Mithrify

How Mithrify works

Eight stages between disposable chat history and a local, user-owned memory layer. Your AI tools remember inside their own walls; Mithrify remembers across them.

  1. Import Conversations and Sessions

    Bring in the work where it happened. A Conversation is the complete interaction with an AI system; a Session is a bounded period of work. Claude Code is the initial importer focus, and additional importer support rolls out progressively through early access. Voice dictation rides along as a supporting input.

  2. Preserve the original source

    The complete original conversation or session lands on your disk, locally controlled. The Source Archive keeps the evidence behind every note, and machine metadata stays out of sight under .mithrify: no YAML frontmatter cluttering your notes, and no vendor between you and your own history.

  3. Crystallize into Markdown

    Long threads are distilled into clean working notes: decisions, solutions, architecture, prompts, bugs fixed, open questions. The Source Archive keeps everything; crystallization keeps what you'll need again. Each crystallized note is derivative working memory, not a replacement for the source.

  4. Review in Inbox

    New crystallized memory enters Inbox for review before it is routed into your permanent Stratum structure. Mithrify preserves the result immediately, but it does not place AI-generated memory into your permanent organization without review.

  5. Organize into your Stratum

    Your Stratum is the local memory base where approved Markdown knowledge, source-linked context, and organized project continuity remain under your control.

  6. Connect related work through Topics

    A Topic connects related Conversations, Sessions, decisions, and source records over time, so project context survives when one chat ends and another begins.

  7. Search and Recall

    One search across everything you've worked through with AI: previous decisions, bug fixes, prompts that worked, architecture changes, tradeoffs, and research. Crystallized knowledge stays traceable to its source evidence.

  8. Serve context to agents

    Bounded, source-linked Context Packs will be prepared for future AI use, and a read-only MCP Memory Server will serve approved context back to compatible agents, so the next session starts where the last one ended instead of from zero.

Build the memory layer your AI tools are missing.

Import your AI work. Crystallize what matters. Recall it when the next session starts.