MemoryCustodian

MemoryCustodian is a tool designed to provide durable project memory for AI coding agents. It aims to enhance the capabilities of agents such as Codex, Claude Code, and Gemini by giving them a persistent and accessible memory of project decisions, constraints, and context. This memory is stored directly within the project's repository as plain Markdown files, making it transparent, manageable, and shareable. The core problem MemoryCustodian addresses is the ephemeral nature of AI coding agent memory. Without a persistent memory, agents tend to forget crucial project details, decisions, and constraints with each new session. This leads to inefficiencies, repeated mistakes, and a lack of continuity in development. Important information often gets lost in lengthy conversation histories or requires manual re-entry, which is unsustainable for complex projects. One of the key features of MemoryCustodian is its repo-native storage of memory. Unlike solutions that rely on external databases or cloud services, MemoryCustodian stores all memory as plain Markdown files within the repository itself. This means that decisions, constraints, rejected approaches, and project context can be reviewed, versioned, shared, and even deleted using familiar Git workflows, just like any other piece of code. This approach ensures that memory is always aligned with the codebase it pertains to and is easily auditable. Another significant capability is the selective loading of memory through a manifest. Instead of injecting the entire project history into every prompt, which can lead to context window limitations and performance issues, MemoryCustodian uses a manifest to load only the memory relevant to the current task. This ensures that the agent has the necessary context without being overwhelmed, making the process more efficient and effective. The system also emphasizes the reviewability and versioning of memory. Since memory is stored in plain Markdown, developers can easily inspect, edit, and track changes to the memory files over time. This transparency is crucial for understanding how an AI agent's behavior evolves and for maintaining a clear history of project decisions. The ability to version memory alongside code ensures that specific decisions are tied to the code versions they influenced. MemoryCustodian is built with cross-agent compatibility in mind. It explicitly supports popular coding agents like Codex, Claude Code, and Gemini, as well as generic coding-agent workflows. This flexibility allows developers to integrate MemoryCustodian into their existing toolchains and leverage its memory management capabilities across different AI assistants. The overall approach of MemoryCustodian is to treat project memory as a first-class citizen, akin to source code. By storing memory in plain text files within the repository and managing it through Git, the tool provides a robust, transparent, and version-controlled system for AI agent context. The use of a manifest for selective loading further optimizes the process by ensuring that only pertinent information is accessed for each task. The benefits for users include improved efficiency, reduced errors, and enhanced consistency in AI-assisted development. By remembering past decisions and constraints, agents can avoid repeating mistakes, generate more contextually relevant code, and contribute to a more streamlined development process. The transparency and versioning of memory also foster better collaboration and auditing capabilities. Concrete use cases for MemoryCustodian include maintaining architectural decisions across long-term projects, ensuring that AI agents adhere to specific coding constraints, recalling rejected implementation paths to avoid redundancy, and providing agents with the necessary context for complex refactoring tasks. For example, an agent working on a new feature can be reminded of a previous decision to avoid a particular library due to licensing issues, thus preventing a potential conflict. MemoryCustodian is an open-source, local-first tool, making it free to use and prioritizing data privacy and control. It is designed for developers and teams who utilize AI coding agents and require a reliable method for managing their persistent memory. The tool integrates seamlessly with Git workflows, making it accessible to anyone familiar with version control systems. In summary, MemoryCustodian offers a novel and practical solution for AI coding agent memory by leveraging the power of repository-native, plain text storage and selective context loading, thereby enhancing the efficiency, consistency, and manageability of AI-assisted software development.