Your data never reaches the model.
Server-side enforcement, self-hosted detection, and an audit trail for every decision. Built so your security team can verify it instead of taking our word for it.
Enforcement that lives on the server and not in the browser
Every message takes the same path. Detection, masking, and the encrypted map all run inside our boundary, so protection never depends on the client doing the right thing.
Raw text stops at our server. It is scanned and masked there, before a single token reaches the model provider, so only placeholders ever leave our boundary. An encrypted map, kept per chat, restores the real values when the conversation renders back to you.
Detection is self-hosted. The model that reads raw text runs on our own infrastructure, cached locally, and is never called out to a third-party inference API. It sits inside the same boundary as the database.
Security controls that ship with every workspace
The real surfaces admins use to govern a workspace, working exactly as they do in the product.
Replacement map
Detected values become stable tokens the model can reason with.
{"[NAME-1]": "Jane Example","[EMAIL-1]": "jane@example.com","[SSN-1]": "000-00-0000","[CARD-1]": "4242 4242 4242 4242"}PII audit logs
Every decision the gate makes lands in an append-only log. Review it, filter it, and export it when the audit asks.
Admin enforced policies
Admins decide which models and modes a workspace may use.
Workspace labels
Choose which entity groups the scanner enforces, workspace-wide.
Pick how strictly every chat is enforced
Each chat runs in one mode, locked at creation and set by workspace policy. The same message, handled four different ways.
Smart Mask
Values become stable tokens before the message leaves. The reply is restored on screen for the reader.
Block
Sending pauses when sensitive data is found. Nothing leaves until it is redacted or removed.
Sensitive data detected
This message contains sensitive data.
Warn
The send pauses and the user decides. The original text goes through only after Send Anyway.
Review before sending
This message contains sensitive data.
Highlight
Values are highlighted for visibility and the message goes through unchanged. The model sees it verbatim.
Built for the people accountable when AI meets data
IT leadership, data protection, compliance, security operations, and the teams doing the work. Each gets the control their role answers for.
IT Leadership
Adopt AI without surrendering control. Policy, models, and modes stay decided at the top and apply to every chat the same way.
Data Protection Officers
Detection runs on our own infrastructure and raw text is purged after masking. Replacement maps stay encrypted and retention windows stay yours.
Compliance and Legal
Every masking decision lands in an append-only log with CSV and PDF exports ready for review. Audit answers come from records, not recollection.
Security Operations
Enforcement runs server-side where it cannot be clicked away. Role-based access and usage logging cover every workspace surface.
Workspace Admins
Model allowlists, enforced modes, and label groups apply workspace-wide from one settings surface. No member can opt out.
Department Teams
Members just chat. Masking, placeholders, and restores happen around the conversation, so everyday work keeps its pace.