Kilo Code Documentation

Kilo Code allows you to run language models locally on your machine, providing privacy, offline access, and cost savings.

Supported providers include:

• Ollama: Open-source tool supporting a wide range of models.
• LM Studio: Desktop application that provides a local server emulating the OpenAI API.
• Atomic Chat: Open-source local AI with TurboQuant-optimized inference and an OpenAI-compatible API on port 1337. Kilo Code can automatically discover models if you use the provider.atomic-chat setting, atomicChat.autoDetect, or a model name prefixed with atomic-chat/....

Kilo Code provides specialized agents that you can switch between depending on your task:

• Code: The default agent. Implements and edits code using natural language.
• Plan: Designs architecture and writes implementation plans before writing code.
• Ask: Answers questions about your codebase without modifying files.
• Debug: Troubleshoots and tracks down issues.
• Review: Reviews changes to identify performance, security, style, and test coverage issues.

You can also create your own custom agents.

Kilo Code is an AI pair programmer designed to assist with code generation, refactoring, debugging, and shipping software. It operates across multiple interfaces including IDEs, terminals, and browsers, providing AI that understands your specific codebase and context.

Core Capabilities

• Code Generation & Refactoring: Use Autocomplete for inline suggestions and Code Actions for AI-powered refactoring and fixes.
• Task Management: Utilize Task & Todo Lists to decompose complex engineering tasks and Checkpoints to save or restore specific working states.
• Advanced Automation: Leverage Browser Use to automate web interactions and Git Commit Generation to create meaningful commit messages.
• Prompt Optimization: Use Enhance Prompt to automatically improve the quality of your instructions to the AI.

Kilo Code provides specialized agents that can be switched depending on the task. You can also create your own custom agents.

• Code: Implements and edits code from natural language.
• Plan: Designs architecture and writes implementation plans before writing code.
• Ask: Answers questions about the codebase without modifying files.
• Debug: Resolves and tracks issues.
• Review: Reviews changes to find issues in performance, security, style, and test coverage.

Kilo Code provides specialized agents that you can switch between depending on the task:

• Code: Implements and edits code from natural language.
• Plan: Designs architecture and writes implementation plans before coding.
• Ask: Answers questions about the codebase without modifying files.
• Debug: Diagnoses and tracks down issues.
• Review: Reviews changes to identify performance, security, style, and test coverage issues.

Kilo provides specialized agents that can be switched depending on the task. You can also build custom agents.

• Code: The standard agent. Implements and edits code using natural language.
• Plan: Designs architecture and writes implementation plans before writing code.
• Ask: Answers questions about your codebase without modifying files.
• Debug: Troubleshoots and tracks down issues.
• Review: Reviews changes to find issues with performance, security, style, and test coverage.

Kilo Code provides specialized agents that you can switch between depending on your task. You can also create custom agents.

• Code: The default agent. Executes and modifies code based on natural language instructions.
• Plan: Designs architecture and writes implementation plans before writing any code.
• Ask: Answers questions about your codebase without modifying files.
• Debug: Explores and traces issues.
• Review: Reviews your changes to detect performance, security, style, and test coverage issues.

Kilo includes specialized agents that can be switched depending on the task. You can also create custom agents.

• Code: The default agent. Implements and edits code from natural language.
• Plan: Designs architecture and creates implementation plans before writing code.
• Ask: Answers questions about the codebase without modifying files.
• Debug: Performs troubleshooting and issue tracking.
• Review: Reviews changes to detect performance, security, style, and test coverage issues.

The Model Context Protocol (MCP) is a standard used to extend Kilo Code's capabilities by connecting it to external tools and services. MCP servers allow Kilo Code to interact with resources beyond its built-in features, such as databases, custom APIs, and specialized local or remote functionality.

Key concepts include:

• Client-Server Architecture: Kilo Code acts as the client, interacting with MCP servers that provide tools and resources.
• Transports: Servers can communicate via STDIO (typically for local processes) or SSE (Server-Sent Events, typically for remote services).
• Permissions: MCP tools are managed using the same allow / ask / deny permission system used for built-in tools, allowing you to control whether tools are auto-approved, prompt for user permission, or are blocked.