Rutter, a new open-source tool published on Hacker News, aims to solve the fragmentation of AI assistant memory by creating a portable, tool-agnostic record of your coding decisions. Unlike vendor-specific memory features that vanish when you switch platforms, Rutter stores session summaries as plain Markdown files within your existing notes directory. The tool was released on October 10, 2026, and currently supports Claude Code, Grok, Codex, and the Antigravity command-line tool, allowing developers to maintain a continuous thread of thought across different AI environments without relying on proprietary cloud storage.

The Mechanics of Portable Memory

The core innovation of Rutter is its use of cryptographic hashes to bind AI decisions to the specific state of the codebase or notes at the moment of capture. When an AI session concludes a task, a 'Stop hook' intercepts the final output and writes a one-line summary into a dated Markdown file in a _librarian/sessions/ directory. Crucially, Rutter calculates the SHA-256 hash of every file cited in that decision itself, rather than trusting the AI client to report it. This ensures that the memory record remains accurate even if files are renamed or moved, as the tool can follow a file's new path if the content hash remains unchanged.

Limitations and Honest Design

The documentation for Rutter is refreshingly blunt about its limitations, explicitly stating that it is a reference implementation with no support commitment. The tool does not currently flag when a note is edited in place at the same path, meaning a decision might be based on outdated information without warning. Furthermore, Rutter does not verify that the AI model actually read the cited files; it only records that the client listed them. The design deliberately avoids using a model to interpret stale references, instead marking them as 'unresolved' and leaving the judgment call to the human developer.

Why This Matters for Builders

For developers juggling multiple AI assistants, Rutter offers a way to escape the 'vendor lock-in' of context windows. Most AI tools consolidate memory by merging or dropping old entries, which destroys the historical sequence of your reasoning. Rutter uses an append-only log, preserving every decision exactly as it was made, including the wrong turns. This allows for a more accurate audit trail of your architectural choices, turning ephemeral chat logs into a durable, greppable knowledge base that lives alongside your actual code.

Key Takeaways

  • Rutter uses plain Markdown and SHA-256 hashes to create a portable memory layer for AI coding tools.
  • The tool currently supports Claude Code, Grok, Codex, and Antigravity via Model Context Protocol (MCP).
  • Memory is append-only and stored locally in your notes folder, avoiding vendor-specific consolidation that flattens historical context.
  • The tool is MIT-licensed but explicitly lacks support for detecting in-place file edits or verifying if an AI actually read a cited file.

The Bottom Line

Rutter is a pragmatic solution for the messy reality of using multiple AI assistants. It trades the convenience of automatic semantic cleanup for the reliability of an immutable, hash-verified audit trail, making it a valuable tool for serious builders who need to track their reasoning over time.