Visual Read
Read the active screen, application state and visible interface before acting.
RUNTIME CAPABILITIES
From agent intent
to real-world execution.
A model-agnostic bridge to computers, browsers and software. Structured interfaces first. Governed computer interaction when the task demands it.
Semantic UI Access
Use structured interface information when available instead of relying only on raw pixels.
Keyboard & Pointer Control
Perform authorized typing, clicking, scrolling, window focus and other human-computer actions.
Browser & Desktop Workflows
Move across websites and local applications inside one governed execution environment.
Multi-Path Routing
Prefer API/MCP when possible and fall back to GUI execution when structured interfaces cannot finish the job.
Evidence & Verification
Return observable results, logs and checkpoints rather than treating an input action as proof of success.
Durable Recovery
Preserve task state and resume safely after interruption, reconnecting the calling agent when execution can continue.
Human Override
Keep consequential permissions, approvals and stop controls available to the human principal.
THE EXECUTION LOOP
AgentINTENT
RouterPOLICY & AUTHORITY
API / MCPWHEN AVAILABLE
UARWHEN GUI IS REQUIRED
Evidence back to AgentVERIFY & CONTINUE
Direct paths remain preferred when sufficient. UAR bridges the steps they cannot complete.