GhostFilter Docs

GhostFilter AI is a safety firewall for humans and AI agents.

It protects against two connected problems:

• scams, phishing, impersonation, unsafe links, and risky files for people
• prompt injection, jailbreaks, secret extraction, and unsafe tool-use instructions for GhostGPT and other AI agents

The reusable developer SDK is published on npm:

npm install ghostfilter-ai

Open the npm package

What the SDK is

An SDK is a software development kit: code another developer imports into their own application. ghostfilter-ai packages GhostFilter's trained logistic-regression weights, scam rules, prompt-injection firewall, safe-context generator, and command guard as a small TypeScript/JavaScript library.

The SDK does not embed the GhostFilter website. It gives developers callable functions:

import { ghostfilter } from "ghostfilter-ai";

const result = await ghostfilter.protect({
  input: externalEmailBody,
  mode: "full",
});

if (result.verdict !== "safe") {
  console.warn(result.reasons);
}

Core checks run in the developer's Node.js process. No account or API key is required.

What you can build

• Add scam/phishing checks to your app
• Add prompt-injection checks before an AI agent reads external content
• Wrap untrusted content in safe GhostGPT context
• Guard explicit shell commands in CLIs or developer tools
• Use the browser extension to scan selected webpage text or whole pages

Main surfaces

Trust model

GhostFilter is a pre-processing safety layer, not an antivirus or a guarantee that content is safe. Applications should retain human approval for payments, credential changes, message sending, file deletion, and other consequential actions.