Script-first browser automation for AI agents
Launch, snapshot, and execute deterministic browser actions across managed, extension, and CDP modes from one script-first platform.
Launch, snapshot, and execute deterministic browser actions across managed, extension, and CDP modes from one script-first platform.
CLI commands
Built-in tools
Coverage gate
Runtime modes
Three value rails summarize deterministic execution, secure operations, and modular workflow outcomes.
Snapshot to refs to actions replaces brittle selectors with stable references for repeatable automation behavior.
Token auth, origin checks, and redaction defaults create strong execution boundaries without slowing down delivery.
Research, shopping, and product video surfaces convert low-level actions into high-level outcome pipelines.
The validated seven-step workflow from launch to close.
Start managed mode, extension relay mode, or CDP connect mode with one lifecycle model.
Capture deterministic page state before actions.
Run click, type, select, press, and scroll via stable references.
Poll console, network, trace, and perf for runtime confidence.
Use packaged research, shopping, and product-video modules with per-run cookie policy controls.
Capture screenshots, annotations, and cloned component/page artifacts.
Check status and disconnect sessions with explicit completion markers.
Ten required lanes spanning QA, extraction, operations, and workflow-driven outcomes.
Run repeatable browser checks with trace-backed pass/fail evidence.
Operate safely in logged-in sessions using extension relay controls.
Extract structured page intelligence from DOM and workflow outputs.
Capture screenshots and annotations for fast UI review cycles.
Clone pages/components into reusable frontend artifacts quickly.
Diagnose regressions early with status, console, network, and perf signals.
Generate time-bounded multi-source research outputs in one workflow run.
Compare offers across providers with normalized pricing and confidence signals.
Find high-confidence opportunities with market-aware ranking and filters.
Collect product visuals/copy artifacts for UGC videos and product slides.
Concise trust posture aligned to enforceable runtime and operational controls.
Unsafe transport and raw CDP options are disabled by default.
Token checks and origin policy gate extension relay channels.
Sensitive values are redacted from logs and diagnostics outputs.
MIT licensed now, roadmap-driven next, and structured for contributions.
Core runtime, 48 tools, 55 commands, extension relay, workflow wrappers, and skill packs are already available.
Open source, auditable, and structured for AI agent workflows from day one.