Beta programme · 10 August – 10 November 2026
itAtlas has been tested hard against data we built ourselves — four corpora and about 1,400 automated checks. That is rigorous, and it is exactly the blind spot: we have never watched a stranger use it on their own mess. We are looking for five to ten enterprise architects to run it against a real, messy estate and tell us where it falls over.
The deal
The usual version of this asks for your data. This one does the opposite, and the difference is structural rather than a promise — see Security.
The full product, nothing gated — free while the Beta runs, then 40% off for life as a founding partner.
A direct line to the person building it, not a support queue.
Whatever your estate turns into. It is yours, and it never leaves your machine.
Run it against a real, messy estate — the actual exports, not a tidied sample.
Give the estate a shape: a reference model, or teach it your host-naming convention.
Send the structural report, tell us what confused you, and give us twenty minutes on a call.
Realistically two to three hours across three months.
Your estate. itAtlas has no server and no telemetry, and Beta does not change that — no analytics code was added for this programme.
Progress is tracked in your own browser, from your own database, and is never transmitted.
An NDA if your organisation wants one. Happy to sign yours.
What actually leaves your machine
This is the part to forward to your security or privacy person. There are exactly two files, you generate both, and the app shows you their full contents before you save them.
Contains: how each file was read — what it was classified as, how columns were mapped, value shapes (hostname-like, id-code), counts, and where the same key spans files.
Never contains: cell values, hostnames, IP addresses or connection strings. Values appear only as salted SHA-256 hashes, purely so duplicates can be counted. The salt is generated per export and never stored, so a hash means nothing outside one report.
Contains: your own answers, your own written feedback, and counts — how many assets, capabilities and lenses.
Never contains: asset names, hostnames, IP addresses, file contents or cell values.
Each is checked by the same leak detector that guards the app's optional AI features, and it shows you what it found rather than silently removing it. If you paste a server name into a feedback box by accident, it says so.
Optional AI is off unless you supply your own Anthropic key. With no key, nothing is sent anywhere at all. With a key, identifiers — server names, IPs, emails, database names — are tokenised before anything leaves, unconditionally, and the token vault stays on your device.
Who we are looking for
Best fit: higher education. Government and health estates are close cousins.
Estate size: a few hundred assets upward. Under 100 is welcome but teaches us less — the interesting failures need messy, multi-source data.
Data: you can get approval to run real exports through a local tool. If you can only use sample data, we already have that.
Formats: spreadsheets and CSV exports. The messier the better — that is the part being tested.
Sign-up is in the app. It saves locally, tells you straight away whether your estate is the kind that will teach us something, and sends nothing until you decide to.