CodePilot: Agentic Coding Assistant

An autonomous, iterative AI coding agent built with Python and the OpenAI API. This agent is capable of interacting securely with the local file system and executing Python scripts to solve complex user requests autonomously.

Features

• Autonomous Agent Loop: A restricted up-to-20 iteration loop allows the agent to reason, act, observe, and summarize multi-step tasks.
• Secure Sandbox Execution: All operations are strictly bounded to a designated working directory (e.g., ./calculator) to prevent unauthorized file access.
• Tool Suite:
  • get_files_info: Lists contents of directories.
  • get_file_content: Reads text files (truncated at 10k characters).
  • write_file: Creates or overwrites files with automatic parent directory creation.
  • run_python_file: Executes Python scripts securely with a 30s timeout and captures standard output and standard error.

Workflow Execution

The agent operates in a "Think-Act-Observe" cycle, communicating with the LLM via defined JSON Schema tools.

graph TD
    A[User Request] --> B[LLM Evaluates & Reasons]
    B --> C{Tool Needed?}
    C -->|Yes| D[Emit JSON Tool Call Request]
    C -->|No| E[Return Final Text Response]

    D --> F[Dispatcher 'call_function.py']
    F --> G[Execute Function within Secure Sandbox]
    G --> H[Return String Result]
    H --> I[Append Result as 'role: tool' to Conversation History]
    I --> B

Loading

Usage

Ensure you have your OpenRouter API key inside a .env file first:

OPENROUTER_API_KEY=your-api-key-here

To run the agent, simply execute:

uv run main.py "your prompt here"

To see detailed multi-step reasoning and function calls:

uv run main.py "your prompt here" --verbose