Opportunities › AI agent control for mobile devices › Proposed app
Tapwright
Let vision AI agents see and operate a real iPhone through a small plug-in board
- Score
- 48
- confidence 49
- Evidence
- 3
- items on the page
- Rivals
- 0
- named in sources
- First version
- 30 d
- estimate
- Type
- Desktop app
- PWA: runs in the browser, installs on Mac and Windows
- Market since
- no product yet
- New market · AI estimate
ScreensConcept
Desktop app: AI engineers monitor agent runs through dashboards, logs, and guardrails—intensive work that demands wide screens and precise configuration at a workstation. Sample screens on a computer, in the proposed brand. Their numbers are sample data.
New run
Board online- DeviceiPhone bench 2
About this appAI proposal
Tapwright is a control console and hosted API for running AI agents on real iPhones. A small ESP32 board acts as a keyboard and pointer over USB, while the phone's screen is mirrored over AirPlay so a vision model can see it. No Mac, no developer account, no jailbreak.
It is built for AI engineers and automation teams who want agents grounded in a real device, not a simulator. Point it at any OpenAI-compatible vision model, write a goal in plain words, and watch each step, screenshot and action in a replayable run log.
What is different: the phone stays unmodified, runs are recorded and repeatable, and several boards are managed from one place instead of a shelf of cables and scripts.
- Tagline
- Give your AI agent a real finger and a real screen.
- Voice
- PracticalDirectCalm
- Why this name
- A wright is a maker. Tapwright joins tap, the basic phone gesture, with a builder of repeatable flows.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
Bring any vision model
Connect any OpenAI-compatible endpoint with your own key. Swap models per task and compare how each handles the same flow.
No member yet
Board pairing in minutes
Flash the ESP32 from the browser, pair it to the phone and confirm the screen feed. No desktop toolchain needed.
No member yet
Step-by-step run log
Every run stores the screenshot, the model's reasoning and the tap or key sent. Replay any run to find where it went wrong.
No member yet
Guardrails for actions
Set allowed apps, a step limit and confirm-before-action rules, so an agent cannot wander into payments or settings.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#14212BDeep Ink · primary, text#F6F3ECPaper · surface#E8590CSignal Orange · accent#1E8C8CTeal Glass · data#D9D4C7Fog · border
Typography
Aa
Tapwright
ABCDEFGHIJKLMNOPQRSTUVWXYZ 0123456789
Give your AI agent a real finger and a real screen.
PricingAI proposal
The material gives no competitor prices. These tiers are a starting point to test, based on the suggested subscription and concurrent-device premium model.
Hobby
Free
- 1 board
- 50 agent steps a day
- Bring your own model key
- 7-day run log
Builder
$29 a month
- 3 boards
- Unlimited steps
- Saved flows and replay
- 30-day run log
- HTTP API
Team
$149 a month
- 10 boards
- Queue across devices
- Guardrail policies
- Shared workspace
- 90-day run log
What it could be worthEstimate
What a software business making this much sells for today, from real deals. Pick a monthly revenue.
If it makes
Worth about
$190,000
Sale range $132,000 to $263,000
$120,000 a year in revenue
From 87 software businesses listed and sold on Empire Flippers in the last year, revenue vetted by the broker: they sell for about 19× their monthly revenue, most between 13× and 26×.
Measured on profit, the middle deal is 32× monthly net profit: about 2.7 years of profit.
An estimate, not an offer or financial advice. A real price depends on profit, growth, churn, how much of the work is the owner's, and the buyer.
How it would growAI proposal
How the product would reach new customers through its own use, and how the first ones would be found. A plan to test, not a result.
- Built-in loop
- Shared outputs. Users share replayable run links with teammates, clients or bug reports, showing the screen, steps and result, with a Tapwright footer on each link.
- Do they become customers?
- Viewers are often engineers who could build their own board. Test whether shared run links convert viewers to sign-ups, as opposed to just being viewed once.
- First customers
- Small AI agent and mobile QA teams of 2 to 15 engineers who already test on physical iPhones and lack a Mac fleet or cloud device lab budget.
- Where to reach them
- GitHub project community: Reach people starring and discussing the open-source repo with a hosted console that saves them setup time.
- AI agent builder forums: Post demo run logs of real iPhone tasks in agent framework Discords and communities.
- Direct outreach to QA leads: Offer a free two-week pilot to mobile QA leads on LinkedIn and in testing groups.
- First milestone
- Ten teams run flows weekly for a month, and five of them pay for Builder or Team.
Why people already pay for this
- VerifiedGitHub repository gained 2879 stars in 5 days
- Public claimThe solution provides a low-cost way to control real iPhones with AI models using hardware costing $7, eliminating the need for Mac, developer account, or jailbreaking.
- Public claimThe approach works with any OpenAI-compatible vision model, indicating compatibility with the growing ecosystem of LLMs.
Tech stack
Every proposal builds on the same PWA base, installable from the browser on any phone or computer. Dashed boxes are what this app adds.
- AppNext.js and TypeScript
- PWAManifest and service worker
- UITailwind CSS
- Data and authSupabase Postgres
- PaymentsStripe
- EmailResend
- HostingVercel
- This app · Device firmwareESP32 firmware for USB HID, open sourceThe board must act as a USB keyboard and pointer and be flashed from the browser.
- This app · Screen captureAirPlay receiver service on a small workerReceives the phone's mirrored screen and turns it into frames for the vision model.
RisksAI inference
- TechnicalReliable AirPlay mirroring and HID pointer control on current iOS versions may vary by version and be slow or fragile. Latency may make some agent tasks unreliable.
- PlatformApple could change AirPlay or accessory behaviour in an update and break the approach without notice.
- CommercialThe 2,879 stars in 5 days show interest, but not whether teams will pay. The open-source project may be enough for many.
- RegulatoryAutomating apps may breach some apps' terms of service, and agents acting on real accounts raise liability and privacy questions.
- AIVision models can misread screens or tap wrongly, so wrong actions on a real phone are possible. Guardrails reduce this but do not remove it.
Everything marked AI proposal or AI inference was written by a model from the evidence on the opportunity page. The proposed name must be checked against trademark registers before it is used.