Use AI without exposing your sensitive data
Incognitify masks PII, secrets and identifiers in your prompts before they reach an LLM — then restores the real values in the response. Your team gets clean answers; the provider never sees the raw data.
No credit card required · Keys encrypted at rest · Vault never written to disk
Your prompt
Look up the account for john@example.com, SSN 123-45-6789, callback 415-555-0132.
What the provider sees
Look up the account for ⟦EMAIL_1⟧, SSN ⟦SSN_1⟧, callback ⟦PHONE_1⟧.
Masked before it left your machine
- EMAILjohn@example.com
- SSN123-45-6789
- PHONE415-555-0132
How it works
A round trip: mask on the way out, restore on the way back. Your data never leaves in the clear.
- 01
Detect & mask
Incognitify scans your prompt for emails, phone numbers, cards, SSNs, API keys and more — replacing each with a safe, reversible token.
- 02
Call your own LLM
The masked prompt is sent with your own Claude or OpenAI key. The provider only ever sees tokens like ⟦EMAIL_1⟧, never the real values.
- 03
Rehydrate the reply
Tokens in the response are swapped back to the original values in memory, so your team reads a complete, natural answer.
Built for teams that take data seriously
Everything you need to put a safety layer between your prompts and your LLM provider.
Bring your own keys
Use your own Claude and OpenAI keys. No per-token markup — you keep full control of model choice and spend.
Detectors built in
Emails, phone numbers, credit cards, SSNs, IP addresses, API keys and UUIDs are caught out of the box.
Ephemeral vault
The token-to-value map lives in memory for a single request and is never written to disk.
Encrypted key storage
Provider keys are sealed with envelope encryption (AES-256-GCM). Only the last four characters are stored in plaintext.
Team & org controls
Share keys across your organization or keep them personal. Invite members and assign roles.
Audit & usage
Every masked request is logged with what was detected, so you can review activity and track usage.
Map your own data
Beyond the built-in detectors, you'll be able to define custom rules for the identifiers unique to your business — internal customer IDs, project codenames, employee numbers, account references — so they're masked automatically, every time.
- Point at a field or pattern and choose how it's masked
- Reusable rule packs shared across your organization
- Consistent tokens, so your model still reasons correctly
Generate this image elsewhere, then replace this block. Click the prompt to select it.
A clean, modern SaaS UI screenshot of a "custom data mapping" rule builder. Light theme, indigo (#4F46E5) accent, Inter font, rounded corners, soft shadows, generous whitespace, enterprise aesthetic. Left panel: a list of detection rules (Customer ID, Project codename, Employee number). Center: a rule editor with a field-name input, a pattern field, and a "mask as" dropdown. Bottom: a sample-text preview where matched values are highlighted and replaced with tokens like ⟦CUSTOMER_ID_1⟧.Simple, transparent pricing
Start free. Upgrade when your team needs shared keys and oversight. You always bring your own LLM keys — no token markup.
Free
For solo work and trying things out.
$0
Start free- Personal provider keys
- All built-in detectors
- Masked chat playground
- Ephemeral in-memory vault
Team
For teams that share access and need oversight.
$20 / seat / mo
Get started- Everything in Free
- Shared organization keys
- Unlimited members & roles
- Audit log & usage tracking
Enterprise
For organizations with security and scale needs.
Custom
Contact sales- Everything in Team
- SSO & advanced security
- SLA & dedicated support
- Custom contract & invoicing
Frequently asked questions
Which LLM providers are supported?
Anthropic (Claude) and OpenAI-compatible APIs today, with more provider adapters on the way. You connect your own key for each.
Do you ever see my API keys?
Your keys are encrypted at rest with envelope encryption. At request time the proxy decrypts a key in memory to call your provider — it is never logged or stored in plaintext. Only the last four characters are kept, for display.
Is my prompt data stored?
No. The token-to-value vault is in-memory and scoped to a single request — it is never written to disk. We record metadata for the audit log (which types were detected), not your raw values.
How accurate is detection?
Incognitify removes a large class of sensitive data automatically, but no detector is perfect. Treat it as a strong safety net that reduces risk — not a guarantee, and not a replacement for your own compliance review.
Can I self-host?
The core masking engine is open source, so you can run it yourself. Incognitify Cloud is the hosted, team-ready layer with shared keys, billing and audit.
Browse the source on GitHub →Start masking your prompts today
Connect your Claude or OpenAI key and send your first protected request in minutes.