Add a privacy layer before model calls without forcing your startup to adopt a specific AI provider. PrivoNest detects sensitive text, swaps it for scoped aliases, sends the protected prompt to the selected provider, and restores exact aliases in the application response.
Detect → Alias → Selected AI → Restore
The API does the PII detection. Your chosen LLM does the generation/reasoning.
The v1.0.5 SDK includes adapters for OpenAI, Anthropic/Claude, Google Gemini, Ollama, LangChain and LlamaIndex; integration-test against the exact third-party SDK versions your application uses.
The Guard calls the PrivoNest PII API before the model request so raw detected values can be replaced before leaving your application path.
Temporary aliases such as [PN_NAME_1] can be restored in short-lived application-side memory after the selected AI responds.
Keep both the PrivoNest API key and model-provider key in your backend environment, not browser JavaScript.
OpenAI is one adapter. The customer chooses the AI provider; PrivoNest remains the privacy layer.
Automatic provider wrapping remains focused on textual LLM inputs. Supported PDF, DOCX, XLSX, CSV, image, email, DICOM and transcript protection remains available through the PrivoNest REST /v1/hide document workflow.
Your backend receives normal text.
PrivoNest identifies sensitive spans.
Your chosen model receives scoped aliases.
Exact aliases are restored in the application response.