Common questions

Questions, answered.

For AI labs

What do you deliver?

Rights-cleared, frame-aligned sessions: gameplay video, render buffers, world state, inputs, actions, events, and queryable metadata. The full stack by default.

What formats?

Standard mp4 for video and buffers. State ships in VTX, our open Apache-2.0 format; one CLI command converts it to Parquet or JSON. Rerun playback built in.

How do I get access?

Browse the catalogue in the lab portal, agree an order across the titles you want, and the data is granted to your workspace. Sample packs and same-day eval access on request.

Is the behaviour human?

Yes. Real players, captured to a brief and QA'd. No agent rollouts.

How is it normalised?

One coordinate system, scale, and convention across every title. A multi-title grant behaves like one dataset.

Can you capture what you don't have?

Yes. Specify titles, scenarios, signals, and rates: we capture to order.

For IP owners

Do I keep my IP?

Yes. A non-exclusive licence to capture and commercialise training data. You retain the IP, the rights, and your options.

What do I have to do?

Nothing. No SDK, no build, no engineering. We capture on the released game.

How do payouts work?

We recover direct costs, then split net revenue 50/50. Quarterly statements.

Could a model recreate my game?

Labs are contractually barred from training models to recreate titles. And we never ship source code, assets, or mods: nothing that could rebuild your game.

My game is niche. Will it sell?

Labs buy libraries, not single titles. Your game earns inside a normalised catalogue that sells together, so it doesn't have to carry a deal alone.

Is this "selling my game to AI"?

No. A controlled, non-exclusive licensing channel with scope you approve and the right to opt out.

For players & communities

Is this just AI taking from creators again?

No, and the anger at that model is justified. Nothing here is scraped: every game is licensed with its creator's consent, scope control, and the right to opt out, and half of net revenue goes back to the studio. If AI learns from a game through us, its makers said yes and get paid.

Will this be used to clone games?

No. Labs are contractually barred from training models to recreate titles, and we never provide source code, assets, or mods.

What does this data actually build?

World models, robotics, and embodied AI: systems that need to understand space, physics, and cause and effect. Games are the richest source of that understanding ever made, and this way their creators share the upside.

Does this replace game developers?

No. It pays them. The data is valuable because their worlds are: this is game craft being licensed like any other creative work, not harvested.

Legal & security

Where does the right to train come from?

Every session is captured under a direct licence with the IP owner. The licence ID and rights scope travel in every session's manifest.

Is anything scraped or from public sources?

No. Licensed capture only. If it's in the catalogue, there's a signed licence behind it.

How is access controlled?

Data is granted per order to your workspace: key-based auth, versioned delivery, access scoped to your grant.

Build on real worlds.

Building frontier AI?Talk to us →
Have a catalogue?List your game →