Human-verified web
trajectories, at scale.
Slesh collects and verifies real browser interaction data across thousands of domains, so your agent learns from how humans actually complete tasks.
Real tasks. Fully captured. Verified.
Slesh generates structured web tasks and collects human demonstrations across real websites at scale. Each trajectory is end-to-end and verified.

Data ready for training
Every trajectory is a complete record of a successful completion of a real web task and the full action sequence from start to finish.
Data is collected through our crowdsourced platform, then verified for correctness before delivery.
{ "task": "Find a 2-bedroom Airbnb in Tokyo under $200/night", "domain": "airbnb.com", "result": "success", "actions": [ { "action_type": "input", "description": "Entered “Tokyo” into the destination search field", "element": { "tag": "input", "selector": "input[name='query']" }, "screenshot": "step_001.png", "dom": "step_001.html" }, { "action_type": "click", "description": "Applied price filter under $200 per night", "element": { "tag": "button", "text": "Price" }, "coordinates": { "x": 1032, "y": 312 }, "network_calls":[{ "request": {"method": "GET","url": "/api/v3/Search","params": {"price_max": 200}}, "response": {"status": 200, "body": {"results": 42,"fields": ["listing_id", "price", "rating"]}} }], "screenshot": "step_002.png", "dom": "step_002.html" } ] }
Internal API capture — Actions include the full network exchange tied to the exact UI interaction that triggered it
Built for teams building agents and automation
Whether you're fine-tuning an existing model or building training infrastructure from scratch.
Fine-tune on the exact workflows where your agent is failing. Target specific domains and task types and use real trajectories to debug behavior and iterate faster.
Access structured data from live websites through indexed internal API calls mapped from browser workflows.
Build robustness into pipelines with human-demonstrated recovery paths across thousands of real sites.
Start with a pilot dataset
We'll put together sample verified trajectories matched to your domain and task requirements.
Tell us about specific workflows, domains, or task types, and we'll collect targeted trajectories for those exact cases.