v0.4.4 — Beyond parity (142 HTTP endpoints)
5 features that competitors don't have:
Action Caching
/cache/set?key=login_btn&ref=e3&action=click → repeated workflows skip LLM inference. Subsequent runs cost ~$0.
Set-of-Marks Screenshot
/screenshot/som → screenshot with numbered red bounding boxes on interactive elements. Vision models (GPT-4V, Claude) reference elements by number.
Hybrid Snapshot
/snapshot?include_screenshot=true → a11y tree + base64 screenshot in one call. Saves a round-trip for multimodal agents.
Smart Diff
/snapshot/changes → only what changed since last snapshot. 10-100x fewer tokens than full re-snapshot.
Recording Export
/recording/export → recorded actions as replayable /batch JSON. Record once, replay deterministically.
macOS binaries signed with Developer ID Application: Rachit Pradhan (WWP9DLJ27P).