AI STARTUP
🤖
Protect User Prompts Before Any LLM
Wrap the model client so names, emails and local identifiers can be replaced before an external LLM receives the prompt.
View scenario →PrivoNest combines multilingual PII protection and AI prompt guardrails with API Security Preflight v1.0.5 for developers building sensitive-data and AI-powered applications.
37 production languages · selected examples shown below
A developer layer for multilingual PII protection, AI prompt privacy and pre-deployment API code security.
Production language packs span Latin, Arabic, Indic, CJK, Cyrillic, Greek, Thai and other supported scripts.
153 country profiles include 52 strong-format and 101 baseline-contextual profiles for local identity, tax, health, banking and registration workflows.
spaCy NER, contextual patterns, normalization and checksum validators work together instead of relying on one technique.
Label, redact, partial mask, hash, tokenize, detokenize and remove sensitive values for different workflows.
Protect supported text, JSON, PDF, DOCX, XLSX, CSV, images, email, DICOM and transcript workflows.
A Python guardrail SDK can protect prompts before OpenAI, Claude, Gemini, Ollama, LangChain or LlamaIndex calls.
Create PrivoNest PV1 tokens for trusted workflows and reveal them through an authenticated API when authorized.
Language and country inference can activate relevant packs without forcing every application to hard-code routing logic.
Scan supported Python/JS/TS projects with 62 detector categories, AST/cross-file analysis, private dependency intelligence and CI reporting.
Three steps from raw content to policy-safe output.
POST content to /v1/hide with your API key. Language and country hints can be explicit or inferred.
Multilingual NER, deterministic recognizers, context rules, normalizers and validators identify sensitive entities.
Return labels, redaction, partial masks, hashes, reversible tokens or removed values depending on the selected strategy.
The same PrivoNest engine can detect sensitive values, transform them for a downstream system, preserve useful structure, and restore reversible tokens only when your application chooses to.
Add a privacy layer between your application and an LLM. Sensitive values are detected before the model call, replaced with scoped placeholders, and matching placeholders can be restored in the application response.
from openai import OpenAI from privonest_guardrail import Guard # Two-line privacy integration guard = Guard() client = guard.wrap(OpenAI()) # App prompt "My name is Ali and my email is ali@example.com" # What the model can receive "My name is [PN_NAME_1] and my email is [PN_EMAIL_1]"
PrivoNest is the privacy layer; your client keeps control of the model provider and its own provider API key. The PrivoNest SDK is v1.0.5 stable; third-party provider adapters should still be integration-tested against the exact provider/library versions used in production.
Run PrivoNest Preflight against supported Python, JavaScript and TypeScript project files. The engine uses AST, cross-file context and private dependency intelligence to detect security signals that simple file-local signatures can miss.
$ privonest . --fail-on high Source code ↓ PrivoNest API Security Engine ↓ AST + cross-file + dependency checks ↓ Root-cause triage + reachability ↓ Findings + confidence + fingerprints ↓ CI policy / baseline / SARIF External AI for source scan: NO
Protect PII before it moves into analytics, support systems, documents, AI applications or downstream APIs.
Patient identifiers, clinical text, health IDs, documents, images and DICOM-oriented privacy workflows.
Cards, IBANs, bank details, tax identifiers, national IDs and customer data before AI or processing.
Prompt protection for copilots, RAG applications, support bots and internal AI tools.
Redact sensitive information in email, chat, tickets, transcripts and exported support data.
Student IDs, contact information, enrollment records and sensitive documents.
Employee, applicant, contract and identity information in internal workflows and AI-assisted processing.
Same endpoint, multiple scripts and country-aware protection — selected language examples below.
These are implementation patterns based on current engine capabilities
Wrap the model client so names, emails and local identifiers can be replaced before an external LLM receives the prompt.
View scenario →Use country-aware rules, validators and code-switching support for workflows where English and local-language data appear together.
View scenario →Send PDFs, office files, images, email, DICOM or transcripts through one privacy engine and receive protected output plus detections.
View scenario →Start free. Scale as you grow.
Try it out
For production workloads
For large-scale usage
Use one PrivoNest account for PII/file protection, AI Privacy Guard and API Security Preflight. The same credit wallet is shared across these developer workflows.