Search: Encrypted Metadata Search

The Problem

Full-text search requires plaintext indexing. Existing search engines cannot index encrypted-at-rest metadata. There is no Gmail-style operator support for encrypted mail.

Every email client relies on plaintext full-text indexes. Gmail, Outlook, and Thunderbird all store searchable text unencrypted. When metadata is encrypted at rest, traditional FTS engines cannot read it. Users lose the ability to search their own inbox.

Semantic search compounds the problem. Embedding vectors expose content patterns even when the source text is encrypted. No existing solution combines keyword search, semantic ranking, and encryption at rest in a single pipeline.

The Old Way

The PRIVATE.ME Solution

A complete search pipeline that encrypts metadata at rest, supports Gmail-style operators (from:, to:, subject:, has:attachment), and fuses keyword and semantic results via Reciprocal Rank Fusion.

FTS5 full-text search operates on metadata decrypted in memory. The index itself is stored encrypted. Gmail-style query operators parse into structured filters that combine with free-text ranking.

Hybrid ranking merges BM25 keyword scores with cosine-similarity semantic scores using Reciprocal Rank Fusion (RRF). This produces better results than either method alone, without exposing plaintext to external embedding services.

The New Way

How It Works

Three-stage pipeline: parse query operators, dual-index search (keyword + semantic), and Reciprocal Rank Fusion to produce final ranked results.

Use Cases

Integration

import { createIndex, search } from '@private.me/search';

// Create encrypted search index
const index = await createIndex('user@example.com', {
  enableSemantic: true,
  embeddingProvider: 'ollama'
});

// Search with Gmail-style operators
const results = await search(
  'from:alice subject:invoice has:attachment',
  index
);

// Results ranked by Reciprocal Rank Fusion
results.forEach(r => console.log(r.score, r.subject));

Security Properties

Ship Proofs, Not Source

Search generates cryptographic proofs of correct execution without exposing proprietary algorithms. Verify integrity using zero-knowledge proofs — no source code required.

Use Cases