Hardware-level AI desktop automation
Give AI
a pair of real hands.
ClawTouch reads your screen with vision models and a browser Sensor extension, then sends keystrokes and mouse events through a real USB HID device — handling repetitive desktop work for you. Trigger workflows manually or run them on a schedule. Your own account. Runs locally on your machine.
01 — How it works
Not just AI software.
Real hands on real keys.
Every desktop action runs through a Perception → Decision → Action loop. Whether you trigger it yourself or a workflow schedules it, the agent reads the screen, decides what to do next, and physically presses the keys — exactly what a person at the keyboard would do.
Perception
Read the screen
A browser Sensor extension and a vision model turn whatever is on screen into something the agent can reason about.
- Sensor extension reads the DOM and visible text — read-only
- Vision model recognizes desktop application UI
- OCR and UI Automation as additional channels
- Handles multi-display setups and DPI scaling
"See what's on screen."
Decision
Plan what's next
A built-in LLM plans the next move from what's on screen and the task you've given it.
- Built-in LLM, no external API key needed
- Planning that respects what's already on screen
- Chains multi-step tasks into workflows
- Save and reuse task templates
"Think what's next."
Action
Press the keys
A real USB HID device sends keystrokes and mouse events through the same OS input path as any external keyboard you'd plug in.
- Physical hardware sends every keystroke and click
- Works across applications — no SDK integration needed
- Doesn't depend on window focus or app permissions
- Keeps working when target UIs get redesigned
"Real hands. Real actions."
02 — Enterprise
AI-assisted operations,
packaged for enterprises.
The standard annual package bundles the hardware, the desktop software, and managed AI usage. When you need more, additional capabilities are available à la carte and priced separately. Every engagement is governed by a Master Service Agreement, a Data Processing Agreement, and a Service Level Agreement. Contact us for pricing and rollout details.
Standard · All-in-one
Included
- HID hardware deviceA custom USB HID device that delivers every action as a real keyboard or mouse event
- Desktop clientRuns locally on your own Windows machine — no compute leaves the box
- WeChat mini-programMobile companion app — monitor sessions, message the agent, take control from your phone
- Built-in AI modelNo external API key needed; a generous monthly usage allowance is included
- Built-in UI vision modelTrained specifically to recognize desktop application interfaces
- Persona libraryBuild and store multiple agent personas for different tasks
- Persona engineKeeps each persona's voice consistent across sessions
- Cross-browser compatibilityWorks in Chrome, Edge, Firefox, and other major browsers
- Sensor browser extensionRead-only DOM access; browser-agnostic
- Custom task templatesParameterized prompts you can reuse and share
- Custom workflowsChain multiple task steps into a repeatable flow
- Bundled CLI toolsImage, document, media, and fuzzy-search utilities the agent can call directly
- 24-hour managed modeRuns continuously, driven by incoming events rather than manual triggers
- Recurring task schedulerRe-run task templates on a schedule or fixed interval
- Operational dashboardsFull session logs, usage statistics, and output reports
- Dedicated supportResponse within 30 minutes on business days, through the WeChat mini-program
- Tax invoices (PRC)Individual and enterprise fapiao supported
Custom · À la carte
Optional
- On-premise private deploymentInstalled inside your network perimeter; data never leaves it
- Custom workflow developmentBespoke scripts written around your operational pipeline
- Platform-specific adaptersWe build adapters for your internal business platforms
- Custom managed-mode featuresCapabilities built on top of the managed execution engine
- Enterprise model trainingFine-tune on your business data; the resulting model weights are yours
- Bespoke feature developmentAnything not on the standard list, built on request
- Compliance audit supportExport full operation logs for your internal review
- Account security hardeningLogin anomaly alerts, activity monitoring, MFA enforcement
- Team trainingRemote or on-site, tailored to your processes
- 24/7 dedicated operationsRound-the-clock monitoring, incident response, and upgrade assistance
- Fleet managementA dedicated admin console for 50+ device deployments
- API / Webhook integrationConnect to your CRM, ticketing, or other internal systems
Open source · MIT
The hardware layer
is open.
The hardware, wire protocol, and skill templates are MIT-licensed and public: clawtouch-mcp, clawtouch-hid, and clawtouch-skills. Verify the device and protocol yourself; the integrated commercial product stays closed-source.