Agent Mode

Enable Agent Mode to leverage AI-powered tools for automatically executing code modifications, fixing vulnerabilities, running commands, and other operations. This mode supports direct adjustments to your codebase through natural language instructions.

In Agent Mode, the AI model not only possesses conversational capabilities but is also equipped with a complete toolkit for various coding tasks. It can autonomously decide which tools to call and how to operate them, eliminating the tedious steps of manually searching for context, editing files, or executing commands.

💡 Learn how to choose the right mode

💬 Chat Mode

Learning and discussion only, no code modification.

Core Positioning: Communicating with professional colleagues

Suitable Scenarios: Explaining concepts, comparing solutions, code review discussions.

👀 Plan Mode

Safely explore and plan using read-only tools.

Core Positioning: Architect surveying before renovation

Suitable Scenarios: Understanding codebase, troubleshooting bugs, planning implementation solutions.

🤖 Agent Mode

Execute actual modifications with full tool access.

Core Positioning: Contractor executing confirmed design blueprint

Suitable Scenarios: Implementing features, fixing bugs, running tests and commands.

How to Use Agent Mode

You can switch to "Agent" mode using the mode selector below the chat input box. The mode selector offers three options:

• Chat Mode: No tools available, pure conversation only

• Plan Mode: Provides read-only tools for safe exploration and planning

• Agent Mode: Provides all tools, supports code modification operations

How to Interact in Agent Mode

Agent Mode shares the same interface as Chat Mode, so you can send messages using the same input methods, while still being able to use manual context provision methods, such as through the "@" context provider, or adding highlighted code from the editor.

How to Use Natural Language in Agent Mode

In Agent Mode, you only need to provide natural language instructions, and the model will automatically complete the subsequent work.

How to Grant Permissions to Agent Mode

By default, Agent Mode requests permission before using tools. Click Allow to permit Agent Mode to execute tool calls, or click Cancel to deny.

How to View Tool Responses

All data returned by tool calls is automatically fed back to the model as context items. Most errors are also captured and returned so Agent Mode can decide on subsequent actions.

How Agent Mode Works

Agent Mode retains all the functionality of Chat Mode while additionally providing the model with a set of callable tools and building a complete tool interaction process:

1. User sends a request in Agent Mode

2. System sends the currently available tool list along with the request to the model

3. Model generates one or more tool call instructions in its response

4. If the tool policy is not set to "automatic", the system prompts the user for authorization

5. After authorization, the system executes the tool through built-in functions or MCP services

6. Tool execution results are returned to the model as context for the next step of reasoning

7. Model generates new responses based on the results, which may include a new round of tool calls, cycling until the task is completed

ⓘ Tool availability varies by mode:

Chat Mode: Does not include any tools
Plan Mode: Includes only read-only tools
Agent Mode: Includes all tools

Available Tools in Agent Mode (All Tools)

In Agent Mode, all tools are available, including the aforementioned read-only tools and the following:

• Create New File (Tool ID: create_new_file): Create new files within the project

• Edit Files (Tool ID: edit_file): Modify existing files

• Run Terminal Commands (Tool ID: run_terminal_command): Run commands from the workspace root directory

• Create Rule Blocks (Tool ID: create_rule_block): Create new rule blocks in the .vjsp/rules directory

• All other write/execute tools for modifying the codebase