.maki@gmail.com>
Co-authored-by: Anthony <anthony@zed.dev>
In zed logs you can see these logs of lmstudio connection refused.
Currently zed connects to lmstudio by default as there is no credential
mechanism to check if the user has enabled lmstudio previously or not
like we do with other providers using api keys.
This pr removes the below annoying log and makes the zed logs less
polluted.
```
2025-09-01T02:11:33+05:30 ERROR [language_models] Other(error sending request for url (http://localhost:1234/api/v0/models)
Caused by:
0: client error (Connect)
1: tcp connect error: Connection refused (os error 61)
2: Connection refused (os error 61))
```
Release Notes:
- N/A
---------
Signed-off-by: Umesh Yadav <git@umesh.dev>
<img width="598" height="311" alt="Screenshot 2025-09-11 at 9 39 12 PM"
src="https://github.com/user-attachments/assets/b526e648-37cf-4412-83a0-42037b9fc94d"
/>
This is for both ACP and the regular agent. Previously we would always
include the whole file, which can easily blow the context window on huge
files.
Release Notes:
- When `@mention`ing large files, the Agent Panel now send an outline of
the file instead of the whole thing.
This PR adds support for using external agents in SSH projects via ACP,
including automatic installation of Gemini CLI and Claude Code,
authentication with API keys (for Gemini) and CLI login, and custom
agents from user configuration.
Co-authored-by: maan2003 <manmeetmann2003@gmail.com>
Release Notes:
- agent: Gemini CLI, Claude Code, and custom external agents can now be
used in SSH projects.
---------
Co-authored-by: maan2003 <manmeetmann2003@gmail.com>
For read_file and edit_file, show the worktree-relative path if there's
only one visible worktree, and the "full path" otherwise. Also restores
the display of line numbers for read_file calls.
Release Notes:
- N/A
Added schemars annotations to generate inline enums instead of
references ($ref) in the JSON schema passed to LLMs.
Concerns :
- "timezeone" parameter for "now" tool function
- "mode" parameter for "edit_file" tool function
Should be the same for futures tools/functions enums. This is easier for
LLMs to understand the schema since many of them don't use JSON
references correctly.
Tested with :
- local GPT-OSS-120b with llama.cpp server (openai compatible)
- remote Claude Sonnet 4.0 with Zed pro subscription
Thanks in advance for the merge.
(notice this is my first PR ever on Github, I hope I'm doing things
well, please let me know if you have any comment - edit: just noticed my
username/email were not correctly setup on my local git, sorry, it's
been 5 years I've not used git)
Closes#37389
Release Notes:
- agent: Improve "now" and "edit_file" tool schemas to work with more
models.
Closes #ISSUE
Centralizes the references to the `ZED_STATELESS` env var into a single
location in a new crate named `zed_env_vars`
Release Notes:
- N/A *or* Added/Fixed/Improved ...
Closes: #32668
- Add
[tool_call_timeout_millis](https://github.com/cline/cline/pull/1904)
field to ContextServerCommand, like in Cline
- Update ModelContextServerBinary to include timeout configuration
- Modify Client to store and use configurable request timeout
- Replace hardcoded REQUEST_TIMEOUT with self.request_timeout
- Rename REQUEST_TIMEOUT to DEFAULT_REQUEST_TIMEOUT for clarity
- Maintain backward compatibility with 60-second default
Release Notes:
- context_server: Add support for configurable timeout for MCP tool
calls
---------
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Before we were setting up lots of test setup regardless of if we were
actually going to be making real requests or not.
This will hopefully help with intermittent test errors we're seeing on
Windows in CI.
Release Notes:
- N/A
This is just copying over the same workaround here:
a790e514af/crates/agent/src/thread.rs (L1455-L1459)
Into the agent2 code.
Release Notes:
- agent: Fixed an issue where some tool calls in the Zed agent could
return an error like "`tool_use` ids were found without `tool_result`
blocks immediately after"
Closes: https://github.com/zed-industries/zed/issues/37089
Instead of looking for the gemini command on `$PATH`, by default we'll
install our own copy on demand under our data dir, as we already do for
language servers and debug adapters. This also means we can handle
keeping the binary up to date instead of prompting the user to upgrade.
Notes:
- The download is only triggered if you open a new Gemini thread
- Custom commands from `agent_servers.gemini` in settings are respected
as before
- A new `agent_servers.gemini.ignore_system_version` setting is added,
similar to the existing settings for language servers. It's `true` by
default, and setting it to `false` disables the automatic download and
makes Zed search `$PATH` as before.
- If `agent_servers.gemini.ignore_system_version` is `false` and no
binary is found on `$PATH`, we'll fall back to automatic installation.
If it's `false` and a binary is found, but the version is older than
v0.2.1, we'll show an error.
Release Notes:
- acp: By default, Zed will now download and use a private copy of the
Gemini CLI binary, instead of searching your `$PATH`. To make Zed search
your `$PATH` for Gemini CLI before attempting to download it, use the
following setting:
```
{
"agent_servers": {
"gemini": {
"ignore_system_version": false
}
}
}
```
Instead of manually constructing the venv we now ask the python
toolchain for the relevant information, unifying the approach of vent
inspection
Fixes https://github.com/zed-industries/zed/issues/27350
Release Notes:
- Improved the detection of python virtual environments for terminals
and tasks in remote projects.
Closes#37043
Previously claude sonnet 4 was missing from copilot as it was colliding
with zed's claude-sonnet-4 model id. Now we do deduplication based upon
model and provider id both.
| Before | After |
|--------|--------|
| <img width="784" height="950" alt="CleanShot 2025-08-28 at 18 31
28@2x"
src="https://github.com/user-attachments/assets/d49d5a17-7271-417d-bb5e-bc380071e810"
/> | <img width="720" height="876" alt="CleanShot 2025-08-28 at 18 31
42@2x"
src="https://github.com/user-attachments/assets/a5100c05-994e-4e19-ab20-34c0258b977c"
/> |
Release Notes:
- Fixed an issue where models with the same ID from different providers
(such as Claude Sonnet 4 from both Zed and Copilot) were incorrectly
deduplicated in the model selector—now all variants are shown.
Previously, we wouldn't finalize the diff if an error occurred during
editing or the tool call was canceled.
Release Notes:
- N/A
---------
Co-authored-by: Antonio Scandurra <me@as-cii.com>
Adds support for per-session prompt capabilities and capability changes
on the Zed side (ACP itself still only has per-connection static
capabilities for now), and uses it to reflect image support accurately
in 1PA threads based on the currently-selected model.
Release Notes:
- N/A
Using prompt injection, the agent may be tricked into making a fetch
request that includes unexpected data from the conversation in the URL.
As agent conversations may contain sensitive information (like private
code, or
potentially even API keys), this seems bad.
The easiest way to prevent this is to require the user to look at the
URL
before the model is allowed to fetch it.
Thanks to @ant4g0nist for bringing this to our attention.
Release Notes:
- agent panel: The fetch tool now requires confirmation.
We were rendering a Markdown link like `[Read file x.rs (lines
Y-Z)](@selection)` while the tool ran, but then switching to just `x.rs`
as soon as we got the file location from the tool call (due to an
if/else in the UI code that applies to all tools). This caused a
flicker, which is fixed by having `initial_title` return just the
filename from the input as it arrives instead of a link that we're going
to stop rendering almost immediately anyway.
Release Notes:
- N/A
