Human-layer data · 32 countries

Egocentric video data at real scale.

Thousands of hours of first-person video, on demand.

We collect first-person video of real work tasks, captured by professionals using head-mounted cameras across Latin America, South Africa and the Middle East. Built to train the most demanding AI models.

See dataset coverage

Consent-first capture Every contributor paid Provenance per clip Owned, never scraped

Built for teams training

VLA models World models Humanoids Mobile manipulation Industrial automation Video generation

5,000

Hours per 4-week cycle

Work domains

Task types catalogued

Countries in network

72h

From brief to first clips

Standing capacity across the network.

The physical AI stack

We do one layer.
Completely.

Robot foundation models are trained on three layers of signal. Most vendors claim all three and are thin on each. We produce the human layer — the demonstration signal — and we produce it end to end, from recruiting the person to the signed release.

Not us Layer 03

World layer

Simulation, scene reconstruction and context that world models train against.

This is us Layer 01

Human layer

First-person capture of real work. The source signal for how people actually move, reach, grip and recover from mistakes.

Not us Layer 02

Robot layer

Teleoperation logs and teacher-follower demonstrations that bridge human intent into robot execution.

Why us

Built for volume,
not for demos.

Scalable high-volume delivery

Engineered to capture, process and deliver thousands of hours of egocentric footage per month. The pipeline flexes for volume spikes and hard deadlines instead of breaking under them.

Global ground network

Trained, equipped collectors across emerging and key international markets, giving you demographic, environmental and domain coverage that a single-country vendor cannot reach.

Tailored collection frameworks

Custom recording protocols: specialised head-mounted gear, multi-modal sensor logging (IMU, audio, depth), script-driven task execution, and organic unscripted daily workflows.

Sample clips

This is what
you receive.

Unedited clips straight from the network — head-mounted, shot on the contributor's own phone, in their real workplace. No staging, no reshoots, no cleanup. Tap one to preview it, or watch all three at full size below.

Every shift.
Every trade.

Tap a clip to play it in colour.

Clip 01

Clip 02

Clip 03

Clip 04

Clip 05

Clip 06

Want the full sample pack, with metadata and the consent record attached? — it is free.

The clips, uncut Unedited, head-mounted, shot on the contributor's own phone. Nothing trimmed, colour-graded or stabilised.

Clip 04

Clip 05

Clip 06

Clip 07

Clip 08

Clip 09

Coverage

What we cover,
and where.

Fifteen work domains, eighty-four catalogued tasks and thirty-two countries already running. Everything below is live in the network and can be commissioned today.

Domains and tasks

Hands-on work in commercial settings — the cramped, cluttered, badly-lit places that break models trained on clean lab footage. All fifteen are live in the network and can be commissioned today.

Tap any domain to see its tasks. Need a domain that isn't here? We recruit for it. Two to three weeks to stand one up.

Programme mix

Our standard spread across the network — balanced so a model sees variety rather than one kitchen a thousand times, and weighted toward the regions most datasets under-sample. Every slider moves: tell us the mix you want and we build the run to it.

Share of a standard 5,000-hour programme. Configurable to your spec at no extra cost.

Capture network

Thirty-two countries with city leaders in place and contributors onboarded in the capture app. Any of them can be switched on for a commissioned run, and new ones take about three weeks to stand up.

Adding a country takes about three weeks: recruit a city leader, onboard contributors, secure business permissions.

How it works

From brief to
training-ready.

Scope

Tasks, scenes, minimum clip length, camera placement, and what must never be captured. We turn the spec into a task list the contributor sees on their phone before they hit record.

Capture

City leaders recruit inside the trade and secure written permission from each business. People record their own shift through our app: pick the task, hit start, it uploads by itself. Nothing is assembled after the fact.

Review & redact

Every session is rated by a human before it counts. Unusable footage is rejected and never paid for. Faces, documents, screens and anything you excluded are handled to your rules.

Deliver

Clips with task and scene labels, country, device, duration and timestamps, plus the consent and payment record behind each one. Your bucket, your structure, your manifest.

What ships

Data you can put in
front of legal.

Egocentric footage is filmed inside someone's workplace, often around their coworkers. Done badly, it becomes your problem long after delivery. So we built the paperwork before we built the library.

Everyone signs, everyone is paid

Each contributor signs an agreement before recording and is paid per approved hour at a rate they know in advance. They can check what they have earned in their own dashboard, any time. Nobody works for exposure.

The business agrees in writing

Nothing is filmed on a business's premises without written permission, and the owner can stop it at any moment without giving a reason.

Hard limits on what is filmed

No minors, no documents, no screens, no internal pricing, no identifiable customers. Those rules live in the signed agreement, not just in the training deck.

Traceable clip by clip

For any clip we can produce who recorded it, where, when, on which device, under which agreement, and what they were paid. Full chain of custody on request.

Every clip we sell can be traced to a person who signed an agreement, a business that gave written permission, and a payment that went out. That is the part your legal team will ask about — and the part most of this industry cannot answer.

How to work with us

Commission
the capture.

Most teams start with a free sample pack to check the fit against their pipeline, then commission a run built to their spec.

Commissioned capture

Built to your spec

5,000 hrs

production capacity every 4 weeks

We recruit inside the trade, not from a general crowd panel.
First clips back for review within days of the brief.
Scales by adding cities, not by squeezing the same contributors.
Free sample pack, so quality gets checked before money moves.

FAQ

What buyers
ask first.

How fast can you stand up a new domain?

Two to three weeks from brief to first clips in a domain we do not already run. In a domain we already cover, first clips land in days.

Can we see samples before committing?

Yes, free. Tell us the domains and we send a sample pack from the library carrying the same metadata and consent record a full delivery does.

Who owns the footage, and can we train on it?

We own it. The contributor agreement assigns it to us, explicitly including use for AI training, and they are paid for that. You receive a clear licence for training and evaluation, with the provenance documentation behind it.

What hardware is it shot on?

The contributor's own phone, head-mounted. We log the exact model per clip so you can filter or stratify by sensor. If your model needs specific hardware, we equip the run for it.

Where do you operate?

Thirty-two countries, with the deepest bench across Latin America and a second cluster in Asia-Pacific and the Gulf. That matters because the big coverage charts in this industry are dominated by two or three countries. Different regions mean different tools, layouts, product mixes and working practices — precisely the distribution shift that breaks a model in deployment.

Do you provide annotation?

On request: step segmentation, hand-object interaction tags, or your own taxonomy. Every clip already ships labelled by domain and task from a fixed catalogue.

How is it priced?

Per approved hour. It depends on the domain, how hard it is to recruit for, and whether you need annotation. Tell us the volume and we come back with a number.

How do you handle privacy in someone's workplace?

Exclusions are written into the signed agreement and enforced at review: no minors, no documents, no screens, no identifiable customers. Face and object redaction is applied to your rules before delivery, and anything that breaks the rules is rejected and never paid for.

Egocentric video data at real scale.