Hand it your knowledge, your methods and your routines. It works right on your Mac, with a cloud model or one that runs on the machine, and takes work wherever you already chat.
Coach it once, hand it work for good. Everything you teach lives in its workspace: a real folder on your Mac that you can open and edit.
Documents, handbooks, templates, who you are. It remembers across conversations.
Built-in skills: law lookup, official letters, offline OCR. Add your own.
Describe a multi-step job in plain words. It becomes a workflow with stops for your sign-off.
Morning briefs, weekly reports: it runs on time, even while you're away.
Each correction becomes a lesson. It applies one only after you approve it, and measures whether it helps.
One late-night WhatsApp message, and here is what happens. At the approval step, try pressing Approve.
DuberWorker has been at work every day since July 2026: answering messages, running schedules, writing and drafting. These are the measured numbers.
of daily use: several agents, messaging channels and scheduled runs
work turns, including – tool calls
tokens processed
of input tokens served from cache, far cheaper and faster
of turns ran on a fast model; only hard work stepped up to a stronger one
of turns finished without an error
Measured on the team's own Mac, 17 Jul 2026 to –.
In the cloud or on your Mac, switch whenever you like. Your colleague keeps its knowledge, skills and memory.
Set a default, a fallback for when limits run out, or a model per task.
Connect Duber to the messaging apps you use, or scan a QR code once to work from your phone over the Internet.
Telegram, WhatsApp, Zalo, Slack, Discord, Microsoft Teams, Google Chat, iMessage, Signal, LINE, Lark, Matrix, Mattermost.
There's no server of ours in between. Only what goes to the model reaches the AI provider you picked; with a local model, nothing leaves at all.
Chat shows every step: which file it read, what it ran. It asks before anything important.
The Context Inspector shows exactly what each turn sends to the model, with token counts.
API keys are encrypted with the macOS Keychain. Remote access is off by default; each device gets its own key, revocable at once.
Everything it needs is bundled: no Node, Python or Xcode to install.
The setup guide checks your Mac and helps you sign in to an AI account or choose a local model.
Give it a name and a role, tell it about you, connect a messaging app, and hand it the first task.
The macOS app is being finished, and runs for real every day before it reaches you.
The app is free during the beta. Cloud models are paid directly to their provider; local models cost nothing.
Yes, with a local model (Ollama, LM Studio). Messaging channels and cloud models need a connection.
It works inside its own workspace and asks before anything important. You can review every step it took.
For now, Apple Silicon Macs only.