Skip to main content
The Agent Panel is Glass’s primary AI interface, providing an interactive environment for complex coding tasks. Access it via cmd-shift-a (macOS) or ctrl-shift-a (Linux/Windows).

​
Agent Types

The panel supports multiple agent implementations:

​
Zed Agent (Native)

The built-in agent with full Glass integration:
  • Tool execution - Edit files, run commands, search code
  • Context awareness - Accesses project rules and structure
  • Streaming responses - Real-time output with tool use
  • Multi-model support - Works with any configured provider
Default model: Claude 4.6 Sonnet

​
Panel Features

​
Thread Management

1

Create New Thread

Click the + icon or use cmd-n to start a fresh conversation
2

View History

Access previous threads via the history menu (top-left)
3

Thread Titles

Threads are automatically titled based on content, or edit manually
4

Search Threads

Filter thread history by content or title

​
Message Composition

The message editor supports rich input: 
@file: Reference specific files
@symbol: Reference functions/classes
/command: Execute slash commands
Use shift-enter to add a newline. Press enter alone to send the message.

​
Context Menu

Right-click in the panel for quick actions:
  • Copy thread - Copy entire conversation
  • Export as markdown - Save conversation to file
  • Share thread - Generate shareable link
  • Clear history - Remove old threads

​
Tool Execution

Native agents can execute tools with your approval:

File Operations

  • Read file contents
  • Write new files
  • Edit with unified diffs
  • Delete files

Terminal

  • Run shell commands
  • Execute scripts
  • Install dependencies
  • Git operations

Code Search

  • Project-wide search
  • Semantic search
  • Symbol search
  • Regex patterns

LSP Actions

  • Go to definition
  • Find references
  • Get diagnostics
  • Symbol information

​
Tool Permission System

Configure how tools are authorized: 
{
  "agent": {
    "tool_permissions": {
      "default": "confirm"
    }
  }
}
Hardcoded security rules prevent destructive operations like rm -rf / regardless of settings.

​
Model Selection

Change models mid-conversation:
1

Open Model Menu

Click the model name in the panel header
2

Browse Providers

Models are grouped by provider (Anthropic, OpenAI, etc.)
3

Select Model

Choose a model - the next message uses it
4

Thinking Models

Some models (Claude Sonnet 4.6 Thinking, o1) show reasoning steps
For different tasks:
TaskRecommended ModelWhy
Code generationClaude 4.6 SonnetBest code quality and tool use
Fast responsesGPT-4o MiniQuick, cost-effective
ReasoningClaude 4.6 Sonnet ThinkingShows reasoning steps
Long contextClaude 4.6 Sonnet200k context window

​
Context Management

​
Automatic Context

Native agents automatically include:
  • Active file - Current editor content
  • Project structure - Directory layout
  • Rules files - .zed.md, .rules, .clinerules
  • Related files - Semantically relevant code

​
Manual Context

Add explicit context: 
@README.md can you update the installation instructions?

@src/lib.rs @tests/integration.rs help me write tests

​
Context Servers

Extend context via Model Context Protocol (MCP) servers: 
{
  "context_servers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/project"]
    }
  }
}
Context servers can provide:
  • External file systems
  • Database access
  • API documentation
  • Custom knowledge sources

​
Agent Configuration

Customize agent behavior: 
{
  "agent": {
    "enabled": true,
    "button": true,
    "dock": "right",
    "default_width": 400,
    "default_view": "agent",
    "notify_when_agent_waiting": "always",
    "play_sound_when_agent_done": true,
    "use_modifier_to_send": false,
    "message_editor_min_lines": 3
  }
}

​
Slash Commands

Context server prompts appear as slash commands: 
/search find all TODO comments
/git show recent commits
/docs search for authentication examples
Configure via context servers in settings.

​
Agent Profiles

Create custom agent configurations: 
{
  "agent": {
    "default_profile": "write",
    "profiles": {
      "write": {
        "default_model": {
          "provider": "anthropic",
          "model": "claude-4.6-sonnet"
        }
      },
      "review": {
        "default_model": {
          "provider": "openai",
          "model": "o1"
        }
      }
    }
  }
}
Switch profiles via the profile menu in the panel.

​
Keyboard Shortcuts

ActionmacOSLinux/Windows
Toggle panelcmd-shift-actrl-shift-a
New threadcmd-nctrl-n
Focus editorescapeescape
Send messageenterenter
New lineshift-entershift-enter
Historycmd-hctrl-h

​
Troubleshooting

  • Check API key is configured
  • Verify provider authentication
  • Check network connectivity
  • View logs: cmd-shift-p → “Open Logs”
  • Review tool permissions in settings
  • Check file/directory permissions
  • Verify terminal access
  • Review hardcoded security rules
  • Restart context servers
  • Check context server logs
  • Verify MCP server configuration
  • Ensure files are indexed

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us