AGI One
AGI One
General-purpose agent control system
A voice-native agent system for macOS. Give it a goal in your own language; it understands the active workspace, coordinates the right models, acts across apps and tools, and verifies the result — on your own API keys.
A visible agent, not a static assistant
Enable the optional digital-human mode and the assistant appears as a realtime, audio-synchronized presence. Its face and voice follow the same response that drives the agent system; reasoning, tools, memory, and control stay on the primary path.
This is the actual in-app android. Digital-human rendering is optional and only starts after you configure its providers.
Audio synchronized
- Wake
- Hear
- Decide
- Act
- Verify
Currently for Apple silicon Macs · macOS 13+ · one-time purchase · bring your own API keys
One spoken sentence, five stages
Every turn runs the same loop. Nothing is faked between stages — if a step can't be verified, it says so instead of claiming success.
- Wake — an offline Vosk model on your Mac listens for the wake word. No audio leaves the machine until you're talking to it.
- Hear — realtime speech, full duplex. Interrupt it mid-sentence and it stops, like a person would.
- Decide — it reads intent from your words, not keywords. Clear instruction: it just does it. Ambiguous or irreversible: it asks first.
- Act — commands first, GUI only when there's no command path. Real cursor and keyboard events, 1:1 logical coordinates.
- Verify — it looks again and checks the result before telling you it's done.
What it does
Talks in real time
Full-duplex speech. Interrupt it mid-sentence and it stops, like a person would. Voice stays live while work is running, so you can ask what's happening or change your mind without waiting.
Acts across the system
The agent selects the fastest available execution path, coordinates tools, handles applications and dialogs, and falls back to the visual interface when needed. Every action stays inside one supervised, verifiable control loop.
Uses your tools
Anything you can do on this Mac, it can do the same way — including the coding tools you already have installed. No integrations to configure: it operates them like you would, and reports back out loud.
Finishes multi-step work
Give it a goal, not a click. It plans, acts, checks the result, and changes approach when a step fails — instead of stopping at the first obstacle or claiming it's done.
Appears as a live digital human
Optional audio-synchronized rendering gives the agent a visible presence with matched speech and motion, without replacing the realtime voice or execution brain.
Uses your own keys
Your OpenAI and Anthropic keys stay on your Mac and requests go straight to the providers. You pay them directly; there is no subscription and no server of ours in between.
Measured, not demoed
Demo videos are easy to cut. So we built a 29-task benchmark — 20 command and file tasks, 9 long-chain GUI tasks — with programmatic checkers. The score is what the checker found on disk and on screen, never what the model said it did.
| Model | Basic 20 | GUI chain 9 | Total | Wall clock | False success |
|---|---|---|---|---|---|
| Claude Sonnet 5 | 20/20 | 9/9 | 29/29 | 9 min | 0 |
| Qwen3.7-Plus | 20/20 | 5/9 | 25/29 | 22 min | 2 |
| Kimi K3 | 18/20 | 4/9 | 22/29 | 95 min | 0 |
Ships on Claude Sonnet 5. The others are swappable with one config line — the loop doesn't care which brain it's running.
What leaves your machine
Stated plainly, because a thing that watches your screen owes you this.
- Stays local — wake-word detection. The offline model runs on your Mac; microphone audio is not sent anywhere until you've woken it and are speaking to it.
- Goes out while in use — your speech, and screenshots of your screen, go to the model providers you configured (OpenAI for voice, Anthropic for agent execution) to be understood. That is how it sees.
- Optional digital human — only when you enable it, the android image and assistant output audio go to LemonSlice; synchronized audio/video returns through your configured LiveKit project. Your microphone, screen, and task text are not sent through this path.
- App activity is not collected by us — there is no Hugging Monkey application-data server. Your keys stay on your Mac and model requests go straight from your machine to the providers.
The camera is different: it only ever opens when you ask for it out loud, enforced in code against the actual transcript — not left to the model's judgment.
The website's download service keeps limited, privacy-preserving delivery metrics so we can count unique completed downloads and diagnose failed transfers. It never receives your voice, screen, files, API keys, or activity inside the app. Details are in the privacy policy.
Get it
One payment. No monthly AGI One subscription.
- AGI One for Apple silicon Macs
- Real-time voice interaction
- General-purpose agent control system
- Multi-step task execution
- Screen understanding
- Result verification
- Bring your own AI API keys
- Updates for the current major version
Currently for Apple silicon Macs only. Requires macOS 13 Ventura or later. AI usage is not included: you use your own supported AI provider API keys and pay those providers directly.
Founding 1,000: the first 1,000 completed downloads are free — direct download, no checkout, no card. When the founding batch is gone, the link comes down and the price is $29.90. Buying now simply supports the project.
Direct download: 257 MB dmg, Developer ID–signed and notarized by Apple. Auto-updates from 1.0.3 on.
SHA-256 cef3a67366edcc1819af1d0dc5ceb5e6d8b11b5d94209be10d58d55faebd913c
Purchase → checkout by Polar → your download link for the macOS build.
Developer ID–signed macOS build · Apple silicon only · macOS 13 Ventura or later · needs Screen Recording and Accessibility permission to drive your apps.