Skip to main content
Zed’s AI features are highly configurable, allowing you to customize models, permissions, and behavior to match your workflow.

​
Settings Location

AI settings are configured in your Zed settings file:
  1. Open settings: Cmd+, (macOS) or Ctrl+, (Linux/Windows)
  2. Navigate to the agent section
  3. Or edit ~/.config/zed/settings.json directly

​
Basic Configuration

​
Enable/Disable AI Features

{
  "agent": {
    "enabled": true,  // Master switch for all AI features
    "button": true    // Show agent panel button in status bar
  }
}
Disabling AI features will:
  • Hide the agent panel button
  • Disable inline assistant
  • Remove AI-related menu items

​
Model Configuration

​
Default Model

Set the primary model for all AI features: 
{
  "agent": {
    "default_model": {
      "provider": "anthropic",
      "model": "claude-3-5-sonnet-20241022"
    }
  }
}

​
Task-Specific Models

Use different models for different tasks: 
{
  "agent": {
    "default_model": {
      "provider": "anthropic",
      "model": "claude-3-5-sonnet-20241022"
    },
    "inline_assistant_model": {
      "provider": "openai",
      "model": "gpt-4o"
    },
    "commit_message_model": {
      "provider": "zed.dev",
      "model": "claude-3-5-haiku-20241022"
    },
    "thread_summary_model": {
      "provider": "openai",
      "model": "gpt-4o-mini"
    }
  }
}
Model types:
  • default_model - Used for Agent Panel conversations
  • inline_assistant_model - Used for inline code generation
  • commit_message_model - Used for generating git commit messages
  • thread_summary_model - Used for summarizing conversation threads

​
Favorite Models

Pin frequently used models to the top of the model selector: 
{
  "agent": {
    "favorite_models": [
      {
        "provider": "anthropic",
        "model": "claude-3-5-sonnet-20241022"
      },
      {
        "provider": "openai",
        "model": "gpt-4o"
      },
      {
        "provider": "ollama",
        "model": "qwen2.5-coder:32b"
      }
    ]
  }
}

​
Model Parameters

Customize model behavior with parameters: 
{
  "agent": {
    "model_parameters": [
      {
        "provider": "anthropic",
        "model": "claude-3-5-sonnet-20241022",
        "temperature": 0.7
      },
      {
        "provider": "openai",
        "temperature": 0.8  // Applies to all OpenAI models
      },
      {
        "model": "gpt-4o",
        "temperature": 0.5  // Override for specific model
      }
    ]
  }
}
Parameters cascade: Settings are applied from most general to most specific. The last matching entry wins.

​
Extended Thinking

Enable extended reasoning for complex tasks: 
{
  "agent": {
    "default_model": {
      "provider": "anthropic",
      "model": "claude-3-7-sonnet-20250219",
      "enable_thinking": true,
      "effort": "medium"  // Options: "low", "medium", "high"
    }
  }
}
Thinking mode:
  • Model spends more time reasoning
  • Shows thinking process during generation
  • Better for complex, multi-step problems
  • Uses more tokens (costs more)

​
Tool Permissions

Control what the AI can do in your workspace:

​
Global Default

{
  "agent": {
    "tool_permissions": {
      "default": "confirm"  // "allow", "confirm", or "deny"
    }
  }
}
Permission modes:
  • allow - Execute without asking
  • confirm - Always prompt user (default)
  • deny - Never allow

​
Per-Tool Configuration

{
  "agent": {
    "tool_permissions": {
      "default": "confirm",
      "tools": {
        "edit_file": {
          "default": "confirm"
        },
        "create_file": {
          "default": "confirm"
        },
        "terminal": {
          "default": "confirm",
          "always_allow": [
            {"pattern": "^cargo test"},
            {"pattern": "^npm test"},
            {"pattern": "^git status"}
          ],
          "always_deny": [
            {"pattern": "rm -rf"},
            {"pattern": "sudo", "case_sensitive": false}
          ]
        },
        "fetch": {
          "default": "allow",
          "always_deny": [
            {"pattern": "localhost"},
            {"pattern": "127\\.0\\.0\\.1"}
          ]
        }
      }
    }
  }
}
Pattern matching:
  • always_deny takes precedence over everything
  • Then always_confirm is checked
  • Then always_allow is checked
  • Finally, the default mode applies
Regex rules:
  • Patterns are case-insensitive by default
  • Set "case_sensitive": true for case-sensitive matching
  • Use standard regex syntax
  • Escape special characters: \. for ., \\ for \

​
Built-in Security Rules

Zed has hardcoded protections that cannot be overridden: 
rm -rf /
rm -rf ~
rm -rf $HOME
rm -rf .
rm -rf ..
These commands are always blocked for safety.

​
Agent Profiles

Create profiles for different workflows: 
{
  "agent": {
    "default_profile": "write",
    "profiles": {
      "write": {
        "name": "Code Writer",
        "tools": {
          "edit_file": true,
          "create_file": true,
          "delete_path": false,
          "terminal": true
        },
        "default_model": {
          "provider": "anthropic",
          "model": "claude-3-5-sonnet-20241022"
        }
      },
      "review": {
        "name": "Code Reviewer",
        "tools": {
          "edit_file": false,
          "create_file": false,
          "terminal": false,
          "read_file": true
        },
        "default_model": {
          "provider": "openai",
          "model": "gpt-4o"
        }
      },
      "debug": {
        "name": "Debugger",
        "tools": {
          "terminal": true,
          "read_file": true,
          "edit_file": true
        },
        "default_model": {
          "provider": "anthropic",
          "model": "claude-3-7-sonnet-20250219",
          "enable_thinking": true
        }
      }
    }
  }
}
Switch profiles from the Agent Panel settings menu.

​
UI Configuration

​
Panel Position and Size

{
  "agent": {
    "dock": "right",        // "left", "right", or "bottom"
    "default_width": 800,   // pixels, for left/right dock
    "default_height": 400   // pixels, for bottom dock
  }
}

​
Message Editor

{
  "agent": {
    "use_modifier_to_send": false,     // Require Cmd+Enter to send
    "message_editor_min_lines": 4      // Minimum height of input
  }
}

​
Display Options

{
  "agent": {
    "expand_edit_card": true,          // Show full diff preview
    "expand_terminal_card": true,      // Show full command output
    "enable_feedback": true,           // Show thumbs up/down buttons
    "show_turn_stats": false,          // Show timing and token stats
    "single_file_review": true         // Show edits in source files too
  }
}

​
Notifications

{
  "agent": {
    "notify_when_agent_waiting": "primary_screen",  // "all_screens", "never"
    "play_sound_when_agent_done": false
  }
}

​
Advanced Settings

​
Default View

Choose the starting view for the Agent Panel: 
{
  "agent": {
    "default_view": "thread"  // "thread" or "text_thread"
  }
}

​
Inline Assistant Options

{
  "agent": {
    "inline_assistant_use_streaming_tools": true,
    "inline_alternatives": [
      {"provider": "anthropic", "model": "claude-3-5-sonnet-20241022"},
      {"provider": "openai", "model": "gpt-4o"},
      {"provider": "ollama", "model": "qwen2.5-coder:32b"}
    ]
  }
}

​
Terminal Integration

{
  "agent": {
    "cancel_generation_on_terminal_stop": true
  }
}
When true, clicking the stop button in a terminal tool card also cancels the AI generation.

​
Context Server Configuration

Configure MCP (Model Context Protocol) context servers: 
{
  "agent": {
    "profiles": {
      "write": {
        "enable_all_context_servers": false,
        "context_servers": {
          "git": {
            "tools": {
              "git_log": true,
              "git_diff": true
            }
          },
          "filesystem": {
            "tools": {
              "read_file": true,
              "write_file": true
            }
          }
        }
      }
    }
  }
}

​
Complete Example

A comprehensive configuration: 
{
  "agent": {
    "enabled": true,
    "button": true,
    "dock": "right",
    "default_width": 800,
    
    "default_model": {
      "provider": "anthropic",
      "model": "claude-3-5-sonnet-20241022"
    },
    
    "inline_assistant_model": {
      "provider": "openai",
      "model": "gpt-4o"
    },
    
    "favorite_models": [
      {"provider": "anthropic", "model": "claude-3-5-sonnet-20241022"},
      {"provider": "openai", "model": "gpt-4o"},
      {"provider": "ollama", "model": "qwen2.5-coder:32b"}
    ],
    
    "model_parameters": [
      {
        "provider": "anthropic",
        "temperature": 0.7
      }
    ],
    
    "tool_permissions": {
      "default": "confirm",
      "tools": {
        "terminal": {
          "default": "confirm",
          "always_allow": [
            {"pattern": "^cargo (test|build|check)"},
            {"pattern": "^npm (test|run)"},
            {"pattern": "^git (status|log|diff)"}
          ],
          "always_deny": [
            {"pattern": "rm -rf"},
            {"pattern": "^sudo"}
          ]
        }
      }
    },
    
    "expand_edit_card": true,
    "use_modifier_to_send": false,
    "notify_when_agent_waiting": "primary_screen"
  }
}

​
Best Practices

  1. Start conservative - Use "confirm" mode while learning the AI’s behavior
  2. Whitelist safe commands - Add common commands to always_allow once you trust them
  3. Profile per task - Create separate profiles for coding vs reviewing vs debugging
  4. Choose models wisely - Faster models for simple tasks, powerful models for complex ones
  5. Use local models - Consider Ollama for privacy-sensitive work
  6. Review tool permissions - Regularly audit what the AI is allowed to do

​
Troubleshooting

​
Settings not taking effect

  • Save your settings file (Cmd+S)
  • Reload Zed: Cmd+Shift+P → “zed: reload”
  • Check for JSON syntax errors

​
Invalid model configuration

  • Verify the provider name matches exactly
  • Check that the model is available for your provider
  • Ensure API keys are configured in Settings → Language Models

​
Tool permissions not working

  • Test regex patterns in a regex tester
  • Remember patterns are case-insensitive by default
  • Check the order: always_deny > always_confirm > always_allow > default