Privacy on
the edge.
An AI inference proxy that detects sensitive data in your prompts and routes it to edge models — so PII never leaves the node closest to your user.
One proxy.
Two destinations.
Privedge sits between your app and the AI API. Every prompt is inspected at the edge — sensitive data is answered locally, everything else continues to the cloud.
Each prompt is scanned for PII before a single token is sent anywhere.
Sensitive prompts go to the edge model. The data never leaves the node.
Clean prompts hit OpenAI or Anthropic as usual — zero added latency.
Watch a prompt get routed.
Detection runs on every keystroke, exactly like the worker at the edge. Add an email or an SSN and watch the route flip.
ROUTED → EDGE7 sensitive entities detected. This prompt stays on the local node — nothing reaches a third-party API.
{
"routed_to": "edge",
"model": "@cf/meta/llama-3.2-1b",
"pii_matches": 7,
"latency_ms": 179
}Sensitive data, caught
before it leaves.
v1 ships a fast regex detector. v2 adds a named-entity model on Workers AI.
Change one import.
Keep your whole stack.
@privedge/sdk mirrors the OpenAI client one-to-one. Swap the import, point at your worker, and routing happens transparently.
import Privedge from '@privedge/sdk' // Same surface as the OpenAI SDK — change one import. const ai = new Privedge({ apiKey: 'YOUR_OPENAI_KEY', workerUrl: 'https://privedge.workers.dev', compliance: 'hipaa', }) const res = await ai.chat.completions.create({ model: 'gpt-4', messages: [{ role: 'user', content: prompt }], }) console.log(res.routed_to) // 'edge' | 'cloud'
Built for compliance from request one.
US health data — PHI never reaches a third-party API.
EU personal data stays within the processing region.
Card numbers are detected and kept off the cloud path.