Skip to main content
Glama

Geniuz

Your AI starts every conversation from zero. Geniuz keeps what matters, the calls you made and the reasons behind them, on your own computer, where any AI you use can find it.

Free and open source. No account, no cloud, no telemetry.

Install

You have

Do this

A Mac (Apple Silicon, Sonoma or later)

Download Geniuz.dmg and drag it to Applications

Windows 11

Download Geniuz-Setup.exe

Linux (x86_64 or arm64)

Run the command below

curl -fsSL https://raw.githubusercontent.com/jackccrawford/geniuz/main/install.sh | bash

Coming from an earlier version

Remove the old version first, then install this one as above. Your memories aren't stored in the app, so they stay where they are and appear in a folder called My memory.

You have

Remove it with

A Mac

Quit Geniuz and drag it to the Trash

Windows 11

Settings → Apps → Geniuz → Uninstall. If you picked your own data folder, pick it again when you install

Linux

Nothing to remove: the install command replaces the old version

Don't delete .geniuz in your home directory. That's where your memories live.

Related MCP server: potluck

Connect your AI

You use

Do this

Claude Desktop

Connects by itself when Geniuz opens. Restart Claude Desktop once. To connect by hand: geniuz mcp install

Anything else that speaks MCP

Add geniuz mcp serve as a server

Anything that can run a command

geniuz skill prints a guide to give your agent

Then keep talking. Your AI saves what matters as you go. Tomorrow, in a new chat, ask it what you told it today.

Try it

geniuz keep -c "Acme wants invoices rounded down. They failed an audit over it in 2024." -g "client: Acme rounding"
geniuz recall "how does Acme want amounts rounded"

geniuz folder create Work -d "Client projects. No passwords here."
geniuz keep --in Work -c "Dana is the contact at Acme." -g "client: Acme contact"
geniuz recall --all "who do I talk to at Acme"

The first keep downloads a small model (about 130 MB) that runs on your computer.

What makes it different

  • It's yours. Your memories are files on your computer, and search runs there too.

  • It isn't tied to one AI. Change tools and your memory comes with you.

  • Nothing is overwritten. A correction is added beside the original, so you can trust what you find.

  • It finds things two ways. By meaning, for what you can't quite name, and exactly, for names and numbers.

  • Folders keep things apart. Work, Home, one per project, or a place agents leave each other messages.

The idea, if you only take one thing

One table: a UUID, a JSON payload, a parent, a time. Rows are added and never changed. That shape holds far more kinds of work than you'd expect, and it spares you most of the design a normalized schema demands up front.

Borrow it and tailor it. Make something updatable only where a concrete need says so, and write that need down beside the schema. Ours is schema/schema.sql.

Learn more

MIT licensed. See LICENSE.

Available Tools

3 tools
foldersA

List the folders you can use: each is a separate store of memories with its own purpose. Check a folder's description before writing to it, since it may say what belongs there. Pass a name to read one folder's whole description. Pass create to make a new folder, when the person wants memories kept apart.

Inputs: face, name, create, filter, +1 more.

keepA

Saves a note for later chats: a decision and why, a preference, project context, what to pick up next. It's found later by meaning, not exact words. Each note goes in a named folder, with a one-line gist.

Inputs: gist (required), folder (required), content (required), images, +1 more.

recallA

Finds notes kept in earlier chats, by meaning: 'budget priorities' finds a note about 'retention focus, $40K' though the words differ. With no query it shows the most recent notes, which is a quick way to see what is already known when a conversation starts.

Inputs: all, full, grep, image, +6 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.6/5.0

    Scored across 3 tools

    Disambiguation4/5

    recall (read by meaning) and keep (write) are clearly opposite operations, and folders is the only tool touching folder metadata. The mild wrinkle is that folders multiplexes list, describe, and create behind an argument, which could momentarily confuse an agent expecting a separate create tool.

    Naming Consistency3/5

    Two tools are verbs (recall, keep) while folders is a noun, so the naming convention is mixed rather than a predictable verb_noun or all-noun pattern. The names are short and memorable, but they don't form a consistent scheme an agent could extrapolate from.

    Tool Count4/5

    Three tools sit at the low end of the ideal 3-15 range but each earns its place: read, write, and organize. The surface is deliberately minimal for a memory server rather than padded.

    Completeness3/5

    Read (recall), write (keep), and folder listing/creation are covered, but there is no way to update, correct, or delete a note, and no way to rename or remove a folder. Those gaps will bite an agent asked to fix or forget a stored memory.

    Maintenance

    ActivityNo data
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      A
      maintenance
      Privacy-first local document search using semantic search. Runs entirely on your machine with no cloud services, supporting PDF, DOCX, TXT, and Markdown files.
      9
      4,071 npm
      436
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Privacy-first personal knowledge database for your AI that ingests data exports (Google Takeout first) and exposes them via MCP tools like search and list items. Runs locally with no cloud or telemetry.
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables AI agents to maintain persistent, local memory with retrieval-augmented search, knowledge graphs, and context surfacing, without any cloud dependencies.
      276 npm
      MIT