Polygraph

Polygraph is a meta-harness designed to enhance the capabilities of AI coding agents by providing them with crucial context that is typically missing. It addresses the limitations of current AI agents, which often operate with a fragmented understanding of a codebase due to their inability to see across repository boundaries. Polygraph aims to equip these agents with a comprehensive model of the entire codebase, enabling more autonomous and effective work. The core problem Polygraph solves is the inherent limitation of AI coding agents in understanding the full scope of a project when it spans multiple repositories. This lack of holistic visibility leads to inefficiencies, errors, and a reliance on manual context-setting for the AI. Without a unified view, agents struggle to make decisions that maintain code integrity across different parts of a system, especially when changes in one repository impact others. One of Polygraph's key features is its ability to create a unified dependency graph across all connected repositories, both private and public. This is achieved without the need to move any code, preserving the existing project structure. The system indexes repositories semantically, allowing agents to discover and include relevant repositories for their tasks within a session. This cross-repo visibility is crucial for understanding how changes in one area might affect others. Another significant capability is Polygraph's session memory management. It ensures that memory survives beyond a single session, combating 'agent amnesia.' This means developers can resume, reference, or build upon any session initiated by any other developer, on any machine, and even across different agent frameworks. This persistent memory allows for continuity in complex development tasks and facilitates seamless handoffs between team members. Polygraph also facilitates the management of changes across multiple repositories. After modifications are made, Polygraph can create pull requests across all affected repositories simultaneously. It monitors CI statuses in a unified manner and coordinates follow-up actions. Furthermore, it can link npm packages, enabling testing of changes from one repository within another that depends on it, ensuring a cohesive development workflow. The product operates by first establishing a dependency graph of all repositories. It then semantically indexes these repositories to enable agents to find relevant code. When an agent needs to learn or make changes across multiple repositories, Polygraph checks out the necessary code locally and delegates sub-agents to work within them. This approach ensures that agents have access to the actual code, not just embeddings, and can even incorporate open-source repositories into their sessions. The benefits for users include significantly improved autonomy for AI coding agents, reduced context-switching, and enhanced collaboration. Developers can work on features that span multiple repositories without losing context, and team members can seamlessly pick up where others left off. This leads to faster development cycles and fewer integration issues. Concrete use cases for Polygraph include individual developers working on features that touch three different repositories, where Polygraph sets up a single session, manages cross-repo CI, and records all agent actions. For teams, if a change in a shared library affects five downstream repositories, Polygraph allows an agent to validate the change across all of them before any pull requests are opened, managing all PRs and CI as a single unit. Polygraph is designed for individual developers and teams working with multiple repositories. While specific integrations and tech stack details are not explicitly listed, the content implies compatibility with various agent frameworks like Claude Code and Codex. The product is offered for free. The platform is web-based. In summary, Polygraph empowers AI coding agents with the essential cross-repository visibility and persistent memory they need to operate more autonomously and effectively, streamlining complex development workflows across multiple codebases.