A write loop, not a read-only API
Agents lease a phone, look at a fresh frame, act, assert and export evidence. Actions are tied to the frame they were chosen from, so a stale screen or a retried request never taps twice.
Handsy for teams
Real iPhones for your agents. No Mac. No $99 a year.
Plug a small Handsy module into any iPhone and your agent sees the screen and taps it like a person. No Apple Developer account, no Mac, no WebDriverAgent to re-sign. You pay only for results you accept.
- $99/yr
- saved per Apple Developer account
- 0 Macs
- needed to drive the phone
- $0
- when a run fails on our side
Open-source hardware, CLI and MCP server: available now. Hosted Handsy Cloud and racks: pilots with design partners.
Leasing the phone…
Run 7f21 · iPhone 13 Pro · iOS 26PASS
- Tapped “Pair” on a fresh screen
- “Connected” is visible
- Paired exactly once
Accepted · billed
Our failure · not billed
- No jailbreak
- Stock iPhone, stock iOS
- Any agent
- Claude Code, Codex, any MCP client
- From ~$7 per phone
- Small module, open-source hardware
- Your phones or ours
- Your desk, your lab, or Handsy Cloud
Cost and stability
No Apple Developer account. No Mac. Nothing to re-sign.
Scripted iOS automation runs through WebDriverAgent, a test app you build, sign and keep alive on every phone. Handsy talks to iOS as a plain USB keyboard and mouse instead, so there is less to pay for and less to break.
WebDriverAgent stackHandsy
Apple account$99 a year for the Developer Program; a free Apple ID’s signing expires every 7 daysNone for apps already on the phone
Host computerA Mac with Xcode in every labAny computer that runs the CLI
Setup on each phoneDeveloper Mode on; WebDriverAgent installed and trustedPlug in the module, turn on AssistiveTouch
When Xcode or iOS updatesRebuild and re-sign WebDriverAgent, re-check every deviceNothing to rebuild: keyboard and mouse are standard iOS input
What usually breaksExpired profiles, crashed runner sessions, selector driftA loose cable or a dropped link, reported as an infrastructure failure and not billed
Building and shipping your own app still uses Apple’s normal signing. The $99 is Apple’s published Developer Program fee.
The gap
Simulators can’t pair. Scripts need a Mac. Agents can only read.
- 01
Simulators skip the radio
No simulator pairs a real Bluetooth accessory, drops Wi-Fi mid-flow, or behaves like the phone your customers carry.
- 02
Scripted stacks are heavy
Appium and XCUITest need a Mac, Xcode and a signed WebDriverAgent on every device, and break when the UI shifts.
- 03
Agents see reports, not phones
Coding agents can read a test report. They still can’t look at a real screen, tap it, and check what happened.
Why Handsy
Built for the agent that has to finish the job.
Real radios, real phones
Bluetooth pairing, Wi-Fi handoffs, background and foreground switches happen on real hardware. The things simulators can’t see are the point.
No Mac, Xcode or WDA in the input path
For apps already installed, the Handsy module is the keyboard and mouse. Nothing to sign or rebuild per device just to send a tap.
Pay for accepted results
A result is defined before the run. A PASS, or a real bug in your app, is a delivery. Our hardware or network failing is not.
What you can run
One module per phone. The same recipes on your desk, in your lab, or in our cloud.
Agents
Available nowPhones for AI agents
Screenshot, tap, swipe, type and hardware buttons on a real iPhone from Claude Code, Codex or any MCP client. The agent looks again after every action.
iPhone MCP server →Bluetooth & IoT
PilotBluetooth and connected devices
Pair, disconnect and reconnect your real accessory. Each step returns a screenshot, so a failed pairing is visible, not guessed. App bugs and an accessory that wasn’t ready are reported apart.
Network
PilotWi-Fi and network transitions
Switch networks mid-flow and check the app recovers: sync resumes, nothing is submitted twice, the user sees the right state.
Releases
PilotReal-device assertion packs
Fixed checks per build, device and iOS version: PASS or FAIL, screenshots, every action, and a recipe to run it again.
Self-hosted
PilotYour own device lab
Phones and accounts stay on your premises. Single slot today; 4- and 8-slot racks with per-slot power and fault isolation are in design.
Workflows
PilotAuthorized mobile workflows
Internal processes that live in phone apps, across several apps, with a person approving anything that buys, sends or deletes.
How a run works
Look, act, check. Every step leaves evidence.
The CLI runs this loop on your phone today. Handsy Cloud wraps it in device leases, assertions agreed up front and an exportable evidence pack.
- 1LeaseOne run holds the phone. Nothing else types into it.
- 2LookA fresh frame of the real screen. Stale frames are refused.
- 3ActTap, swipe, type or press, bound to that frame with an idempotency key.
- 4CheckAssertions set before the run decide PASS or FAIL.
- 5Hand backEvidence exported; the phone is reset, or kept signed in for you.
What you get back
An evidence pack you can audit
- Each frame and action receipt, in order
- Device, iOS, app build and environment versions
- Assertions with PASS or FAIL and the frame that proves it
- Failure class: app bug, infrastructure, permission, or needs a person
- A recipe to run the same thing again
Run 7f21 · iPhone 13 Pro · iOS 26PASS
- Tapped “Pair” on a fresh screen
- “Connected” is visible
- Paired exactly once
Accepted · billed
Our failure · not billed
Result → payment
Pay for accepted results, not for robot time.
Scope, checks and the run window are fixed before we start. Delivering a result and you accepting it are separate steps, and only accepted results are billed.
- 01Scoped
- 02Run
- 03Delivered
- 04Verified
- 05Accepted
- 06Billed
Our hardware, network or runner failed, or evidence is missingNot billed
Your own phones
The hardware design, firmware, CLI and MCP server are open source. Free.
A real bug counts
A FAIL that finds a real defect in your app is a delivered result, with the evidence to fix it.
No double charge
Retries, re-exports and webhooks never add results. Disputed lines are held, not billed.
Pilot pricing is quoted per scope. No annual contract to start.
Where it runs
Same CLI, recipes, permissions and evidence everywhere.
Available now
On your desk
One module in one iPhone, driven over Bluetooth or Wi-Fi from your laptop.
Pilot
In your lab
A gateway and modular slot boards, each slot with its own power, monitoring and fault isolation.
Pilot
In Handsy Cloud
Lease a phone from the web, CLI, API or MCP, run, collect evidence, release. Dedicated phones keep your sign-ins.
Control
Clicking is easy. Being allowed to click is the product.
- One writer per phone
- A lease gives one run exclusive input. Shared phones are cleaned and checked before the next run.
- Approval gates
- Buying, publishing, sending and deleting wait for a person, even if the button is right there.
- Accounts by reference
- Test accounts you provide. We never deliver passwords, cookies or verification codes.
- Audit and stop
- Every action is logged with its frame. A run can be stopped mid-step; keys are released first.
Handsy vs Mobot
Mobot runs your tests for you. Handsy lets your agent run the phone.
Both use real devices. They differ in who operates the phone and who owns the loop.
| Topic | Handsy | Mobot |
|---|---|---|
| Operating the phone | A small module acting as USB keyboard and mouse, with the screen over AirPlay | Robots physically tap the glass |
| Who runs the tests | Your agent, CI or team, self-serve | Mobot’s AI and QA analysts build and run them for you |
| What an agent can do | Drive the phone through MCP or CLI: screenshot, tap, swipe, type, buttons | Read-only MCP: reports, test cases, observations, logs. Booking and running are planned |
| Checking failures | Evidence for every step; human review on request in pilots | A QA analyst validates every failure |
| Devices | Your own iPhones today; hosted slots in pilot | 300+ hosted iOS and Android devices |
| Self-hosting | Open source today; hardware from about $7 per phone | Mobot Labs, early access from 2027 |
| Pricing | Free on your phones; pilots per accepted result | Credits, or Mobot Unlimited at a flat annual rate, by invitation |
Choose Handsy when your own agent or pipeline should operate real iPhones, without a Mac or a $99 developer account in the loop, or when phones must stay on your premises. For Face ID, camera and sensor flows, Android, or fully managed QA, Mobot fits better.
FAQ
Before you ask
Do I need a jailbreak, a Mac or WebDriverAgent?
No jailbreak. For apps already installed on the phone, the input path needs no Mac, Xcode or WebDriverAgent. Building and distributing your own app still follows Apple’s usual rules.
Is Handsy Cloud available?
Not publicly. Hosted phones and multi-slot racks run as pilots with a few design partners. The open-source hardware and CLI work on your own phone today.
Can it handle Face ID, the lock screen or the camera?
No. A keyboard and mouse can’t present a face or a finger, and can’t replace a real touch on the glass. Tell us your flow and we’ll say plainly whether it fits.
Android?
iPhone only for now.
What happens when a run fails because of you?
Infrastructure failures, missing evidence or unmet required fields are not billed, and any reserved amount is released. They are reported separately from failures in your app.
Bring me a flow.
Tell us what you need to run on real phones.
One device, your app, checks agreed up front. You see the results before you pay.