Skip to content

memex

Memex

A durable, local memory layer for AI coding agents. Memories are readable Markdown files under ~/.memex/docs/; a rebuildable SQLite index makes them searchable. Memex brings relevant pages into coding sessions and records where they came from.

Get started

Requires Python 3.12+ and uv.

  1. Install the CLI from Git:

    uv tool install git+https://github.com/phanijapps/memex.git

  2. Connect your coding agent from your project directory:

    memex install claude   # or codex, pi, copilot

    Run memex install without a name to choose interactively.

  3. Store and find a memory:

    memex write --type preference --title "Deploy on Fridays" \
    --body "The team deploys to production on Fridays only."
memex recall "deploy"

    Use --scope project for workspace knowledge; Memex derives the project identity from the current directory. The guide explains global and project scope.

  4. Open the dashboard with memex viz to browse memories, sessions, projects, and health on your own machine.

Memex dashboard overview showing memory counts and recent pages

The Memories view shows scope and type filters. Both screenshots use sample data.

To disconnect a harness, run memex uninstall <name> from the project where you installed it. To remove the CLI, run uv tool uninstall memex. Your memories and transcripts remain under ~/.memex/.

How Memex works

  • Pull: MCP tools let the model search and write memory when it chooses.
  • Push: Harness hooks inject relevant memories and capture transcripts.
  • Proof: memex verify checks store health and can require memory activity in CI.

The filesystem is the source of truth. The index can be rebuilt from Markdown pages, and captured sessions link memories to their source conversations.

Where to go next