Drafting from redacted packs
Prepare letters, summaries or meeting notes from desensitised packs so cloud models never see raw names or identifiers when you choose a cloud path.
Budbox · Schools
Safeguarding first. Cloud second — and only airlocked.
Teachers and office staff are already using generative AI. Budbox gives the school a locked-down box on site: desensitise documents before any optional cloud call, run local models when needed, and keep egress to allowlisted endpoints only.

Prepare letters, summaries or meeting notes from desensitised packs so cloud models never see raw names or identifiers when you choose a cloud path.
Give staff a school-controlled AI path instead of personal accounts pasting pupil detail into consumer chat products.
Use Budbox as a Dexl node with conditional questions in admin workflows — for example confirming whether a document is cleared before an AI step runs.
Real UK frameworks and guidance named for context only. Budbox is not claimed as certified or endorsed by these bodies on this page.
Schools are controllers of pupil and staff personal data. Budbox is designed so processing can stay on premises and outbound AI calls can avoid raw personal data via the airlock. This page does not claim a specific certification.
The Department for Education has published generative AI guidance for the education sector covering safe, effective use. Budbox supports that direction by keeping a school-owned appliance, local options, and controlled egress — it is not an official DfE product.
Safeguarding records are highly sensitive. Design intent is that such material is not the default outbound payload to cloud AI; local models and airlock desensitisation are the control points. Final policy remains the school’s.
Across every sector page the product facts stay the same: locked-down Docker AI on a dedicated box, airlock desensitisation before optional cloud AI, fully local models when required, a local graph for domain enrichment, Dexl workflow nodes with conditional questions, and egress limited to allowlisted AI endpoints. Hardware is typically from around £1,000–1,500 with a ~12GB NVIDIA GPU-class card; billing is a hardware lease plus AI credits.
No. Budbox is an OnDraft appliance. References to DfE generative AI guidance describe the regulatory context schools already work in, not an endorsement.
Yes for workloads that use fully local models on the box. Optional cloud calls go through the airlock and only to allowlisted endpoints.
AI agents, integrations and operational software—built around your business.