Think. Create.
Stay local.
Zeno is a cross-platform AI assistant that runs entirely on your machine — powered by Ollama, with voice in, voice out, and a plugin system you control. Nothing leaves your device unless you say so.
Minimum RAM to run Zeno
Analytics or telemetry events sent
Zeno runs entirely on your machine. Connect your own Ollama models — nothing leaves your device unless you say so.
Sign upEverything the assistant needs, none of the cloud it doesn't
Zeno bundles model access, voice, security, and extensibility into one desktop app — each piece pulled straight from the codebase, not a marketing layer on top of it.
Ollama integration
Full model management, streaming responses, and context window control against your local Ollama instance.
Voice in, voice out
Web Speech API by default, with optional offline speech-to-text (Whisper or VOSK) and multi-voice text-to-speech.
Security first
Local-only by default, encrypted chat history, sandboxed plugin execution, audit logging, and a strict Content Security Policy.
Polished UI
Light and dark themes, glass-morphism backgrounds, full ARIA support for screen readers, and keyboard shortcuts throughout.
Plugin system
Define your own commands as Python plugins with explicit, per-plugin permission controls.
Low-end PC support
Auto-detects device capability, drops to 1B-parameter models, and trims animations on machines with as little as 2GB RAM.
From ollama pull to a working assistant in four steps
Zeno's own quick-start, laid out as a path — install a model, launch the stack, talk to it, then extend it with plugins you control.
Install Ollama & pull a model
Zeno talks to a local Ollama instance. Install Ollama, then pull a model such as llama2 to get started.
One-click startup
Run Zeno_AI.bat, start_zeno.ps1, or npm run start:all to launch Ollama, the Python backend, and the React frontend together.
Talk or type
Use push-to-talk voice input or the chat window. Requests are streamed back from your local model over a secured WebSocket.
Extend with plugins
Drop a Python plugin into ~/.jarvis/plugins/ and register it — every action still requires your explicit confirmation.
Runtime stack
Every layer runs on your machine — the WebSocket link between the Electron shell and the FastAPI backend is local and secured, and the only network calls Ollama makes are to itself.
A threat model, not a promise
Local-first isn't a slogan on Zeno's landing page — it's the storage layer, the sandbox, and the confirmation prompt in front of every action.
Local-first by design
All data stays on your machine by default. There is no telemetry and no analytics pipeline collecting your conversations.
Encrypted where it matters
Chat history is stored in SQLite and can be encrypted at rest with a password-derived key using PBKDF2 and AES-256-GCM.
Sandboxed by default
User scripts and plugins run in a restricted environment, and every shell command, file access, or network request needs your confirmation.
Built for the hardware you have
From a 2GB dual-core laptop to a quad-core desktop, Zeno auto-detects your device and scales model size and effects accordingly.
The app is free. Support is optional.
Zeno's desktop app is 100% free & open source under the MIT License. These optional plans are for teams who want managed infrastructure and support layered on top.
Starter
For individuals who want a maintained, hosted config.
- Everything in the free desktop app
- Hosted config & plugin backups
- Priority security patches
- Email support
Pro
For power users running larger local models daily.
- Everything in Starter
- Offline Whisper STT bundle included
- Advanced audit log retention
- Custom plugin permission policies
- Priority chat support
Pay As You Use
Flexible metering for teams with variable workloads.
- No monthly minimum
- Meter by active device or session
- Scale support up or down anytime
- Same local-first security model
Questions people ask before they clone the repo
Run your own assistant. Keep your own data.
Clone the repo, pull a model, and Zeno is talking to you in minutes — no account, no cloud, no telemetry.