AiMatch

Rutter

Rutter records AI-session decisions and the Markdown notes each session cited, then makes those records available through MCP. It is for individuals using supported AI clients with a Markdown knowledge vault who want to recall past work across tools.

One of 129 tools in Research

Visit Rutter Compare alternatives Own this tool? Claim it
Rutter screenshot

Who it's for

  • Researchers or writers using a Markdown knowledge vault who want to recall decisions from research, drafting, or planning sessions.
  • People configuring Claude Code, Grok, Codex, or Antigravity who want one personal memory shared across those clients.
  • Developers comfortable configuring an AI client who want append-only session records and hashed references stored in their own notes folder.

Not the right fit if…

  • Teams who need a shared multi-user memory overlay: Rutter’s memory is single-user and entries carry no user identity.
  • People who need semantic search: semantic search is a stub, and the current search uses SQLite FTS5 keyword matching.
  • Organizations requiring a supported product or support commitment: Rutter is a reference implementation with no support commitment.

How it fits your workflow

Set up Rutter against a Markdown notes folder and configure a supported AI client’s capture hook. At the end of a turn, the client emits a summary and cited paths; Rutter appends the record to a dated file and computes hashes for the cited files.

An MCP client can search notes, retrieve a note, review recent records, or recall positions. References that cannot be matched are shown as unresolved rather than silently dropped; position results reflect the last reindex.

Pricing

Free: The page gives no paid pricing information and describes Rutter as MIT-licensed and published as a reference implementation.

The vendor doesn't publish prices. Check with Rutter directly.

Prices checked on Oct 11, 2026 from the vendor's site. They can change; confirm before you buy.

Key features

Append-only session records
A supported client can write a short summary and cited note paths at the end of a session. Rutter appends the record to a dated file in the notes directory rather than rewriting earlier records.
Hashed note references
Rutter computes a SHA-256 hash of each cited note’s bytes when the record is captured. When a note is renamed without content changes, identity matching can bind the old reference to the new path; unresolved references are shown rather than silently discarded.
MCP recall tools
The server exposes read-only tools for ranked full-text search, retrieving a note by path, listing recent sessions, and finding recorded positions by topic or related text.
Position capture and recall
Rutter can record and retrieve positions over time, including when a stance was formed and last revised. Positions are based on the last reindex and become available after the next reindex.
Markdown records in the user’s vault
Records live in a _librarian folder inside the user’s notes directory as plain Markdown. The SQLite FTS5 index is a disposable cache that can be rebuilt from the notes and records.
Prior-engagement annotations
Search results can include an annotation when a past session engaged with a result. The annotation is additive and does not change search ranking.

Works with

  • Claude Code
  • Grok
  • Codex
  • Antigravity (agy)
  • MCP clients

Limitations to know

  • A record exists only if the client writes a suitable line and the hook finds it; capture is not guaranteed to be complete.
  • A cited path records what the client listed, not proof that the model read the note or used it to reach its conclusion.
  • Rutter does not currently report when a note’s contents change while its path stays the same.
  • Semantic search is a stub; search currently uses SQLite FTS5 keyword matching.
  • Memory records are single-user and carry no user identity; sharing the memory overlay between people is out of scope.
  • Rutter is a reference implementation with no support commitment or guarantee that future changes will preserve behavior.

Alternatives to consider

  • Knowl

    Choose it if you want a persistent memory system for AI coding agents that records provenance, evidence, reasoning, and supersession for each fact rather than Rutter’s session summaries and cited-note references.

  • Recalld

    Choose it if you want a memory layer that stores conversations and documents and reconciles new facts, rather than Rutter’s append-only records of decisions and cited-note hashes.

  • Atlaso

    Choose it if you want automatically captured, shared persistent memory across a broader set of AI tools, rather than Rutter’s personal memory overlay with hooks supported for four clients.

Compare Rutter side by side

Getting started

A Markdown notes folder and an AI client configured for Rutter; supported capture hooks are listed for Claude Code, Grok, Codex, and Antigravity (agy).

  1. Choose a supported client and follow its setup path in getting-started.md.
  2. Configure the client’s capture hook and connect it to the notes directory.
  3. Run the capture check and recall check described in the setup instructions.