What is Lowkey?

Lowkey is a workspace you run by talking to it. It runs on a private server dedicated to you, which we host and look after, works with any AI model, and keeps work moving while you're away.

Instead of opening apps, clicking through menus and doing the work yourself, you open a chat and ask. On the other side are AI agents that can do things: read and write your files, run tasks, search the web, send a message, build something, and keep working in the background while you do something else. It looks like a messaging app, because the way you get work done is by having a conversation.

You don't need to know any commands. You type what you want in plain language and an agent works out how to do it. If it needs something from you, like a file, a yes or a sign-in, it asks right there in the chat.

What that feels like

You ask once. The agent does the work, shows you the result, and remembers the context so the next message picks up where you left off.

What it does

Work that outlasts a chat. When something won't finish today, because you're waiting on someone or there's a deadline next week, it becomes a task: a chat with its state attached. Every task says where things stand, whose move it is, and when it will be looked at next. When that date arrives, Lowkey's clock wakes the task in its own chat and the agent picks it up. Nothing depends on anyone remembering to check. Related tasks can live in a program, with a regular review that spots anything drifting. Anything that repeats runs on a schedule, and a run that finds nothing to do stays quiet and costs nothing.

It asks when it matters, and only then. When an agent needs you to choose, you get a decision in your Inbox: a short brief with the background, the options and a recommendation. When something is truly blocked on you and can't wait, it sends a page to your phone. An ordinary answer never buzzes you. Home shows what your agents are doing across projects, with a Work log that keeps a receipt for each action.

Any model, or several at once. Agents think with Claude, GPT, Gemini, or Grok. Switch models per conversation from the gear next to the message box; the agent stays the same, only how it thinks changes. For a question that matters, pick two or more models. Each answers on its own, and you get one reply saying where they agree, what only one caught, and whether they split. If they split, press Debate and they answer each other.

Your own box, and your secrets stay locked. Every Lowkey customer gets their own private instance, not a shared service. Your projects, files, conversations and credentials live on a machine that's yours alone. Credentials go into a vault that agents never see: the box uses a credential on an agent's behalf, a credential belongs to one person, and there is no button that shows a secret, not even for an admin. For the sensitive ones, you approve each use from a card.

Built for your phone. Install Lowkey on your phone as an app and turn on notifications, so pages arrive as ordinary push notifications and a tap opens the chat. Conversations live on your box, so a thread you started on your laptop is waiting on your phone. You can also call an agent hands-free from a chat and hear its progress spoken.

And the rest

• A real browser and your own computers. Agents drive a browser on your box, and can run commands, move files and use the browser on a laptop or PC you pair. Control of your screen is off until you turn it on for a conversation.
• Files you can see. Uploads, documents, spreadsheets, PDFs and previews open in a focus pane beside the chat, each in its own tab. Mark up a preview and your notes go to the agent with your next message.
• Share a page. Publish a report or a mockup to a public link that expires after a week by default, with an optional password.
• Google Drive and a project wiki. Keep a Drive folder in sync with a project, and let agents file what they find into a sourced wiki they consult later.
• Other people. Share a project with your family or team and @mention a person or an agent. Each person keeps their own credentials and approvals.
• You stay in charge of the conversation. Send another message while work runs, stop a turn, or rewind to before a message and ask again.

Why it isn't a chatbot or an IDE

Lowkey agents carry out the work inside a real workspace with real files, and the results stick around: documents you can open, tasks that keep running, a history you can come back to.

Lowkey gives you what a developer's setup gives a developer, agents that read and write files and run commands, without asking you to be one. No editor to learn, no command line, nothing to install. It works from your phone or your browser.

Agents really can act on your behalf, and Lowkey is plain about what that means: an agent works with real access to your workspace. That's why there is one box per customer, why credentials sit in a vault, and why you can stop any task at any time.

Who it's for

• People who want things done, not another tool to manage.
• Anyone who's hit the wall of "the AI can tell me how, but can't just do it."
• Small teams and households that want a shared place where the work gets done.
• Tinkerers who want agents with real hands but don't want to live in a terminal.

Get started

The whole thing is a conversation. Open a project, start a chat, and type what you want, the same way you'd text a person.

Keep reading: Why Lowkey exists · How Lowkey is different · Docs