SCREENWEB AGENTS
Browser Navigation and Web Task Flows
Browser-based task sequences for models that need to reason across navigation, page state and multi-step interaction.
View datasetSCREENCOMPUTER USE
Software Task Screenflows
Computer-use models need more than screenshots. They need to understand how an interface changes after an action and how a sequence of decisions moves a task towards completion. Software Task Screenflows preserve that progression across a continuous desktop workflow.
Screen state can be connected to task steps, action events and visible interface text where the task requires it. This turns a recording into a sequence of model-relevant states and actions rather than leaving the buyer to reconstruct the workflow from video alone.
Technical structure
- Continuous screen recordings linked to task and step identifiers
- Action events and application context where available
- OCR-derived interface text when visible text is relevant to the task
- Temporal segments connecting interface state, action and resulting state
Typical applications
Computer-use agents, GUI grounding, workflow automation, task completion and screen understanding.
Delivery note
The useful action vocabulary and level of temporal detail are defined around the software tasks being modelled.
Delivery
The exact package depends on the collection scope. Where relevant, delivery can include task-specific records, manifests, provenance fields, stable identifiers, SHA-256 hashes, MLCommons Croissant 1.0 metadata and loading instructions.