rcllr
One memory for Claude Code, Codex and other agents: Markdown notes in your own Git repo, one CLI, and an import of what each agent remembers on its own. Agents forget. rcllr recalls.
- TypeScript
- Node
- Git
- Markdown
What it does
rcllr gives Claude Code, Codex and other coding agents one long-term memory. The memory is a Git repository of Markdown notes, the brain. Agents read and write it through one CLI. Hooks commit every change. With a private remote, they push it to your other machines too.
Before an agent answers a question about earlier work, it asks the brain:
$ rcllr context "how does the release work"Ignored: how, does, thetopics/release-process/README.md | rev=c3cdd0ac3e2e | updated=2026-10-02 | status=active | scope=workKey fact: Releases ship on Tuesdays after the smoke test passes on staging.rcllr drops filler words before it searches. Every hit names its file and revision, so the agent can cite the note and update it without overwriting a newer version.
One memory across agents
Claude Code keeps a memory per project and machine. Codex keeps one in ~/.codex/memories/. rcllr import-memory mirrors both into the brain, so what one agent learned, the other can find. Imported notes rank below your own notes and never count as instructions.
Scopes and sensitive notes
Scopes keep work and private notes apart. Notes marked sensitive stay hidden from agents, even in search results, until you name the topic yourself. A brain without a remote never leaves your machine.
Status
Open source under the MIT license, installed with npm install -g rcllr. Up to version 0.4 the project was called dummkopf. Docs and quickstart live at rcllr.schmidbauer.dev.