Integrations

Add protection where your AI work already happens.

For people and teams choosing a setup. Pick the path that matches how you reach an AI model today.

Same swap, different places.

Each path finds sensitive details in your text and swaps them for placeholders before the text reaches an AI model. The swap table stays with you, so real values can be put back in the reply.

AI chat sitesYou use ChatGPT, Claude, Gemini, Copilot or Perplexity in the browser.
Your own applicationYour code calls an AI model and you want protection inside your process.
Local proxyA tool on your machine calls an AI provider and you want protection in between.

Go to downloads ↗

AI chat sites

The browser extension works on five AI chat sites. It swaps details when you press send and shows real values in the answer on your screen. It installs by hand in Chrome or Edge.

See the extension and install steps

Works with the AI tools you already use

ChatGPT

Browser extension compatibility. Manual installation required.

Your own application

The engine runs inside your process. Call protect() before the model call and restore() on the reply. A stream restorer handles streamed replies.

The current SDK is a tarball on the download page. It includes wrappers for the OpenAI and Anthropic clients and needs Node 18 or later. The package still uses the Tightlip name.

npm publishing is planned and not available yet. Until it is, install from the tarball link and do not run npx tightlip on its own.

Protected prompts use credits from your account. Your model provider bills you separately. See pricing.

Install from the current tarball

npm install https://tightlipai.com/downloads/tightlip-0.1.0.tgz

Download page ↗Developer overview SDK reference

Local proxy

The SDK package also includes a proxy command. It runs on your own machine and sits between a compatible tool and its AI provider, so the tool’s prompts can be protected without changing its code. Use it when you cannot add the engine to the tool directly. Setup, provider routing and restore behaviour are in the local proxy guide.

Limits on every path

Detection can miss a name that has no cue around it. The rest of the text can still identify a person.
The browser extension does not scan attachments, and a chat site’s page can read text while you type.
Review protected text before it goes to a model, whichever path you use.

Remove sensitive details before your next AI prompt.

Try the free check ↗