For AI labs

Simulation-grade game data.

Granular enough to rebuild a gameplay moment 1:1 in another engine. Proven live: ZENOS game twins re-render running games in Unreal Engine 5. Every entity, transform, input, and event, frame-perfect. Ordered in hours, delivered as plain files.

pip install zenos-data · scriptable end to end · machine-readable output on every command
The package

Plain files. Open formats.

A session is a folder you can open: standard video files for every visual signal, one open state file for everything else, and a manifest that describes it all. No database to stand up, no runtime to adopt. Read it with tools you already have. And because every title is normalised to one coordinate convention, a 38-title grant behaves like one dataset, not 38 integrations.

Video as mp4

RGB and every render buffer (depth, normals, segmentation) as standard, frame-aligned video files. Any decoder reads them.

State as VTX

World state, inputs, actions, and events, frame by frame, in VTX: our open Apache-2.0 format with a public spec, SDK, and inspector. github.com/ZenosInteractive/VTX

One manifest

Capture specs, session details, licence and rights scope. Every session is self-describing and auditable.

Prefer columns or plain text? One command converts state to Parquet or JSON. Keep the originals, convert what your pipeline wants.

The CLI

One tool. The whole workflow.

Install our Zenos Data CLI in seconds, authenticate with your key, and work your granted data from your shell: search, pull, convert, view. Built to be scripted, with machine-readable output and stable exit codes on every command.

Search your grant

Filter everything you've licensed by title, activity, and around 20 metadata categories. See matching hours instantly, down to frame ranges.

Downloads that finish

Resumable and checksummed. The CLI babysits multi-session orders; a dropped connection or a crash costs you nothing.

Convert on demand

State to Parquet or JSON in one command, in place. Video stays mp4. Nothing else to install.

Example

Search in. Training data out.

the whole workflow · one tool
$ zenos grantsorder_0112 · 50,000 hrs · 38 titles · granted
$ zenos search --label driving --weather rain312 sessions · 1,840 hrs in your grant
$ zenos pull order_0112 --matchvideo + buffers + state + manifest
$ zenos convert ./order_0112 --parquetoptional · state to parquet
$ zenos view session_00428 --rerunframe-synced playback in Rerun
Compatibility

Drops into the stack you already run.

Labs inspect this kind of data in Rerun, so it's first-class: one command streams any session into the Rerun viewer, with video, buffers, and state in frame-synced playback. Everything else is standard by design: mp4 any decoder reads, state that converts to Parquet or JSON in one command, plain files any dataloader can mount.

Rerun · one-command playbackmp4 · any decoderParquet / JSON · one commandYour dataloader · plain files
How access works

Browse everything. Unlock what you license.

The whole catalogue is searchable in the lab portal: titles, labels, and hours, up front. The data itself unlocks per order. Agree a volume of hours across the titles you want; every title is already licensed, each IP owner gives a quick sign-off on the use, and the data lands in your workspace. From there it's self-serve: search at full depth, pull, train. Every session carries its licence ID and rights scope in the manifest.

01

Browse

The full catalogue, searchable in the portal. Locked until licensed.

02

Scope

Agree hours, titles, and signals with us.

03

Signed off

Already licensed. Each IP owner confirms the use, built into the deal.

04

Granted

The data lands in your workspace.

05

Train

Search, pull, convert. Self-serve from here.

Lab portal · catalogue, grants, listsSample pack on requestSame-day eval accessKey-based authLicence ID per session

Start building.

Request a sample pack and a schema walkthrough. Install the CLI and pull it the same day.