What the engine does to a video.
Detect objects, brands, logos, and products
Found in frame with a timestamp.
Transcribe speech — timestamped text
Read on-screen text — words become data
Track actions and motion over time
Describe what is happening
Interpret against criteria you define
Search across all of it
beverage cup 00:14
3 matching moments found
How you use it.
Three ways to reach the capabilities.
Pandora
Our agent. Ask in plain language, it calls the right capability.
Which clips mention the tumbler? → 3 moments found
Zelf MCP
Connect your own agent. Zelf becomes a tool it can call.
agent → zelf → vision output
Zelf API
Call any capability directly in code.
curl https://api.hellozelf.com/api/v1/creator/vetting/ -H ‘Authorization: Api-Key YOUR_KEY’ -d ‘{profile_url: https://tiktok.com/@creator}’
What people build with it.
Brand and product detection in video
Find a brand or product inside video, with the timestamp where it appears.
cup · 00:14 · 00:27
Video datasets and annotation at scale
Turn footage into labelled training data, including dense action annotation over time.
grasp · move · place
Sentiment and share of voice
Read how an audience talks about a subject, including social listening.
positive · neutral · negative
Creator and content screening
Vet a creator against your rules before you sign them.
criteria checked · review ready
Brand discovery and monitoring
Surface rising brands early and track them.
01 rising · 02 watch · 03 new
Vision AI, without the ML team.