Skip to content
Refact Documentation

Agent Tools

The Refact.ai Agent is designed to operate autonomously, extending its capabilities beyond simple integrations. While integrations allow external services to be connected, Agent Tools empower the agent to perform key operations using built-in functionalities. Below is an overview of the available Agent Tools, typical use cases, and how they can enhance your development workflow.

Core Tools

The Agent has access to several powerful tools that help it understand and modify your codebase:

Context and Search Tools

  • search
    Find similar pieces of code or text using vector database.
    Use Case: When you ask the Agent to modify code, it uses this tool to find similar patterns across your codebase to maintain consistency.

  • definition
    Read definition of symbols in the project using AST.
    Use Case: The Agent uses this to understand function signatures, class structures, and type definitions when working with your code.

  • references
    Find usages of symbols within the project using AST.
    Use Case: Before modifying a function or class, the Agent checks all its usages to ensure changes won’t break existing code.

  • tree
    Get a files tree with symbols for the project.
    Use Case: The Agent uses this to understand project structure and locate relevant files.

File Operations

  • cat
    Read multiple files and understand their content.
    Use Case: The Agent can read and analyze multiple files at once, including images and skeletonized code views.

  • locate
    Find relevant files for a specific task.
    Use Case: When given a task, the Agent can quickly identify which files need to be modified.

Code Modification

  • patch
    Apply changes to files in a controlled manner.
    Use Case: The Agent uses this to make actual changes to your codebase, with your approval.

Planning and Analysis

  • think
    Analyze complex problems and create execution plans using o3 mini reasoning model.
    Use Case: Before making changes, the Agent plans out the steps needed to complete a task successfully. The o3 mini model helps break down complex problems into manageable steps and create a clear execution strategy.

Web Interaction

  • web
    Fetch and read web pages in plain text format.
    Use Case: The Agent can read documentation, specifications, or other web resources to help solve problems.

How Agent Tools Work Together

The Agent combines these tools strategically to complete complex tasks. Here’s a typical workflow:

  1. Understanding Phase

    • Uses tree to understand project structure
    • Uses locate to find relevant files
    • Uses search to find similar patterns
    • Uses definition and references to understand code context

  2. Planning Phase

    • Uses think to create a detailed plan
    • Uses web if external documentation is needed

  3. Execution Phase

    • Uses cat to read necessary files
    • Creates changes using patch tool
    • Verifies changes using search and references

Best Practices

When working with the Agent, consider these tips:

  • Let the Agent gather context before making changes
  • Review proposed patches carefully before approving
  • Use the Agent’s planning capabilities for complex tasks
  • Provide clear, specific instructions for best results

Next Steps

Once you’re familiar with the core tools, you might want to explore:

For specific integration guides and advanced usage scenarios, refer to our detailed documentation sections.