Show HN: Stirrup – A lightweight and customizable foundation for building agents (github.com)

Stirrup has launched a new lightweight framework designed for building customizable AI agents. Unlike existing frameworks that often dictate rigid workflows, Stirrup allows the model to determine its own strategies for task completion, enhancing performance and adaptability. The framework incorporates best practices gleaned from leading agents like Claude Code and Codex, ensuring efficient context management and providing essential tools such as web browsing, local execution, and document handling. Users can either utilize it as a package or as a starting template for their own agents. The significance of Stirrup lies in its flexibility and ease of adoption for AI/ML developers. It features a generic Tool class for easy tool creation and context management, which automatically summarizes conversation history when limits are approached. Pre-built support for OpenAI-compatible APIs and multimodal capabilities allow agents to process various formats, including images and videos. These attributes position Stirrup as a comprehensive foundation for developing dynamic agents tailored to specific needs, ultimately driving innovation and efficiency in the AI/ML community.