What is rutter?
rutter is a personal memory layer for AI agents that records decisions, cited notes, and changes in thinking from AI sessions. It stores those records as dated Markdown files in a notes folder you own, where multiple agents can use the same memory. Capture hooks are available for Claude Code, Codex, Grok, GitHub Copilot CLI, Antigravity CLI, and Muse Code; other MCP clients can search and recall. It is intended for people who keep Markdown notes and are comfortable configuring AI clients.
rutter Features:
Session Capture: A client hook can store a record after each turn when the model supplies a line for capture.
Shared Agent Memory: Supported clients can capture and recall records from the same notes folder, so memory can travel between agents.
Cited Note References: Each record lists the notes cited in the session and includes a content hash taken when the record is captured.
Dated Markdown Records: Records are appended as dated Markdown files in a _librarian folder beside your notes, ready to inspect or commit to git.
Note Reindexing: Reindexing scans the notes folder and can resolve a renamed note when its stored content hash still matches.
MCP Recall: Any MCP client can be configured to search and recall records, though capture requires a working client hook.
rutter Benefits:
Continuity Across Agents: You can change AI clients without having to start over with decisions already recorded in the shared memory.
Decision History: Append-only records keep later decisions beside earlier ones, making it easier to see how your thinking changed.
Inspectable Memory: Because the records are ordinary files in your own folder, you can review them directly instead of relying on a vendor account.
Traceable Context: References to cited notes give you a trail to check alongside a remembered decision, rather than leaving only a summary.
Visible Uncertainty: When a reference cannot be placed after a note changes, rutter marks it unresolved instead of guessing at its location.
Use Cases:
Cross-Agent Projects: Use one record of project decisions when moving between coding agents such as Claude Code and Codex.
Research Continuity: Recall which notes informed a research discussion in a later session, including when you use a different MCP client.
Drafting Decisions: Keep a history of editorial choices, such as changing a newsletter from weekly to fortnightly, with cited notes.
Planning Work: Record and revisit planning decisions across sessions instead of relying on one agent's built-in memory.
Markdown Note Workflows: Add searchable agent recall to an existing Markdown notes folder while keeping the underlying records in files you control.