Skip to content

SDK & local proxy

Protect before the model call.

Run detection and restoration in your own process. Keep the swap table inside your application’s trust boundary.

Small surface.
Explicit data flow.

The core exposes protect and restore. Your application sends the protected text to its model provider, then restores the response using the same local vault.

The example illustrates the core workflow; callYourModel is your own integration. The SDK provides client wrappers and streaming helpers. Review the full documentation before using it in production.

Read the SDK docs ↗

TypeScript · core workflow

import { protect, restore }
  from 'tightlip';

const result = protect(prompt);

// Your model integration
const reply = await callYourModel(
  result.text
);

const readable = restore(
  reply, result.vault
);

Use the SDK

Wrap OpenAI or Anthropic clients. Preserve the vault for the full request/response lifecycle, including streamed output.

View downloads ↗

Use the local proxy

Point a compatible tool at a proxy in your environment. Review network access and metering behavior in the docs.

View downloads ↗

THE LOCAL PROTECTION WORKFLOW

Conceptual diagram

01 / YOUR DEVICE

Find & replace

Client emailConsistent placeholder

The original values stay in your local swap table.

02 / YOUR REVIEW

Check what remains

Instructions + context
+ placeholders

Check missed details and identifying context before sharing.

03 / YOUR MODEL INTEGRATION

Send reviewed text

The provider receives the text you choose to share.

Restore supported replies locally using the same swap table.

The free browser check demonstrates protection and restoration only; it makes no AI request. Sending text and handling replies depend on your chosen workflow. Account and usage data are separate.

Remove sensitive details before your next AI prompt.

Try the free check ↗