Request sample access
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This listing is a preview. The production dataset (250,000 traces, ~15,000 hours of screen activity) is rights-cleared and delivered directly under a commercial license. Approved requesters get the full annotation schema and data dictionary in this repository, and can request a review package with real trace samples, metadata, and QA summaries. Requests are reviewed within 1 business day.
Log in or Sign Up to review the conditions and access this dataset content.
Computer-Use Traces Dataset
250,000 real-world computer-use traces with screen states, browser sessions, UI actions, task instructions, and completion outcomes for training AI agents.
This repository is a specification and preview listing. The production dataset is rights-cleared and delivered directly to buyers. Request access to see the full schema and get a real sample package.
Overview
The Computer-Use Traces Dataset is a large-scale collection of 250,000 real-world computer interaction traces for training and evaluating AI agents that operate browsers, websites, desktop interfaces, SaaS tools, and productivity software.
Each trace captures the full process of completing a digital task, from the user's starting instruction to the final outcome: task prompt, screen recordings, screenshot states, mouse movements, clicks, scrolls, keyboard inputs, typed text, timestamps, UI element interactions, intermediate steps, and completion status. Depending on the subset, traces also include browser metadata, page URLs, DOM snapshots, accessibility tree data, visible text extraction, form fields, button labels, menu interactions, tab changes, error states, recovery steps, and human-reviewed success or failure labels.
This makes the data directly useful for state-action-outcome modeling, UI action prediction, visual grounding, browser automation, desktop automation, and agent evaluation: models learn when to click a button, fill a form field, open a dropdown, scroll to hidden information, switch tabs, copy information between tools, and stop when the task is complete.
At a glance
| Traces | 250,000 |
| Screen activity | ~15,000 hours |
| UI actions | ~10M timestamped mouse, keyboard, scroll, click, and text-entry events |
| Screen states | ~8M screenshots / visual UI states |
| Task categories | 100+ |
| Avg. trace length | 3 to 5 minutes |
| Avg. actions per trace | 30 to 50 |
| Outcome labels | Human-reviewed success / failure with recovery steps |
Task coverage
100+ task categories across real software workflows, including:
web search, form filling, ecommerce navigation, spreadsheet editing, calendar scheduling, email workflows, CRM updates, dashboard navigation, invoice review, file organization, account settings, software onboarding, travel booking, data entry, product comparison, customer support tooling, internal admin workflows, and multi-step research tasks.
This diversity teaches models how users move across interfaces, locate relevant information, enter data, revise actions, recover from mistakes, and complete tasks in realistic digital environments.
Technical specifications
- Data modalities: screen recordings, screenshots, task instructions, action logs, UI metadata, DOM snapshots where available, accessibility trees where available
- Annotation coverage: task category, user intent, step labels, UI targets, action timestamps, success/failure outcome, recovery steps, reviewer notes
- QA: human-reviewed completion labels, PII redaction review, malformed trace filtering, duplicate trace filtering
- Delivery format: CSV metadata, JSON action logs, linked MP4 screen recordings, screenshot folders, optional DOM/accessibility files
Annotation schema
The gated file annotation_schema.json in this repository contains the full trace schema with an illustrative example record. A typical trace record includes:
| Field group | Contents |
|---|---|
| Task | instruction text, task category, intent label |
| Actions | ordered event log with type, target element, coordinates, typed text, timestamp |
| States | screenshot references, URL, DOM/accessibility snapshot references |
| Outcome | success/failure label, error states, recovery steps, reviewer notes |
How to evaluate this dataset
- Request access using the form above. Requests are reviewed within 1 business day.
- On approval you get the gated files in this repository: full annotation schema, data dictionary, and access instructions.
- Request a review package and we deliver real trace samples, metadata CSVs, JSON action logs, QA summaries, and licensing documentation within 2 business days.
All samples are delivered with structured CSV metadata and JSON annotation files where available. Buyer review packages include representative media files, metadata samples, annotation schema, QA summaries, and data dictionary documentation.
Licensing
The production dataset is rights-cleared for commercial AI training and licensed directly by Datoric, with chain-of-custody and consent documentation. Subset, exclusive, and custom-collection options are available.
About Datoric
Datoric supplies rights-cleared, spec-exact training data for frontier AI labs and enterprise model teams: computer-use traces, expressive multilingual voice, egocentric and industrial video, human manipulation data, robot episodes, and gameplay trajectories. We also run managed collection pipelines for custom specifications.
Contact: nikhil@arzule.com
- Downloads last month
- 10