AI agents are becoming the front door for customer conversations. But when a customer shares an email, phone number, account number, PAN, Aadhaar, or card detail, that data can end up directly inside an LLM prompt. I built Pryvra to change that. Pryvra replaces sensitive values with stable tokens before the model call. The model receives useful context without receiving the underlying customer data. The encrypted mapping stays in your environment and can be restored only where it is trusted, such as a human-agent handoff. What Pryvra supports today: • Detection for emails, phones, PAN, Aadhaar, IFSC, bank accounts, payment cards, and common API keys• Encrypted PII-to-token mappings• Conversation-aware stable tokens, preserving context without exposing identity• Controlled rehydration for trusted systems and human agents• Custom detectors for company-specific identifiers• Integrations for LiveKit and Pipecat• An open-source TypeScript SDK, available on npm Pryvra is useful for voice agents, chatbots, copilots, support workflows, healthcare, fintech, SaaS, and any AI product that handles customer data. I’d genuinely love feedback from builders: which framework, data type, or integration should we support next? GitHub: https://github.com/sahil-rajput/...: https://www.npmjs.com/package/@p... / Pipecat: https://pypi.org/project/pryvra-...

Pryvra

Keep customer PII out of AI prompts

Keep customer PII out of AI prompts