Inspect the whole request
Prompts, extracted document text and retrieved context need inspection. Production deployment adds malware scanning, OCR and sandboxed parsers.
YOUR KNOWLEDGE. YOUR BOUNDARY.
Enterprise AI.
Policy comes first.
One search experience. Approved AI tools. Client-specific rules that check questions and attachments before model processing.
Explore content verification ↘Designed for deployment in an enterprise’s approved private environment.
INTERACTIVE PRODUCT SHOWCASE
See the decision before the answer.
Use fictional data only. This page runs illustrative checks in your browser. It sends no prompts or attachments to a server or AI model, and does not save their contents. It is not a production security scanner or a GDPR certification.
01
Question & context
The selected policy applies to this run. Corporate sign-in, repository permissions and network enforcement are simulated.
02
Policy decision
Ready to verify
Choose an example and inspect the question before it reaches a model.
SESSION DECISION TRAIL
Evidence without storing the content.
Only decision metadata is shown in memory. Reset or reload clears it.
- No checks yet.
DEFENCE THROUGH MULTIPLE CONTROLS
More than a prompt filter.
Keep client boundaries
Combine enterprise and client rules. Check source permissions before retrieval and enforce model and tool permissions outside the LLM.
Answer from approved evidence
Use authorised sources, cite the evidence, and decline when evidence is missing. Check generated output and each proposed action again.
A PRIVATE DEPLOYMENT FOR EACH ENTERPRISE
Bring AI search
inside your boundary.
The target enterprise edition keeps uploads, extraction, classification, indexes, embeddings, inference and audit storage in the customer’s approved environment.
External models are optional and require an explicitly approved route. Even content verification must run internally when client policy requires it.
Production integration scope
- Corporate SSO, roles and client workspaces
- Permission-aware document connectors
- Internal models and controlled outbound gateways
- Versioned enterprise and client policies
- Retention, encryption and security monitoring
- Leakage testing and independent security review
This showcase demonstrates decisions only; these integrations are not connected here.
What these checks can and cannot establish
The demo uses a small set of patterns and fictional examples. A “permitted” result means no configured sample rule matched; it does not prove the content is safe. Real client data may have no obvious identifiers. Redaction may leave identifiable or confidential context. Prompt injection cannot be eliminated by a text filter. UK GDPR compliance also depends on lawful processing, contracts, rights, retention and applicable transfer arrangements.
Only UTF-8 plain-text and CSV files are inspected here. PDFs, Word files, images, encrypted files and unsupported content fail closed. Full document analysis and actual model responses are outside this demo.