in&space home Join the waitlist

Developer preview · Linux (X11) first

Give agents hands. Keep control.

Local computer control for AI agents

in&space is a small driver that runs on your computer. It lets an AI agent see and use the apps you choose, and it stops to ask before anything that publishes, deletes or pays.

Join the waitlist See how it works

What it does

An agent that can use your apps, on your terms.

Agents are good at working out what to do next. in&space gives them a careful way to do it on a real desktop.

  • It sees what you see

    The driver reads windows through the system’s accessibility interfaces and takes screenshots of the apps you allowed, so the agent knows what is on screen.

  • It acts as a person would

    Clicks, typing, scrolling and keyboard shortcuts, sent to the app in front of it. Every action is checked against the apps you chose.

  • It stops to ask

    Sending, publishing, deleting and paying wait for your click on an approval card. Nothing else counts as a yes.

How it works

Four parts. One of them is you.

Your agent lives in a host app: a chat app, an editor, a script of your own. The host app talks to the in&space driver on your computer, and the driver works the apps you chose.

  1. You

    Choose the apps, watch the indicator, approve or stop.

  2. The agent

    Runs in your host app and decides the next step.

  3. The in&space driver

    Runs on your computer. Reads the screen, acts, and keeps to your limits.

  4. Your apps

    Only the ones you allowed, only for this session.

Stays on your computer

  • The driver and everything it reads from the screen
  • Your consent choices and the list of allowed apps
  • A local record of what the agent did, with secrets left out
  • Everything in&space does: no telemetry, no account, no update checks

May leave, through your host app

The host app decides what to send to its AI model. If it sends screenshots or window text to the model’s provider, that is covered by the host app’s terms. in&space itself sends nothing anywhere.

Safety and approvals

You stay in charge at every step.

The rules live in the driver, not in the agent’s good manners.

An approval card on a desktop, waiting for a person to click before a message is sent.
You choose the apps.
The agent works only in the apps you allow for a session, and only while that session lasts.
Big actions wait for your click.
Publishing, deleting and paying stop at an approval card. Typed text, voice, model output and the agent’s own clicks never approve.
You can always see it working.
A visible indicator stays on screen while the agent has control. Move the mouse or type, and control comes straight back to you.
One click stops everything.
The kill switch ends every session at once, from the indicator or from a terminal.
Password fields stay private.
The driver refuses to read or type into password fields.
No telemetry.
The driver, the CLI and the SDKs have no network code. Nothing is reported back to us.

For developers

Bring your own agent.

in&space speaks the Model Context Protocol and comes with a command-line tool and two SDKs. Each one offers the same tools, the same approvals and the same stop.

Developer preview. Linux (X11) first; macOS and Windows are in progress.

  • MCP server

    Connect any MCP client. Tools are marked read-only or destructive, and destructive ones go through the approval card.

  • The inspace CLI

    Check your setup, start the MCP server, print the settings for your MCP client, and stop everything from a terminal.

  • TypeScript SDK

    Typed tools from the driver contract, session events, approvals and a stop call. No runtime dependencies.

  • Python SDK

    The same surface for Python, using only the standard library.

A code editor beside a quiet panel that lists an agent’s tool calls, one of them waiting for approval.

Nothing is published yet, so there is nothing to install today. Install steps arrive with the preview. See what the docs will cover

Pricing

Free while it is a preview.

The developer preview costs nothing. Paid plans for teams will come later.

Consent, approvals, the indicator and the kill switch are never a paid extra.

Questions

Plain answers.

What is in&space?

A small program, the driver, that runs on your computer and lets an AI agent see and use the apps you allow. You stay in control: big actions wait for your click, and you can stop it at any time.

Which systems does it run on?

The developer preview runs on Linux with the X11 display server. Wayland is not supported yet, and the driver tells you so instead of half-working. macOS and Windows are in progress.

Does in&space send my data anywhere?

No. The driver, the CLI and the SDKs have no telemetry, no accounts and no update checks. The host app you connect may send what the agent sees, such as screenshots, to its AI model. That is up to that app and its provider.

Can the agent post, delete or pay on its own?

No. Those actions stop at an approval card on your screen and go ahead only after your click. Typed text, voice, model output and the agent’s own clicks never count as approval.

How do I stop it?

Move the mouse or type and control comes back to you at once. To end everything, use the stop button on the indicator, or the stop command in the CLI.

Can it read my passwords?

No. The driver refuses to read or type into password fields, and its local record leaves secrets out.

Which agents work with it?

Any MCP client, and any app built with the TypeScript or Python SDK. The agent and its model are your choice.

What does it cost?

The developer preview is free. Paid plans for teams will come later. Safety features are never a paid extra.

Under which licence will it ship?

The licence is not final yet. It will be published before the public preview.

How do I get it?

Join the waitlist below. We will write to you when the developer preview is ready.