Product
A workspace with a boundary that holds.
Your people get a full workspace on sensitive data, with their usual tools and private AI. Inside the boundary they work freely. At the boundary, every attempt to move data out is stopped, and your security team is told.
Drag your pointer outside the box and try to pull the data out.
Work the way you already do.
People open a browser workspace with everything already set up. They read, compute, train and ask questions of the real dataset instead of a sanitized copy. Only moving data out is controlled.
Three rings. The innermost one decides.
The control plane decides and alerts
Policies, approvals, alerts and the audit chain live in your own Command Center, one isolated stack per customer. Blocked attempts reach your security team in real time.
The agent carries it out
A small agent on each host connects outbound only and runs an allowlisted set of commands. There is no inbound SSH.
The kernel enforces and reports
Inside every workspace the kernel denies the forbidden operation itself and reports the attempt. If the dashboard falls silent, enforcement holds.
A chain of custody for every export.
Nobody approves their own request. Every window is temporary, and every decision is kept.
- i
Requested
A developer asks to move a specific file to a specific place.
- ii
Approved by another
A project manager or admin decides. Never the requester.
- iii
Window opened
A temporary allowlist opens for the hours approved.
- iv
Window closed
It closes on schedule, or sooner if someone revokes it.
- v
Sealed
Request, decision and revoke are chained so any later edit shows.
For the people who sign off, and the people who do the work.
Evidence your auditors can use: proof that sensitive data stayed where policy says it must.
- Kernel-level denial plus real-time alerts
- Two-person export approval
- Tamper-evident audit chain
- Single-tenant, air-gap capable
Work on the real dataset instead of a sanitized copy, with the tools you already know.
- VS Code, Jupyter and Python
- Private models, RAG and chat
- Exports in a few clicks
One Command Center for every host, workspace and alert, built for the 2 a.m. incident.
- Outbound-only agents
- Allowlisted commands only
- Golden-image provisioning
Questions about the product.
Does this slow my team down?
People work in a browser workspace with VS Code, Jupyter and Python ready. Reading and computing on data is unaffected; only moving data out is controlled.
What happens when someone needs data out?
They file an export request. A second person approves it, a temporary window opens, and it closes automatically. Every step is in the audit chain.
Who finds out when someone tries?
Your security team, in real time. Every blocked attempt raises an alert with the user, workspace, file and destination, and the same record lands in the audit chain.
Can we use AI on this data?
Yes. Private models run inside the protected workspace, on infrastructure you control. Public AI chatbots are blocked by default, like any other route out, and every attempt to use one is reported.
See the boundary hold, live.
Thirty minutes. We run the exfiltration test on a real workspace and size a deployment for your data.