Computer-Use Trajectories
24,000 browser and desktop trajectories with per-step screenshots, accessibility observations, grounded actions and final-state checks.
Discuss this productWhat the agent could see.
The image and accessibility tree belong to the same point in the application’s state.
- screenshot
- Rendered application state before the action
- a11y_tree
- Element roles, names and states
- step
- Position in the recorded sequence
Observations are aligned with the action that follows.
The delivered unit
Every observation, action and result.
The capture connects what the agent could see to what it did and what changed.
- 01
Per-step observations
The screenshot and accessibility tree available before each action.
- 02
Grounded actions
The recorded tool action, target role and coordinate grounding for the interaction.
- 03
Observed transitions
The change in application state after the action, kept in the original sequence.
- 04
Outcome checks
A final-state check against the environment, evaluated after the task is complete.
Built around the task
Train the interface between seeing and doing.
A click is only useful when it is grounded in the right observation. Screenshots, accessibility trees, actions and state changes stay aligned through the trajectory.
Select the applications, task families, observation format and action interface for your model. Pair trajectories with environment delivery where the program requires it.
What should your agent operate?
Share the task, tool surface and scale. We’ll help you select and evaluate the material.