Swival – A coding agent for open models (swival.github.io)

Swival, a new coding agent designed for use with open models, has been announced, offering a seamless way to interface with machine learning models via LM Studio or the HuggingFace Inference API. This innovative tool requires minimal setup, automatically discovering the loaded LM Studio model or allowing users to select models from HuggingFace. Unique for its simplicity, Swival is built with just a few thousand lines of Python code, making it highly hackable and easy to modify for specific workflows. The significance of Swival lies in its user-friendly, out-of-the-box functionality that enhances the efficiency of coding tasks. Key features include built-in tools for file editing, search, and structured problem-solving, along with robust benchmarking capabilities that generate JSON evaluation reports detailing performance metrics. This positions Swival as a powerful resource for developers and researchers, facilitating systematic comparisons of different models on real-world coding challenges and streamlining the development process. Users can quickly deploy commands to refactor code or engage in interactive sessions, exemplifying how accessible AI tools can drive innovation in programming workflows.