For frontier model labs
Datasets, evals, and RL environments.
Built from traces nobody else captures. Accepted by frontier-lab quality gates.
- Full page state, every step
- Runtime truth below the agent
- Deterministic rewards
Agent traces · Work traces
The hypervisor
for RL.
Ground truth others can't trace.
- HypervisorRuns your agent unchanged and records processes, network egress, policy decisions, and file changes below the harness.
- W8 browser engineOur Chromium engine records the full page state at every step: exact values, controls, and off-screen content.
Our team turns both into training data for frontier labs and custom models for enterprises.
Agentagent traces
Expertwork traces
W8 browserPage state at every step
HypervisorProcesses · egress · file changes
WootzApp teamEngineered into training data
Datasets · evals · environmentsfor model labs
Custom small modelsfor enterprises
Our work
All work →- Benchmark· Oct 2026W8 DOM-diff Bench
What screenshot recordings lose against structured browser state, measured on the same 100 agent runs.
100 tasks639 criteriaPaired evaluation4.07% → 0.31% missing evidence - Dataset· Sep 2026Browser Agent Tasks
Public-web browser tasks with paired screenshot and W8 recordings of each run. The data behind W8 DOM-diff Bench.
100 tasks90 sitesPaired runs930 recorded browser states - Open-source runtimeOpenrind Shell
Runs coding agents in an isolated OpenShell sandbox, with the project filesystem on a PostgreSQL-backed FUSE mount.
OpenShellFUSEPostgreSQLAll /sandbox/work I/O through one daemon
For research and data partnerships: founders@wootzapp.com
For enterprise AI teams
Small models trained on your agents' real work.
Bring your production harness. Change almost nothing.
- No harness rewrite
- Only verified runs in training
- Frontier fallback at your gateway
How it works
From trace to training data.
- 1
Capture
The hypervisor and W8 record every run.
HypervisorW8 browser - 2
Engineer
Our team builds datasets, evals, and environments.
WootzApp team - 3
Deliver
To frontier labs, or as a small model for your team.
Others train on what your harness exposes.
We reveal what it cannot see.
The technology
Two observation points we built.
OpenShell sandbox
Your harnessunchanged
- Model request
- Proxy decision
- Workspace update
- Egress attempt
01Hypervisor
Below the harness
An OpenShell-based sandbox records processes, egress, policy decisions, and file changes, whatever the agent logs.
Architecturescreenshot0.pngScreenshot run
dom_model0.txtW8 run
ethche released this 02 Sep 17:40
02W8 browser
Inside the browser
Our Chromium browser records the page state screenshots miss: exact values, controls, and off-screen content.
See the benchmarkBuilt on NVIDIA OpenShell; no NVIDIA endorsement implied.
FAQ
Questions.
What does WootzApp sell?
Datasets, evals, and RL environments for frontier model labs, and custom small models for enterprises. Both are built from traces captured by our hypervisor and the W8 browser.
What is a work trace?
A recording of an expert doing a real task inside our hypervisor and W8: actions, runtime events, and page state at every step. An agent trace is the same record produced by an AI agent.
Do we need to change our agent?
No rewrite of core agent logic and no tracing SDK. Setup covers image packaging, sandbox configuration, credentials, and gateway routing.
Do you resell enterprise data?
Not by default. Enterprise training and model-lab data are separate contracts. External licensing requires explicit rights and permitted use.