macOS desktop agent · pre-release beta

Tell your Mac what to do. Watch it do it.

ScreenHand reads your screen, plans the steps, and operates your real apps — Mail, Notes, Calendar, Safari — to finish multi-step tasks you describe in plain English. You supervise; it works.

macOS only Bring your own AI key Supervised — you stay in control Paid beta · invite-based
Request beta access See how it works No public installer or billing yet — early access is invite-based while we harden reliability.
How it works

It uses your apps the way you would — just faster, and it doesn't forget a step.

No per-app integrations, no scripting, no brittle pixel coordinates. You describe the outcome; ScreenHand figures out the clicks and the typing.

You describe it

Type a goal in plain English — "find the latest email from Priya and save it as a note."

It sees the screen

It reads the accessibility tree and screenshots to understand what's actually in front of it.

It plans & acts

It plans the steps, then clicks, types, and navigates real apps — recovering when something shifts.

It verifies

It checks the result actually happened, instead of assuming success.

What works today

A handful of real workflows we've proven — not "automate anything."

We'd rather ship a few things that work reliably than promise everything. These run on the apps macOS exposes well. New ones are added only after they clear a hard reliability bar.

Mail → Notes

Capture an email as a note

Find the latest message and save its subject and sender into Apple Notes.

Calendar

Daily briefing note

Read today's events and assemble a clean briefing in Notes.

Mail

Inbox triage

Sort unread mail into a morning standup note — urgent, follow-up, informational.

Research → Notes

Research and save

Search the web on a topic and drop a sourced summary into Notes.

Supervised by design

An agent that controls your machine should be one you can stop.

The constraints are the product. ScreenHand is built to be watched, interrupted, and trusted — not to run loose.

Approve every script

Before it runs any generated script, it shows you the exact code and waits for your Approve or Deny.

Instant kill switch

One hotkey stops a run mid-flight — even when another app has focus.

Won't run away with your bill

A per-run cost cap stops the agent before it burns through your AI budget.

Asks before destructive actions

Deleting, sending, or overwriting prompts you first — nothing irreversible happens silently.

Your keys, your machine

You bring your own Claude or Gemini key; it runs on your Mac, not our servers.

It verifies its own work

Every run is checked against the real result, so "done" means done — not "the agent thinks so."

The honest part

What you should know before you ask for access.

We'd rather you trust us than be surprised. Here's the unvarnished state of things.

It's a supervised beta
Pre-release software running an autonomous agent on your Mac. You watch it. It's reliable on the proven workflows and best-effort beyond them — it will sometimes fail. Not for unattended, mission-critical use.
Your screen goes to a cloud AI
To understand the screen, ScreenHand sends screenshots and on-screen text to the AI provider you choose (e.g. Anthropic Claude or Google Gemini). We disclose this and ask for consent — we don't hide it. Read the privacy details.
Mac only, for now
macOS on Apple Silicon. Windows and tool-specific (engineering) support are on the roadmap, not in this beta.
Bring your own AI key
You supply and pay for your own Claude/Gemini API key. There's no hosted inference and no free tier in the beta.
It needs real permissions
Accessibility, Screen Recording, and app automation — the access a power-user automation tool needs. You grant them; you can revoke them anytime.
FAQ

Straight answers.

Is this fully autonomous — can I set it and forget it?

No. It's supervised by design. You run it and watch it, and it pauses for your approval before running scripts or doing anything destructive. Treat it as a fast assistant you oversee, not a hands-off robot.

Can it automate any app?

No — and we won't claim it. It's reliable on a small set of proven Mac workflows today (Mail, Notes, Calendar, Safari) and best-effort elsewhere. We add new workflows only after they clear a strict reliability bar.

Is my data private?

Your screen contents are sent to the cloud AI provider you choose so it can read the screen — that's how it works, and we disclose it up front. It runs on your machine with your key; we don't run inference or store your screen on our servers. See the privacy details.

How much does it cost?

The beta is invite-based and paid; pricing is being finalized. You also pay your own AI provider for the API usage (bring your own key).

Windows? CANoe / MATLAB / engineering tools?

Roadmap, not beta. The current product is Mac-only on consumer apps. Tell us your use case when you request access — it helps us prioritize.

Who builds this?

Clazro Technology Private Limited. It's early, it's honest, and it's improving fast.

Want in while it's early?

Access is invite-based while we harden reliability on real machines. Tell us what you'd automate and we'll be in touch.