Agent Panel

The Agent Panel is where you interact with AI agents that can read, write, and run code in your project. It's the core of Zed's AI code editing experience — use it for code generation, refactoring, debugging, documentation, and general questions.

Open it with agent: new thread from the Command Palette or click the ✨ icon in the status bar.

Getting Started

If you're using the Agent Panel for the first time, you need to have at least one LLM provider or external agent configured. You can do that by:

subscribing to our Pro plan, so you have access to our hosted models
using your own API keys, either from model providers like Anthropic or model gateways like OpenRouter.
using an external agent like Gemini CLI or Claude Agent

Overview

With an LLM provider or external agent configured, type in the message editor and press enter to submit. Expand the editor with shift-alt-escape|shift-alt-escape if you need more room.

Responses stream in with indicators showing which tools the model is using. The sections below cover what you can do from here.

Note that for external agents, like Gemini CLI or Claude Agent, some of the features outlined below may not be supported—for example, restoring threads from history, checkpoints, token usage display, and others. Their availability varies depending on the agent.

Creating New Threads

By default, the Agent Panel uses Zed's first-party agent.

To choose another agent, go to the plus button in the top-right of the Agent Panel and pick either one of the external agents installed out of the box or a new Text Thread.

Editing Messages

Any message that you send to the model is editable. You can click on the card that contains your message and re-submit it with an adjusted prompt and/or new pieces of context.

Queueing Messages

Messages sent while the agent is in the generating state get, by default, queued.

For the Zed agent, queued messages get sent at the next turn boundary, which is usually between a tool call and a response, whereas for external agents, the message gets sent at the end of the generation.

You can edit or remove (an individual or all) queued messages. You can also still interrupt the agent immediately if you want by either clicking on the stop button or by clicking the "Send Now" (double-enter) on a queued message.

Checkpoints

Every time the model performs an edit, you should see a "Restore Checkpoint" button at the top of your message, allowing you to return your code base to the state it was in prior to that message.

The checkpoint button appears even if you interrupt the thread midway through an edit, as this is likely a moment when you've identified that the agent is not heading in the right direction and you want to revert back.

Right-click on any agent response in the thread view to access a context menu with the following actions:

Copy Selection: Copies the currently selected text as Markdown (available when text is selected).
Copy This Agent Response: Copies the full text of the agent response you right-clicked on.
Scroll to Top / Scroll to Bottom: Scrolls to the beginning or end of the thread, depending on your current position.
Open Thread as Markdown: Opens the entire thread as a Markdown file in a new tab.

Navigating the Thread

In long conversations, use the scroll arrow buttons at the bottom of the panel to jump to your most recent prompt or to the very beginning of the thread.

Navigating History

To quickly navigate through recently updated threads, use the cmd-shift-j|ctrl-shift-j binding when focused on the panel's editor, or click the menu icon button at the top right of the panel. Doing that will open a dropdown that shows you your six most recently updated threads.

To view all historical conversations, reach for the View All option from within the same menu or via the cmd-shift-h|ctrl-shift-h binding.

Thread titles are auto-generated based on the content of the conversation. But you can also edit them manually by clicking the title and typing, or regenerate them by clicking the "Regenerate Thread Title" button in the ellipsis menu in the top right of the panel.

Following the Agent

Follow the agent as it reads and edits files by clicking the crosshair icon at the bottom left of the panel. Your editor will jump to each file the agent touches.

You can also hold cmd/ctrl when submitting a message to automatically follow.

Get Notified

If you send a prompt to the Agent and then put Zed in the background, you can choose to be notified when its generation wraps up via:

a visual notification that appears in the top right of your screen
a sound notification

These notifications can be used together or individually, and you can use the agent.notify_when_agent_waiting and agent.play_sound_when_agent_done settings keys to customize that, including turning both off entirely.

Reviewing Changes

Once the agent has made changes to your project, the panel will surface which files, how many of them, and how many lines have been edited.

To see which files specifically have been edited, expand the accordion bar that shows up right above the message editor or click the Review Changes button (shift-ctrl-r|ctrl-shift-r), which opens a special multi-buffer tab with all changes.

You can accept or reject each individual change hunk, or the whole set of changes made by the agent.

Edit diffs also appear in singleton buffers. If your active tab had edits made by the AI, you'll see diffs with the same accept/reject controls as in the multi-buffer. You can turn this off, though, through the agent.single_file_review setting.

Adding Context

The agent can search your codebase to find relevant context, but providing it explicitly improves response quality and reduces latency.

Add context by typing @ in the message editor. You can mention files, directories, symbols, previous threads, rules files, and diagnostics.

Copying images and pasting them in the panel's message editor is also supported.

When you paste multi-line code selections copied from a buffer, Zed automatically formats them as @-mentions with the file context. To paste content without this automatic formatting, use cmd-shift-v|ctrl-shift-v to paste raw text directly.

Selection as Context

Additionally, you can also select text in a buffer or terminal and add it as context by using the cmd->|ctrl-> keybinding, running the agent: add selection to thread action, or choosing the "Selection" item in the + menu in the message editor.

Token Usage

Zed surfaces how many tokens you are consuming for your currently active thread near the profile selector in the panel's message editor.

Once you approach the model's context window, a banner appears above the message editor suggesting to start a new thread with the current one summarized and added as context. You can also do this at any time with an ongoing thread via the "Agent Options" menu on the top right, where you'll see a "New from Summary" button, as well as simply @-mentioning a past thread in a new one..

Changing Models

After you've configured your LLM providers—either via a custom API key or through Zed's hosted models—you can switch between their models by clicking on the model selector on the message editor or by using the cmd-alt-/|ctrl-alt-/ keybinding.

The same model can be offered via multiple providers - for example, Claude Sonnet 4.5 is available via Zed Pro, OpenRouter, Anthropic directly, and more. Make sure you've selected the correct model provider for the model you'd like to use, delineated by the logo to the left of the model in the model selector.

Favoriting Models

You can mark specific models as favorites either through the model selector, by clicking on the star icon button that appears as you hover the model, or through your settings via the agent.favorite_models settings key.

Cycle through your favorites with alt-tab|alt-l without opening the model selector.

Using Tools

The Agent Panel supports tool calling, which enables agentic editing. Zed includes several built-in tools for searching your codebase, editing files, running terminal commands, and more.

You can also extend the set of available tools via MCP Servers.

Profiles

Profiles act as a way to group tools. Zed offers three built-in profiles and you can create as many custom ones as you want.

Built-in Profiles

Write: A profile with tools to allow the LLM to write to your files and run terminal commands. This one essentially has all built-in tools turned on.
Ask: A profile with read-only tools. Best for asking questions about your code base without the concern of the agent making changes.
Minimal: A profile with no tools. Best for general conversations with the LLM where no knowledge of your code base is necessary.

You can explore the exact tools enabled in each profile by clicking on the profile selector button > Configure button > the one you want to check out.

Alternatively, you can also use either the command palette, by running agent: manage profiles, or the keybinding directly, cmd-alt-p|ctrl-alt-p, to have access to the profile management modal.

Use shift-tab|shift-tab to switch between profiles without opening the modal.

Custom Profiles

You can also create a custom profile through the Agent Profile modal. From there, you can choose to Add New Profile or fork an existing one with a custom name and your preferred set of tools.

It's also possible to override built-in profiles. In the Agent Profile modal, select a built-in profile, navigate to Configure Tools, and rearrange the tools you'd like to keep or remove.

Zed will store this profile in your settings using the same profile name as the default you overrode.

All custom profiles can be edited via the UI or by hand under the agent.profiles key in your settings file.

To delete a custom profile, open the Agent Profile modal, select the profile you want to remove, and click the delete button.

Tool Permissions

Note: In Zed v0.224.0 and above, tool approval is controlled by agent.tool_permissions.default. In earlier versions, it was controlled by the agent.always_allow_tool_actions boolean (default false).

Zed's Agent Panel provides the agent.tool_permissions.default setting to control tool approval behavior:

"confirm" (default) — Prompts for approval before running any tool action
"allow" — Auto-approves tool actions without prompting
"deny" — Blocks all tool actions

When the agent requests permission for an action, the confirmation menu includes options to allow or deny once, plus "Always for " choices that set a tool-level default. When Zed can extract a safe pattern from the input, it also offers pattern-based "Always for ..." choices that add always_allow/always_deny rules. MCP tools only support tool-level defaults.

Even with "default": "allow", per-tool always_deny and always_confirm patterns are still respected — so you can auto-approve most actions while blocking or gating specific ones.

Learn more about how tool permissions work, how to further customize them, and other details.

Model Support

Tool calling needs to be individually supported by each model and model provider. Therefore, despite the presence of built-in tools, some models may not have the ability to pick them up. You should see a "No tools" label if you select a model that falls into this case.

All Zed's hosted models support tool calling out-of-the-box.

MCP Servers

Similarly to the built-in tools, some models may not support all tools included in a given MCP Server. Zed's UI will inform you about this via a warning icon that appears close to the model selector.

Text Threads

"Text Threads" present your conversation with the LLM in a different format—as raw text. With text threads, you have full control over the conversation data. You can remove and edit responses from the LLM, swap roles, and include more context earlier in the conversation.

Text threads are Zed's original assistant panel format, preserved for users who want direct control over conversation data. Autonomous code editing (where the agent writes to files) is only available in the default thread format, not text threads.

Errors and Debugging

If you hit an error or unusual LLM behavior, open the thread as Markdown with agent: open thread as markdown and attach it to your GitHub issue.

You can also open threads as Markdown by clicking on the file icon button, to the right of the thumbs down button, when focused on the panel's editor.

Feedback

You can rate agent responses to help improve Zed's system prompt and tools.

Note that rating responses will send your data related to that response to Zed's servers. See AI Improvement and Privacy and Security for more information about Zed's approach to AI improvement, privacy, and security. If you don't want data persisted on Zed's servers, don't rate. We will not collect data for improving our Agentic offering without you explicitly rating responses.

To help improve Zed's system prompt and tools, rate responses with the thumbs up/down controls at the end of each response. In case of a thumbs down, a new text area will show up where you can add more specifics about what happened.

You can provide feedback on the thread at any point after the agent responds, and multiple times within the same thread.