MemoryCustodian is an open-source, local-first developer tool designed to provide repo-native memory for AI coding agents such as Codex, Claude Code, Gemini, and others. Instead of relying on a hosted service or bloating your prompts with redundant context, MemoryCustodian stores project decisions, constraints, rejected approaches, and architectural context as plain Markdown files directly inside your repository.
By treating AI memory like standard code, it allows development teams to review, version-control, share, and delete project context using familiar Git workflows. Additionally, an intelligent manifest system ensures that only the memory specifically relevant to each individual task is loaded, keeping your agent prompts clean, efficient, and targeted.
MemoryCustodian is an open-source, local-first developer tool that gives AI coding agents like Codex, Claude Code, and Gemini durable project memory using plain Markdown files stored directly inside your repository.
It uses a manifest system that loads only the specific memory and context relevant to each individual task, avoiding the need to cram entire project histories into every prompt.
MemoryCustodian is cross-agent and supports popular coding assistants including Codex, Claude Code, and Gemini.
Project memory, decisions, constraints, and rejected approaches live as plain Markdown files in your repo, allowing you to review, version, share, and delete them just like regular code.
0 community reviews for MemoryCustodian
Help the community by sharing your experience. Your review helps others make better decisions.
Machine Learning Engineer
