Multimodal / AI & Machine Learning
Computer-Use Traces Dataset | 250,000 Global Screen, Browser & UI Action Traces for AI Agents - Machine Learning Data
Advertised by providerLicensing rights not independently verifiedPrice on request
Overview
Computer-Use Traces Dataset | 250,000 Global Screen, Browser & UI Action Traces for AI Agents - Machine Learning Data is a multimodal dataset. Typical applications include fine-tuning & instruction data, conversational AI and agents & tool use. Relevant sectors: AI & Machine Learning and Education. The listing mentions timestamps. Pricing is not published and is provided on request.
Primary applications
What the listing offers
- Listing mentions timestamps.
The Computer-Use Traces Dataset is a large-scale collection of 250,000 real-world computer interaction traces designed for training and evaluating AI agents that operate browsers, websites, desktop interfaces, SaaS tools, and productivity software. The dataset contains approximately 15,000 hours of screen activity, 10 million timestamped UI actions, and 8 million visual screen states across more than 100 task categories. Each trace captures the full process of completing a digital task from the original user instruction to the final outcome. The data includes screen recordings, screenshots, mouse movements, clicks, scrolls, keyboard inputs, typed text, timestamps, UI element interactions, step labels, task outcome labels, and completion status. Browser and web-app traces include page URLs, DOM snapshots, accessibility tree data, visible text extraction, form field metadata, button labels, tab changes, error states, and recovery actions. The dataset covers 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. Average trace length is 3 to 5 minutes, with 30 to 50 actions per trace. The dataset is built for computer-use models that need to understand the relationship between an instruction,
Related datasets
More in Multimodal & Agent Training →MultimodalAdvertised
AI Agent Tool-Use Traces Dataset | Multi-Step Trajectories for LLM & SFT Training
This dataset provides a curated, normalized collection of multi-step AI agent execution traces extracted from heterogeneous real-world archives.
- Language
- —
- Volume
- —
- Domain
- AI & Machine Learning +1
- Use case
- Fine-tuning & instruction data
Price on request
More multimodal & agent training datasets
All 159 multimodal & agent training datasets →- Computer Vision Data: Multi-View Drone Imagery & 3D Scenes | Multi-Country
- Conversation Audio | Candidate Speech from Real AI-Led Job Interviews | Mono 48 kHz WAV
- Conversational Text-to-Speech (TTS) Dataset | 20,000 Hours of 48kHz Voice Data for Text-to-Speech, Voice Cloning & Speech AI
- Cross-Disciplinary Thesis Dataset - 641K Thesis, 7.3B Words, 14 Categories | Textual Data
- Custom AI Training Data | Audio, Video, Embodied AI, 3D & Text — Any Modality, Volume | Made-to-Order, Scoped Per Project | Machine Learning (ML) Data
- Custom Manipulation Data Collection | Egocentric Wrist-Camera Video + IMU Recorded to Your Task Spec | LeRobot v3.0
- Customer Service Audio Dataset [Raw Call Recordings, Multi-Industry, U.S.]