Skip to content

Research memory kit

  • Full-text search over past Claude Code and supported Codex sessions.
  • A project brief injected automatically at session start through Claude Code's hook.
  • One-line commands for decisions and check-ins in plain project files.

In our own usage, the parts that needed manual upkeep went unused; the automatic parts survived. This kit keeps that lesson: search updates its own index, the hook supplies context, and the agent skill records decisions and stopping points. There is no dashboard to maintain and no server to run.

Install

Clone the course repository using its clone URL, then enter the checkout:

git clone https://github.com/murnanedaniel/ai-agents-physics-research.git ai-agents-physics-research
cd ai-agents-physics-research
bash kit/research-memory/install.sh --apply-hook

Requires Python 3.9+ on Linux or macOS, with no extra packages. Ensure ~/.local/bin is on PATH for Claude Code. The installer backs up existing settings and merges the hook without duplicates. Add --project to configure only the current project. Without --apply-hook, it prints the hook to copy. Codex users can copy the AGENTS.md instructions from the kit README into their project.

First five minutes

Run these from your research project:

rmem index
rmem search "something you remember"
rmem init
rmem decide "Use the measured baseline" --why "It includes detector noise"

Fill in PROJECT.md once. The skill keeps Status and Next steps current and logs approved decisions and check-ins. Search with --here to focus on this project. Context reads the brief, recent decisions and check-ins without indexing.

Privacy

Everything runs locally, with no network access. The index lives in ~/.local/share/research-memory/index.db (override with RMEM_HOME). Roots default to ~/.claude/projects and ~/.codex/sessions; use colon-separated RMEM_ROOTS to change them. Secret redaction is best effort, so treat the index like the transcripts. Never put secrets in project files.

Uninstall

Remove ~/.local/bin/rmem, ~/.claude/skills/research-memory and the state directory. Remove the rmem context hook from ~/.claude/settings.json (or the project's .claude/settings.json) and any AGENTS.md instructions you added. Project files are yours to keep or delete.

Back to the deck ↩