Autopus-ADK is a Go-based open-source project that acts as an Agent Development Kit (ADK) for AI coding agents. It facilitates multi-model orchestration, allowing agents to work together through consensus, pipelines, or debate mechanisms. The framework emphasizes 'Agent Experience' (AX), designing project documentation and code structure to be easily parsed and understood by AI agents. Key features include Architecture-as-Code for defining system structure, Lore decision tracking to maintain institutional memory across sessions, and a SPEC/EARS engine for managing requirements. The project aims to overcome common limitations of single-agent AI coding tools, such as the inability to handle multi-file refactoring, lack of testing and security considerations, and loss of context between sessions. Autopus-ADK enforces a 300-line limit per source file to optimize agent processing and provides structured markdown rules and YAML frontmatter skills for agent-friendly interaction. The latest release is v0.50.27.
Autopus-ADK: A Multi-Model AI Agent Orchestration Framework for Code Development
Autopus-ADK is an open-source framework designed to orchestrate multiple AI coding agents, enabling them to function as an engineering team. It supports various models like Claude Code and Codex, providing features such as multi-model orchestration (consensus, pipeline, debate), architecture-as-code, and decision tracking. The framework aims to improve AI agent performance by providing structured context and workflows.
Why it matters
Autopus-ADK addresses key challenges in AI-driven code development, such as platform lock-in, inconsistent code quality, and lack of institutional memory for AI agents. By enabling multi-agent collaboration and providing structured environments, it could lead to more reliable, maintainable, and secure code generated by AI, making AI coding tools more practical for complex projects.