Native Android tools
Tap, type, scroll, swipe, navigate Settings panels, wait for UI to settle — every interaction is a typed tool the agent can call. No screen scraping, no overlay hacks.
Features
Six pillars, each enforced in code, each readable on this page.
Tap, type, scroll, swipe, navigate Settings panels, wait for UI to settle — every interaction is a typed tool the agent can call. No screen scraping, no overlay hacks.
The agent reads on-screen text and accessibility metadata through Android's AccessibilityService. It sees what you see, including app names, fields, and buttons.
Each capability is a Markdown file with a YAML header: id, title, description, risk, allowed tools, success criteria, and example phrases. Read, edit, version-control.
A deterministic local router handles simple actions with no network or API key. Optional LiteRT-backed on-device model for richer command routing.
Every medium- and high-risk tool surfaces a confirmation card in chat before it runs. Low-risk observation runs silent. The user is always the last word.
Every model decision and every tool execution is persisted locally and exportable. Sensitive arguments are redacted before they hit disk.