Solutions · Government

Can it run with no connection out?

Wheelhouse runs on your premises and air-gapped. Every record is write-once and tamper-evident. Records are linked so any change shows. CERT-In: within 6 hours. Data Protection Board: first notice without delay, full report within 72 hours.

Wheelhouse runs on your premises and air-gapped. Tower runs inside your own cloud account.

Last reviewed

Illustrative. No customer data.

The difference

Their way, and ours.

Their wayRaidu
Audit evidence held in a vendor's cloud.Write-once records, inside your perimeter.
Verification requires a call to the vendor's API.Records are linked so any change shows. No call to a vendor to check a record.
A different tool for each regulation.One policy stack, tamper-evident evidence, across your AI estate.

Every record is write-once and tamper-evident. Records are linked so any change shows.

Source: Wheelhouse Proof Pack and tamper-evident records

What procurement and security ask

Answered with what the system does.

Can it run with no connection to the internet?

Wheelhouse ships air-gapped and runs on your premises. Every record is write-once and tamper-evident. No outbound call is needed to produce an audit entry.

Can a third party verify the records without vendor access?

Yes. The Proof Pack includes a chain of tamper-evident records. Records are linked so any change shows. Any auditor can check any decision independently, with no vendor access required.

Is there a single policy stack for the whole AI estate?

Yes. Every call from people, applications, and agents is checked against the same policy set. 61 guardrail policies across 36 detection services run on the full prompt before the model is called.

The product

See the screens.

Product windows from the films, with sample data. No customer data.

The record

The decision, on the record.

Every AI call produces a tamper-evident entry. Records are linked so any change shows.

Entry 2,091 Policy checkpoint · 09:31:17Z
Prompt
Document summary · no personal data detected
Policy
Checked · passed
Verdict
Allowed and logged
Location
On premises · no outbound call
Proof
Tamper-evident
sha256 6242a77d4ad05dba4fde78a89eb20330…
entry 2,091 of 11,847 · write-once

The call path

Where the decision runs.

  1. InterceptEvery call, inside your perimeter
  2. InspectPolicy checks on the full prompt
  3. DecideAllow or refuse before the model is called
  4. RecordWrite-once, on premises
  5. ProveRecords linked so any change shows

Also covered

  • On premises
  • Air-gapped (Wheelhouse)
  • Tamper-evident
  • Write-once records

Common questions

Answered straight.

Which components run air-gapped?
Wheelhouse ships air-gapped and runs on your premises with no outbound network required. Tower runs inside your own cloud account.
How are records kept tamper-evident?
Records are write-once and kept inside your own infrastructure. Each record is linked to the one before it, so any change shows.
Can personal data in prompts be kept off the model?
Yes. Raidu detects personal data in the prompt and replaces it with coded stand-ins such as PERSON_1 before the call reaches the model. The substitution is logged and on record.
Is there a single audit trail across the whole AI estate?
Yes. Every call from every user, application, and agent goes through the same intercept path. The audit chain is unified and tamper-evident.
Does any call data leave our perimeter?
No. Raidu runs inside your own infrastructure. No call data, prompt content, or audit record reaches Raidu.

Sovereign by design.