EGC - Extended Global Context: Persistent Memory for AI Agents

Overview

EGC is a local-first runtime designed to give every AI coding tool a persistent memory. It ensures that AI agents never start from zero. This tool eliminates the need to memorize commands by automatically handling context, decisions, and preferences across sessions. Whether you are using Claude Code, Cursor, Gemini CLI, Windsurf, or other tools, EGC ensures a seamless workflow by saving what was learned at the end of a session and loading that state back at the start of the next.

Benefits

EGC offers several key advantages for developers who use AI tools for coding. First, it provides persistent memory. The system saves decisions made, failures encountered, user preferences, and next steps at the end of each session. At the start of the next session, it loads this state automatically without requiring any prompting. Second, it is language agnostic. Users can simply say let us continue or where did we stop in any language, and the AI will already know what to do. Third, it offers universal compatibility. It works with various AI models including Claude, GPT-4o, Gemini, and OpenRouter models like DeepSeek, Qwen3, and Llama 4. Finally, it is a one-install solution. A single installation covers multiple AI coding tools, ensuring consistency across different environments.

Use Cases

EGC is ideal for developers who work on projects over long periods or switch between different AI coding tools. When you open an AI coding tool on a project you have not touched in weeks, EGC automatically loads the previous session state. For example, if you were working on a TypeScript project, EGC might display the project stack, specific agents, skills, and pending tasks. This allows you to resume work seamlessly without re-explaining the project context. The tool is also useful for teams that want to keep all their AI tools in sync. By running egc watch once, any context edited in one tool automatically appears in others. This eliminates manual steps and stale state. Additionally, developers can use the included prompt library which contains 479 components written from real engineering sessions. These include 63 agents, 229 skills, 76 commands, and 111 rules.

Pricing

EGC is free to use. It is built by a single developer and maintained in the open. There are no subscription fees or hidden costs. To support the project, users can star the repository, share it with others, or become a monthly sponsor.

Vibes

The community response to EGC has been positive among developers who struggle with context loss in AI coding sessions. Users appreciate the ability to pick up where they left off without repetitive explanations. The fact that it works across multiple AI models and tools has been highlighted as a major strength. Developers value the human-readable Markdown format for state files, which makes it easy to inspect the memory manually if needed. The open-source nature of the project has fostered trust, as users can see exactly how the memory and safety systems work under the hood.

Additional Information

EGC is built by a single developer and is maintained openly. It relies on community support through repository stars and sponsorships to remain independent. The project includes two main Model Context Protocol servers that work together during every session. The first is the Memory Server which provides 14 tools for managing memory and context. The second is the Context and Safety Server which provides 5 tools to ensure safety and efficiency. State files are stored in plain Markdown format at a specific directory path, making them accessible and editable. The project emphasizes local-first design, meaning all data is stored on the user machine and not sent to external servers for processing.