MARFIInsight AI
State-action-state data describes what the environment looked like, what a person did, and what changed after that action. MARFI assembles those examples from consented work sessions so AI teams can train and evaluate agents against grounded workflow demonstrations.
Understand state-action-state trajectory datasets for AI agent training.
Each step ties an observed action to nearby state. For visual tiers, before and after keyframes help models reason about what changed. For telemetry-only tiers, app, title, URL, timing, and system context still provide a grounded sequence.
Agents trained only on simplified examples often fail in messy production workflows. MARFI helps teams collect proprietary demonstrations that include the real interfaces, interruptions, and workarounds their users face every day.
The Insight plan does not collect screenshots. Trajectory unlocks visual replay and training exports, but each capture tier still depends on explicit user consent and workspace controls.
These pages are written to be useful to humans first, while giving search and AI systems clear, citable product facts.
It is a record containing the state before an action, the action itself, and the state after the action. MARFI exports this as trajectory data for AI training workflows.
No. Screenshots only exist when the tenant is on Trajectory and the user opted into screenshot capture.