freeact: A Revolutionary Agent Library for Code-Driven Actions

Empowering Language Models with Executable Code

freeact is a lightweight library that transforms language models into autonomous agents by empowering them to execute code actions directly. This innovative approach breaks free from traditional constraints and enables agents to solve complex problems with unprecedented flexibility and power.

"freeact's code-based action paradigm unleashes a new level of agency for language models, allowing them to tackle real-world challenges with dynamic problem-solving capabilities." - Dr. Athena Le, AI Research Lead

Key Features for Building Sophisticated Agents

freeact empowers agents with a range of advanced capabilities:

• Autonomous learning and refinement through feedback and experience
• Storage and reuse of successful code actions as custom skills
• Extensibility and transparency for fine-grained control over agent implementation

Secure Execution and Customization

freeact utilizes ipybox, a secure execution environment, to safely run code actions while providing access to the entire Python ecosystem. Its modular architecture enables developers to customize agents with specific skills, such as generative Google Search or API integration.

Evaluation and Benchmarking

freeact has been evaluated using state-of-the-art language models on a comprehensive benchmark dataset. Comparison with alternative approaches demonstrates the library's effectiveness, even without prompting.

Getting Started with freeact

To install freeact:
pip install freeact

Launch an agent with the CLI:
python -m freeact.cli --model-name=claude-3-5-sonnet-20241022 --ipybox-tag=ghcr.io/gradion-ai/ipybox:basic --skill-modules=freeact_skills.search.google.stream.api

Interact directly with agents and witness their abilities in action!