AI coding agents often struggle with session amnesia, limited context windows, and a lack of awareness across multiple code repositories. Code Relay provides a structured protocol to address these issues, enabling agents to maintain persistent memory across conversations, understand global project structures, and operate with safe, scoped execution. This protocol empowers developers to integrate AI agents more effectively into complex, multi-repository software development workflows.

How It Works

Code Relay organizes multiple Git repositories under a unified workspace root, granting AI agents a comprehensive global view of the project. Key concepts include a Boot Sequence for automatic context restoration upon agent startup, and HANDOFF/CHECKPOINT mechanisms that replace lossy context compression with structured, proactive state snapshots, ensuring zero information loss across sessions. The SCOPE feature (in local mode) implements a whitelist for agent write access, preventing unintended modifications, while Worktree Isolation ensures each task develops in a clean, separate environment. This approach offers a robust alternative to context compression, preserving critical reasoning and enabling agents to navigate intricate, multi-repo project landscapes.

Quick Start & Requirements

To begin, copy either the github/ or local/ directory structure into your project. Launch your AI coding agent, pointing it to the AGENTS.md file within the copied structure. The agent self-boots, loads configurations, and prompts for project details on first use, automatically generating necessary resource maps and task definitions. Subsequent launches restore context seamlessly. Requires an AI coding agent capable of reading project files (e.g., OpenCode, Cursor, Claude, Code Windsurf, GitHub Copilot).

Highlighted Details

• Native Multi-Repo Support: Agents gain a global perspective across independent repositories (frontend, backend, etc.), understanding interdependencies and enabling cross-repo search and refactoring.
• Zero-Loss Context Transfer: HANDOFF/CHECKPOINT documents proactively capture and restore complete task state, avoiding information loss inherent in context compression.
• Scoped Execution (Local Mode): The SCOPE feature enforces a strict whitelist for file modifications, enhancing agent safety and preventing unintended changes.
• Dual Collaboration Modes: Offers seamless integration with GitHub (using Issues/Project Boards for state management) or a fully local, platform-independent file-based state management.

Maintenance & Community

The provided README does not detail specific contributors, sponsorships, or community channels (e.g., Discord, Slack).

Licensing & Compatibility

• License: MIT License.
• Compatibility: The permissive MIT license facilitates commercial use and integration into closed-source projects. Compatible with any AI coding agent that can parse Markdown and YAML project files.

Limitations & Caveats

Code Relay is a protocol specification requiring integration with a compatible AI coding agent; it is not a standalone tool. Its effectiveness hinges on the agent's ability to accurately interpret and act upon the structured protocol files. While local mode offers explicit write controls, GitHub mode's safety relies on GitHub's permission structures and issue management.