AI privacy infrastructure · Now in early access
Send AI everything.
Just not your customers.
Cypherz sits between your application and any AI API — OpenAI, Anthropic, Google, or your own model. Sensitive fields are tokenized on the way out and restored on the way back. The model sees structure, never substance.
Your application
Plain request, real data
01 Detected sensitive fields
Waiting for input…
What the AI sees
Tokenized — no real data leaves
— awaiting input —Round-trip status
- Detecting PII
- Forwarding to AI
- Receiving response
- Restoring real values
Built for teams handling regulated, sensitive data
The problem
Every AI call is a data exfil event waiting to happen.
Models retain. Logs leak. Vendors get breached. Your customers don't care which AI you chose — they care that their email ended up in training data. Cypherz makes that physically impossible.
76%
of enterprises have blocked or restricted an AI tool over data leakage concerns
Gartner, 2025
11.2x
average exposure factor of PII in vendor logs vs first-party stores
Verizon DBIR
$4.88M
average cost of a data breach involving third-party AI processors
IBM Cost of a Data Breach
How it works
A four-stage pipeline.
Three of the stages aren't your problem.
Intercept
Your app calls Cypherz with the same payload you'd send to the AI provider.
POST /v1/proxy/openai
authorization: Bearer cyph_…Tokenize
Detectors swap PII for deterministic surrogates. Mapping persists encrypted, scoped to your project.
"Email <EMAIL_a1b2c3>
call <PHONE_d4e5f6>"Forward
We sign and forward to the upstream provider with your key. Cypherz never logs plaintext.
→ api.openai.com
/v1/chat/completionsRestore
Tokens in the response are replaced with the real values, inside your trust boundary.
"Sent to john@acme.com,
called +1 415 555 0142"Drop-in
Two lines of code.
Any AI. Any stack.
Swap your AI client's import line for our drop-in. The rest of your code is unchanged — we keep the same method signatures so nothing else has to move.
- Proxy mode — no client changes, swap baseURL
- SDK mode — drop-in for OpenAI, Anthropic, Google clients
- Local mode — tokenize without ever hitting our servers
import { OpenAI } from "@cypherz/sdk/openai";
const client = new OpenAI({ cypherzKey: process.env.CYPHERZ_KEY });
const resp = await client.chat.completions.create({
model: "gpt-4o",
messages: [
{ role: "user", content: "Email john@acme.com a quote." },
],
});
// ↳ The model sees: "Email <EMAIL_a1b2c3d4e5f6> a quote."
// ↳ You receive: "Email john@acme.com a quote."
Built for production
The stuff that turns a demo into
infrastructure.
Envelope encryption
AES-256-GCM with per-project data-encryption keys, wrapped under a single master key. Plaintext never persists on disk.
Deterministic tokens
Same input maps to the same token within a project — joins, dedupe, and analytics all keep working on tokenized data.
Custom detectors
Built-in detectors for email, phone, IP, SSN, credit card. Add your own regex or schema-aware rules per project.
First-class audit trail
Every tokenize, detokenize, and proxy call is logged with structured metadata. Export to your SIEM in one query.
Bring-your-own keys
Store provider keys with us encrypted at rest, or let Cypherz provision them so your developers never touch a key.
Self-host or hosted
Run the entire stack in your VPC with one docker-compose, or use our managed cloud. Same binaries, same guarantees.
Honest comparison
We're not magic.
We're just the right shape.
| Approach | Setup | Coverage | Restorable | Audit trail |
|---|---|---|---|---|
| No protection | — | 0% | — | — |
| Regex string-replace | Hours | ~40% | Manual | Roll your own |
| DIY redaction service | Weeks | Variable | Often lossy | DIY |
| Cypherz | Minutes | Built-in + custom | Deterministic | First-class |
Pricing
Pay for traffic, not for seats.
Developer
Free
Self-hosted or up to 25k requests on us.
- Unlimited projects
- All built-in detectors
- TypeScript SDK
- Community support
Team
$249/mo
For startups shipping AI features into prod.
- 1M requests / month included
- Audit log retention 90 days
- Custom detectors
- Email + Slack support
- SOC2 report
Enterprise
Custom
Self-hosted, BYO HSM, dedicated SLA.
- Self-hosted in your VPC
- Bring-your-own KMS / HSM
- SSO, SCIM, role-based access
- Dedicated SLA
- DPA, BAA available
Get started
Ship AI features without shipping your customers' data.
Sign up, create a project, copy your API key. The first request is tokenized in under sixty seconds.