The AI Architecture Lab The 30-minute verdict
Security and data

Your data never leaves.

The agent layer lives inside your environment and works through official APIs. The only way out is one logged exit for public lookups, and you hold the kill switch.

The boundary

Watch what crosses, and what does not

Three rules

  • Inside: the agent layer reads and writes your systems of record through official APIs.
  • One exit: an egress proxy permits public reference lookups only, and logs every call.
  • Your switch: you hold the keys, the accounts, the audit log and a kill switch on every run.

Models run in your cloud account or on your own servers. Inbound is public data only.

Ownership

Everything is yours, from the first commit

Source codeYours
Test and evaluation suitesYours
RunbooksYours
Your dataYours
Keys and accountsYours
Audit logYours

No licence fees. We keep no copy of your data and no licence over your code.

During the Blueprint

How we handle your information

First

NDA signed

Before any discovery work begins.

Access

Read-only

Read-only accounts, scoped to the processes in review.

Storage

No copy kept

We do not copy or store your data outside your environment.

Models

No training

Your data never trains any model, ours or anyone else's.

Plain dealing

What we do not claim

No borrowed badges

We do not claim certifications we do not hold. Your security review sees the actual controls: the boundary, the proxy log, the audit trail and the kill switch.

Named subprocessors

Any subprocessor is named in writing before it is used. Inference can run in an India region or on your premises.