VeilPhantom

The product is an on-device privacy engine that detects sensitive data such as names, amounts, and identifiers in text and replaces them with tokens before data reaches AI models. It is for developers, IT and data teams in businesses that use AI tools with confidential transcripts, documents, and chat logs. It is delivered as a desktop application and an open-source SDK that runs locally with tokenization and rehydration on the device.

Key features

• Detect names, amounts, and identifiers in text
• Tokenise sensitive values with structural tokens
• Rehydrate tokens to original values on device
• On-device processing with zero data leaves
• Phonetic embeddings for speech-to-text errors
• On-device transcription with Parakeet TDT v3
• AI summaries via Claude Haiku
• Memory Garden knowledge graph
• Memory Chat to query meeting history
• SDK with 7 detection layers and 19 entity types
• Works with OpenAI and LangChain