markov

Data + environments for training computer-use AI.

We capture how people use computers—and turn it into data models can learn from.

Data samplesTalk to us
150k+ downloads on HuggingFace

Our datasets

Rich computer-use data built around complete workflows, synchronized interaction signals, and verifiable outcomes.

Engineering drawing and assembly model of a four-cylinder engine

CAD environments

AutoCADSolidWorksSiemens NXSketchUpRevitCATIA
  • Screen recordings + synced mouse/keyboard events + narrations
  • Problem statements + input references
  • Gold output files + rubrics
Samples
A recorded coding and browser-use workflow with VS Code and an SVG asset library

CUA SFT data

Browser useCodingDesignSpreadsheetsProductive workflows
  • Screen recordings + synced mouse/keyboard events + narrations

Our open-source work

markov-ai on Hugging Face
A real software-development workflow recorded in VS Code

10,000+ hours

Computer-use recordings

Real workflows across Salesforce, Blender, Photoshop, and more.

View dataset

Train on how people actually use computers.

Talk to usExplore the data