PC Agent provides a framework for autonomous digital agents by transferring human cognition. It targets researchers and developers building AI agents capable of complex, multi-step computer tasks, enabling AI to operate independently on a user's machine.

How It Works

The system employs a three-component pipeline: PC Tracker for large-scale human-computer interaction data collection, a Cognition Completion post-processing stage to convert raw data into cognitive trajectories, and a multi-agent system with planning and grounding agents for decision-making and visual understanding. This approach aims to imbue AI agents with human-like cognitive processes for more robust task execution.

Quick Start & Requirements

• Install via conda env create -f environment.yml and conda activate pcagent.
• Requires Python environment setup via Conda.
• PC Tracker requires Windows for executable building (.\package.ps1).
• Cognition completion requires an OpenAI API key.
• See PC Tracker User Manual for detailed instructions.

Highlighted Details

• PC Tracker is presented as the first lightweight infrastructure for large-scale HCI data collection.
• Demo available showcasing autonomous computer control for complex, multi-step tasks.
• PC Agent-E, a new SOTA open-source model for Windows, has been released.

Maintenance & Community

• Project is actively developed with recent updates and releases.
• Citation details provided for academic referencing.

Licensing & Compatibility

• License details are not explicitly stated in the README.
• Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The PC Tracker component is specifically noted as requiring Windows for executable building. The reliance on OpenAI API keys for a core processing step may introduce external costs and dependencies.