Agenten stärken
Installation vom Herausgeber. Kein PlugMCP-Endpunkt.
Host a copy with our keys and logs.
Noch kein paketiertes Release in der Registry. Folgen Sie der README des Repositorys.
Remote-URL oder ein Befehl auf Ihrem Rechner? Worin sich die beiden Installationen unterscheiden
Auf PlugMCP ausführen
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
StartenVom Herausgeber
Your AI is sharp. It also starts every conversation from zero. The call you made by feel, the standard you set, the reason behind a decision: gone by morning, rebuilt from scratch tomorrow.
Geniuz fixes that. One install, local, private, searchable by meaning. The context compounds instead of resetting at every boundary. Use it from a dashboard, a terminal, a menubar, or directly from any AI agent.
Your AI remembers now. Geniuz stores all your data locally on Mac, Windows, Linux, Raspberry Pi. No cloud. No account. No API keys. It's open source; you can read every line of code.
New in 2.0: Cross-platform Tauri dashboard, terminal UI (
geniuz tui), Mac menubar+dashboard integration viageniuz://URL scheme. See release notes →Note: Screenshots in this README are from a pre-2.0.5 build and will refresh with the next release. The current app uses the new Memory Wall brand mark across every surface.
Geniuz is one Rust core with several ways to reach it. Pick what fits the moment.
Browse memories, save new ones, search by meaning, see recent threads. Native vibrancy on macOS, system tray on Windows + Linux.
Surfaces: Memories · Remember · Find · Detail · Status · Data & Export · Settings.
geniuz tuiSame data, terminal-native. For developers and agents who live in a shell. Two-field compose for /remember, semantic search via /find, sort toggle via /reorder.
$ geniuz tui
Drill into any memory with /detail <prefix> (or /random to surface one at random):
Refuses to launch from non-interactive callers (TTY guard) so agents don't accidentally lock up on the alternate-screen mode.
Ambient presence. Memory count, recent activity, one-click to the dashboard. The menubar app is dock-less on Mac (LSUIElement). It's a residence, not a window.
Underneath everything, the geniuz command. Your agents call it from any shell, any framework.
geniuz remember -c "OAuth token refresh is async but middleware assumed sync. Swapped lines 42-47." -g "fix: auth token refresh — async ordering"
geniuz recall "authentication middleware"
Searched "authentication middleware," found a memory about "OAuth refresh" and "middleware ordering." The meaning matched. No re-investigation. No human re-explaining.
The full surface:
% geniuz --help
Start here: 'geniuz recent' to see what's in your folder.
GENIUZ: Your AI remembers now.
Persistent memory for AI agents. Three R's: remember, recall, recent.
Works with any agent framework — Claude Code, Cursor, Windsurf, Aider,
or anything that can run a shell command.
Usage: geniuz <COMMAND>
Commands:
remember Save a memory — what you learned, decided, or discovered
recall Search your memories — semantic by default
recent Show recent memories
capture Capture files or directories into your folder
watch Watch for new memories in real time
backfill Build embedding cache for semantic search
skill Show usage guide for agents
status Show folder stats
tui Launch the terminal UI for browsing your memories
dashboard Launch the graphical dashboard app
mcp MCP server for Claude Desktop — run, install, or check status
settings User settings — read or change preferences
help Print this message or the help of the given subcommand(s)
Options:
-v, --version
Show version information
-h, --help
Print help (see a summary with '-h')
Examples:
geniuz tui Browse memories in a terminal UI
geniuz remember -c "Fixed the auth bug" -g "fix: token refresh"
geniuz recall "auth" Semantic search
geniuz recent Latest memories
geniuz capture ./notes/ Bulk-load markdown files
geniuz backfill Build embedding cache
Folder: Defaults to .geniuz in your home directory
Override with GENIUZ_HOME to change the folder location
Override with GENIUZ_STATION for a specific memory.db file
Multiple agents can share a folder for shared memory.
Use "geniuz [command] --help" for more information.
This repository is Geniuz Free: MIT-licensed, single-station memory for any AI chassis. Install it, bring your own AI, get memory that survives across conversations.
Geniuz Team is the paid tier from mVara: unlimited continuity, perpetual agents that live in your environment, and a team chassis that coordinates around your customers' reality. Geniuz Team is built on the same substrate as Geniuz Free.
If you want memory for yourself or a developer team using existing AI tools, Geniuz Free is what you want.
If you want a configured team of agents living in your private cloud, remembering your work, and coordinating around your customers, that's Geniuz Team. Contact mVara for pilot inquiries.
Pick the path that matches your setup.
Download Geniuz.dmg, double-click, drag to Applications. Signed and notarized by Managed Ventures LLC. No Gatekeeper warnings.
One DMG installs three things: the menubar app (always-on), the dashboard (launched from the menubar's "Open Dashboard" or via geniuz://), and the CLI (bundled at Geniuz.app/Contents/Resources/geniuz).
Wire the CLI into your shell PATH if you want it on the command line:
sudo ln -sf /Applications/Geniuz.app/Contents/Resources/geniuz /usr/local/bin/geniuz
Or skip the DMG entirely and use the CLI-first install below. That path installs to ~/.geniuz/bin/ and adds itself to your PATH without sudo. Apple Silicon native; Intel Macs run via Rosetta 2 (universal binary coming).
Download Geniuz-Setup.exe Windows 11, per-user, no admin prompt. Signed via Azure Trusted Signing (Microsoft-rooted cert, no "unknown publisher" warning). For managed deployment it runs silently with /VERYSILENT /SUPPRESSMSGBOXES /NORESTART.
After install, the dashboard runs as a system tray app. Left-click → window; right-click → menu.
First-launch SmartScreen note: even with a valid signature, brand-new binaries can hit a "rarely downloaded" reputation gate. If you see "Windows protected your PC," click "More info" → "Run anyway." That's per-binary reputation, separate from cert trust.
curl -fsSL https://raw.githubusercontent.com/jackccrawford/geniuz/main/install.sh | bash
Detects your OS and architecture, downloads the matching CLI binary (with TUI built in), installs to ~/.geniuz/bin/. Best for developers, fleet operators, and anyone using Claude Code, Cursor, Windsurf, Aider, or any agent framework that can run a shell command.
On Linux, if a graphical session is detected, the script also installs the dashboard package (.deb / .rpm / .AppImage, picked from your package manager) with one sudo prompt. Headless servers skip the dashboard step automatically. Opt out explicitly with GENIUZ_NO_DASHBOARD=1. On macOS, the script installs only the CLI. Use the DMG above for the dashboard + menubar.
Supported architectures:
The arm64 build bundles ONNX Runtime 1.22 as a sibling .so and wraps the CLI with an LD_LIBRARY_PATH script, so it runs cleanly on older-glibc systems like Pi OS Bookworm (glibc 2.36). The x86_64 build is a single static binary.
Claude Desktop isn't available on Linux, but geniuz mcp serve works as a stdio MCP server for any Linux-compatible MCP client (Claude Code, Cursor, Windsurf, Aider, custom agents). The dashboard ships as a .deb and .AppImage for desktop Linux; the TUI runs in any terminal.
git clone https://github.com/jackccrawford/geniuz && cd geniuz
cargo build --release --bin geniuz
cp target/release/geniuz ~/.local/bin/
To also build the dashboard locally:
cd desktop/dashboard
cargo tauri build
(Requires tauri-cli. Mac: nothing else. Linux: libwebkit2gtk-4.1-dev build-essential libxdo-dev libssl-dev libayatana-appindicator3-dev librsvg2-dev. Windows: WebView2 runtime, typically preinstalled on Win 10/11.)
Geniuz is a compiled Rust binary backed by SQLite. No cloud. No API key. No account. Your data stays on your machine.
Dashboard ─┐
TUI ─┼─→ db::DatabaseManager ──→ memory.db (SQLite)
Menubar ─┤ ↑
CLI ─┘ │
↓
ONNX (BERT)
The model downloads once (~118MB) on first search. Every memory after that is embedded automatically. No setup. No configuration.
Save what you learned (CLI):
$ geniuz remember -c "Maria prefers retention over acquisition in Q2. Budget is $40K." -g "client: Maria — Q2 retention focus, $40K budget"
✅ Remembered 7A3B29F1
Find it later, by meaning:
$ geniuz recall "Maria's budget priorities"
7A3B29F1 | 2026-03-05 14:23 | client: Maria — Q2 retention focus, $40K budget (0.52)
Browse interactively:
$ geniuz tui
Opens a terminal UI with /recent, /remember, /find, /reorder, /random, /detail, /help. Press / to start typing a command; ↑/↓ to navigate; Enter to open detail; Ctrl-C to exit.
Thread a follow-up:
$ geniuz remember -c "Maria approved the retention plan. Starting in April." -g "client: Maria — plan approved" -p 7A3B29F1
✅ Remembered E5F6A7B8
The full client history, from first meeting to approval, is one chain. Any future session finds the whole story.
Already have notes, docs, or agent memory files?
geniuz capture ./docs/ # all markdown files
geniuz capture --split notes.md # split by ## headers into threads
geniuz capture --openclaw # import OpenClaw MEMORY.md + daily logs
geniuz capture --dry-run ./notes/ # preview without importing
geniuz backfill # embed everything for semantic search
Three commands (capture, backfill, recall) turn any folder of markdown into a searchable memory folder. Local RAG with zero infrastructure.
# The three R's — remember, recall, recent
geniuz remember -c "what happened" -g "category: compressed insight"
geniuz remember -c @notes.md -g "session: review"
echo "content" | geniuz remember -c - -g "piped: from process"
geniuz remember -c "follow-up" -g "update" -p 98672A90
geniuz recall "topic" # semantic search
geniuz recall "topic" --since 7d # only the past week
geniuz recall "reviews by Cubic" # Capitalized names come first
geniuz recall --grep Cubic --grep PR # matching lines, every term
geniuz recall --thread 98672A90 # a whole thread, root first
geniuz recall --keyword "exact words" # keyword fallback
geniuz recall --random # discover something
geniuz recall --full "topic" # include full content
geniuz recall --json "topic" # JSON output
geniuz recent # latest memories
geniuz recent -l 5 # last 5
geniuz recent --full # with content
# Interactive
geniuz tui # terminal UI (requires a TTY)
# Capture and index
geniuz capture ./docs/ # bulk-load files
geniuz backfill # build embedding cache
# Folder
geniuz status # folder stats
geniuz watch # poll for new memories
geniuz watch --exec "echo {uuid} {gist}" # trigger on new memories
# Claude Desktop
geniuz mcp install # add Geniuz to Claude Desktop
geniuz mcp status # check if configured
geniuz mcp serve # run MCP server (used internally)
Add two lines to your agent's instructions:
When you learn something worth keeping:
geniuz remember -c "what you learned" -g "category: compressed insight"
When you need to remember something:
geniuz recall "what you're looking for"
Geniuz Free exposes four user surfaces plus the MCP server for agents:
geniuz tui) — same for terminal users and agents that prefer ratatui over GUI.--json for procedural callers.geniuz mcp serve is the stdio MCP server. Used by Claude Desktop, Claude Code, and any MCP client.All surfaces go through one library: db::DatabaseManager. Same write path, same read path, same invariants. The dashboard's "Remember" button and the CLI's geniuz remember end up at the same SQL INSERT.
The SQLite file (memory.db) is not a public interface. Schema may change without notice. Invariants (memory immutability enforced by triggers, every-memory-has-an-embedding enforced by the library's write transaction) hold at the interface boundary, not at the file boundary. If you want programmatic access, go through MCP or the CLI.
There is no HTTP server in Geniuz Free. Procedural software speaks MCP. This keeps the storage layer free to evolve and the invariants centralized in one place.
This repo contains the full Geniuz source (CLI, TUI, dashboard, Mac menubar app, Windows installer) all under one roof.
Built artifacts are attached to each GitHub release:
Geniuz.dmg — Mac (arm64, Sonoma 14+) — menubar + dashboard + CLIGeniuz-Setup.exe — Windows installer (Inno Setup, x86_64, Windows 11) — CLI, MCP server and the dashboard with system tray.geniuz-linux-amd64.tar.gz — Linux x86_64 CLI + TUIgeniuz-linux-arm64.tar.gz — Linux arm64 CLI + TUI (Pi 5 compatible)Geniuz_<version>_amd64.deb / Geniuz-<version>-1.x86_64.rpm / Geniuz_<version>_amd64.AppImage — Linux x86_64 dashboardThis section is for you, the one reading this README.
Your next session starts cold. No memory of what you built, what broke, what you decided. Every memory you write is a gift to that future session.
A memory that says "fixed auth bug" saves zero time. A memory that says what broke, why, and how you fixed it saves an hour. Write for the version of you that knows nothing.
Memories compound. A single memory is a note. A folder of memories is institutional knowledge.
Search by meaning to discover; search exactly to recall. Every answer ends
with names seen: and pointers: (the next hops) and a searched: line
that says what was searched, so an empty answer carries its scope. When a
question has a time in it, add --since 7d. When you know the name, use
--grep Name. When you've found one piece of a conversation, read the rest
with --thread ID.
geniuz tui only launches with an interactive terminal. Agents calling from a subprocess get a clean refusal with exit code 2. If you want browsing in code, use geniuz recent, geniuz recall, or the MCP tools. They're shell-friendly.
geniuz-embed is an optional background process that holds the ONNX model in memory. Every memory after the first is instant.
geniuz-embed & # start (auto-exits after 5 min idle)
geniuz remember -c "first" # 1.6s (model loads)
geniuz remember -c "second" # 0.04s (model warm)
Your data stays on your computer. Geniuz stores memories in a local SQLite database. The semantic search model runs locally via ONNX Runtime. No data is sent anywhere. The source code is open. Read every line.
MIT
Your AI remembers now. Geniuz stores all your data locally on Mac, Windows, Linux, Raspberry Pi. No cloud. No account. No API keys. It's open source; you can read every line of code.
Tools
Noch keine Tool-Liste. Remote-Server erhalten eine nach dem ersten Health-Check.
FAQ
Kopieren Sie auf dieser Seite die stdio-Konfiguration für Cursor, Claude Desktop, Claude Code oder VS Code. Sie nutzt die Installation des Herausgebers, keinen PlugMCP-Proxy.
In diesem Eintrag ist keine erforderliche Umgebungsvariable aufgeführt.
Geniuz läuft lokal über stdio.
Zuletzt geprüft: 2026-11-15. Your AI remembers now. Geniuz stores all your data locally on Mac, Windows, Linux, Raspberry Pi. No cloud. No account. No API keys. It's open source; you can read every line of code.