Agent traces · Work traces
Ground truth others can't trace.
A model proxy sees none of this.
Our team turns both into training data for frontier labs and custom models for enterprises.
What screenshot recordings lose against structured browser state, measured on the same 100 agent runs.
Public-web browser tasks with paired screenshot and W8 recordings of each run. The data behind W8 DOM-diff Bench.
Runs coding agents in an isolated OpenShell sandbox, with the project filesystem on a PostgreSQL-backed FUSE mount.
For research and data partnerships: founders@wootzapp.com
Where the data comes from
Other training platforms record their traces at the model proxy: what the model was asked and what it answered. Our hypervisor and W8 browser engine record what the agent did and what the page held. A proxy-based trace holds three of the ten kinds of evidence below. Ours holds all ten.
Both records come from infrastructure we built. How they work
| What the traces hold | Othersmodel proxy | WootzApp |
|---|---|---|
| Model layer | via gateway | |
| Prompts in | ||
| Completions out | ||
| Tool calls and results in the model's context | ||
| Below the harness | hypervisor | |
| Processes | ||
| Network egress, including denied attempts | ||
| Policy decisions | ||
| File changes | ||
| Inside the page | W8 browser | |
| Full DOM state at every step | ||
| Exact values and control states | ||
| Off-screen content | ||
| enters a proxy-based trace only if your harness reports it. | ||
For frontier model labs
Recorded below the harness and inside the browser, past the point where a model proxy stops. Accepted by frontier-lab quality gates.
For enterprise AI teams
Bring your production harness. Change almost nothing.
How it works
The hypervisor and W8 record every run below the harness and inside the page.
HypervisorW8 browserOur team builds datasets, evals, and environments.
WootzApp teamTo frontier labs, or as a small model for your team.
A model proxy sees what the model said.
We see what the agent did.
Below the harness. Inside the browser.
The technology
An OpenShell-based sandbox records processes, egress, policy decisions, and file changes, whatever the agent logs.
A proxy sees the model call. The hypervisor sees what the call set in motion.
Architecture
Our Chromium browser records the page state screenshots miss: exact values, controls, and off-screen content.
A proxy sees the screenshot the agent sent to the model. W8 sees the whole page.
See the benchmarkBuilt on NVIDIA OpenShell; no NVIDIA endorsement implied.
FAQ
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.
Production traffic is captured at the model proxy: prompts in, completions out, and the tool calls that pass through the model's context. The hypervisor also records the processes, network egress, policy decisions, and file changes behind each call, and W8 records the full page at every step. The training set holds what the agent did and what the page held, alongside what the model said.
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.
No rewrite of core agent logic and no tracing SDK. Setup covers image packaging, sandbox configuration, credentials, and gateway routing.
Not by default. Enterprise training and model-lab data are separate contracts. External licensing requires explicit rights and permitted use.