react-agent

A powerful, autonomous AI coding assistant built in Rust. This agent uses the ReAct (Reasoning and Acting) pattern to understand complex prompts, plan its actions, and execute them using a suite of integrated tools.

Capabilities

• Autonomous Agent Loop: Implements the ReAct pattern, allowing the agent to "think" (reason) and "act" (execute tools) in an iterative loop to solve multi-step problems.
• Tool Integration:
  • Bash Tool: Execute shell commands directly to run scripts, install dependencies, or explore the system.
  • File Tool: Read and write files within the project workspace, enabling autonomous coding and documentation.
• LLM Agnostic: Integrated with OpenRouter, allowing it to leverage various state-of-the-art models (like Claude 3.5 Sonnet or GPT-4o) while maintaining a consistent OpenAI-compatible interface.
• Flexible Orchestration: Handles context management, tool execution, and response parsing seamlessly.

Project Structure

The project is organized into modular components for easy extensibility:

• src/agent/: The brain of the application. Contains the Orchestrator which manages the ReAct loop and maintains the conversation context.
• src/llm/: The communication layer. Handles requests to LLM providers (via OpenRouter) and parses responses into usable payloads.
• src/tools/: The agent's hands. Each tool (e.g., BashTool, FileTool) is implemented here as a discrete module with specific capabilities.
• src/main.rs: The main entry point that initializes the agent and processes user input.

Getting Started

Prerequisites

• Rust (1.92+)
• An OpenRouter API Key (set as an environment variable)

Running the Agent

The recommended way to run the agent is using cargo run:

cargo run -- "Your prompt here"

How it Works

1. User Prompt: You provide a task (e.g., "Implement a search function in src/lib.rs").
2. Reasoning: The agent analyzes the request and decides which tool to use.
3. Action: The agent executes a tool (like read_file or bash).
4. Observation: The agent receives the output of the tool.
5. Iteration: Steps 2-4 repeat until the agent determines the task is complete and provides a final response.