Zerostack – An agent that's less sloppy than Elon's attempts (github.com)

Zerostack, a new coding agent, was recently announced, boasting a more efficient architecture compared to existing solutions. It supports multiple providers such as OpenAI and Anthropic and features a comprehensive permission system with five configurable modes. Key functionalities include session management tools, a terminal user interface, and the ability to switch between system prompt modes at runtime. With a minimal RAM footprint of around 16MB and low CPU usage, Zerostack aims to be one of the smallest yet most performant coding agents available, containing approximately 30,000 lines of code. This development is significant for the AI/ML community as it presents a viable alternative to larger agents, emphasizing efficiency and ease of use. Its design supports advanced features like integrated Exa search for web tools, persistent memory for context storage, and an innovative advisor model that brings additional strategic guidance during active sessions. Zerostack's approach to permission management and session workflows enhances its usability in coding environments, making it a powerful tool for developers looking for performance without sacrificing capability.