This repository provides an open-source implementation of the "Agent Q" research paper, offering advanced reasoning and learning capabilities for autonomous AI agents designed to reliably complete web-based tasks. It targets researchers and developers building sophisticated AI agents that require complex decision-making and learning mechanisms for web interaction.

How It Works

The project supports multiple agentic architectures, including planner-navigator multi-agent systems, solo planner-actor agents, and actor-critic multi-agent setups. A key differentiator is the integration of Monte Carlo Tree Search (MCTS) based reinforcement learning with Direct Preference Optimization (DPO) fine-tuning, enabling agents to learn and improve their web interaction strategies.

Quick Start & Requirements

• Install dependencies: poetry install
• Requires a Chrome instance running with remote debugging enabled (specific commands provided for macOS, Linux, and Windows).
• Requires OpenAI and Langfuse API keys configured in a .env file. Langfuse tracing is mandatory by default, but can be bypassed with code modifications.
• Run agent: python -u -m agentq run
• Evals: python -m test.tests_processor --orchestrator_type fsm
• Generate DPO pairs: python -m agentq.core.mcts.browser_mcts

Highlighted Details

• Implements the research paper "Agent Q: Advanced Reasoning and Learning for Autonomous AI Agents".
• Supports multiple agent architectures for flexible task completion.
• Integrates MCTS-based RL with DPO fine-tuning for enhanced learning.
• Includes tools for evaluation and generating data for reinforcement learning.

Maintenance & Community

The project is inspired by several notable research papers in the agentic systems space, indicating a connection to active research communities. Specific contributor or community links (e.g., Discord, Slack) are not provided in the README.

Licensing & Compatibility

The repository's license is not explicitly stated in the provided README text. Users should verify licensing terms for commercial use or integration into closed-source projects.

Limitations & Caveats

The setup requires specific browser configurations and mandatory integration with Langfuse for tracing, which might be a barrier for some users. The project appears to be research-oriented, and stability or production-readiness is not explicitly guaranteed.