halilhp

  For those new here: every Claude Code or Cursor session starts from zero.
  MemoTrail auto-indexes all your past sessions and makes them searchable.
  Two commands to set up:

    pip install memotrail
    claude mcp add memotrail -- memotrail serve

  Then ask: "Why did we switch to Redis?" — and it finds the exact session.

  What's new in v0.3.0:

  • Cursor IDE support — MemoTrail now indexes Cursor chat history too
    (parses state.vscdb files). Not just Claude Code anymore.

  • BM25 keyword search — v0.2 only had semantic search. Now there's also
    exact keyword matching (great for error messages, function names) and a
    hybrid mode that combines both using reciprocal rank fusion.

  • Auto session summaries — every session gets a human-readable summary
    automatically. No API keys, no LLM calls — pure heuristic extraction.

  • Auto decision extraction — detects architectural decisions from conversations
    ("decided to use X", "let's switch to Y") and categorizes them. Months later
    you can ask "what tech decisions did we make?" and get real answers.

  • Real-time file watching — new sessions are indexed instantly via watchdog.
    No more restarting the server to pick up new conversations.

  • 3 chunking strategies — token-based (default), turn-based (groups
    user+assistant pairs), and recursive (splits on paragraph/sentence boundaries).
    Different strategies work better for different use cases.

  • VS Code extension — search, index, and view stats directly from VS Code
    sidebar without touching the terminal.

  • 69 tests — up from 12. Covers all new modules.

  Still 100% local. No cloud, no API keys, no data leaves your machine.
  MIT licensed.

  What hasn't changed: ChromaDB + SQLite storage, all-MiniLM-L6-v2 embeddings
  on CPU, MCP protocol, ~/.memotrail/ for all data.

  Next up: GitHub Copilot collector, cloud sync (opt-in), and team memory sharing.

  Repo: https://github.com/HalilHopa-Datatent/memotrail
  PyPI: pip install memotrail

  MemoTrail runs as an MCP server, automatically indexes your past sessions,
  and makes them semantically searchable. It's completely local — no cloud,
  no API keys.

  Currently it supports Claude Code only, but Cursor and Copilot collectors
  are planned. Would love to hear what features you'd find most useful.

  Happy to answer any questions!

  MemoTrail is an MCP server that gives your AI coding assistant persistent
  memory. It automatically indexes every past conversation, embeds them locally
  using sentence-transformers, and makes everything semantically searchable.

  Setup is two commands:

    pip install memotrail
    claude mcp add memotrail -- memotrail serve

  After that, you can ask things like "Why did we choose Redis?" and the AI
  will find the relevant context from any past session — even months ago.

  How it works:
  - Reads Claude Code session logs from ~/.claude/
  - Chunks conversations into meaningful segments
  - Embeds with all-MiniLM-L6-v2 (~80MB, CPU only)
  - Stores vectors in ChromaDB, metadata in SQLite
  - Exposes 6 MCP tools (search_chats, get_decisions, save_memory, etc.)

  Everything runs locally — no cloud, no API keys, no data leaves your machine.
  MIT licensed.

  Currently supports Claude Code. Cursor and Copilot collectors are on the roadmap.

  I built this because I kept losing context between sessions. Would love feedback
  on the approach and what features would be most useful.