Cryptographicdocumentretrieval

The problem PIR solves

Every time you fetch a document from a server, the server learns which document you fetched. That's how your reading history leaks — not through what you typed, but through what you opened.

Private Information Retrieval (PIR) is a cryptographic protocol that lets you fetch a document from a server without the server learning which one. The server processes an encrypted version of your query, returns an encrypted answer, and only your browser can decrypt the result.

How Chuninja does it

Chuninja's PIR Vault is built on two academic results:

• SimplePIR (Henzinger, Laine, Wohlwend, 2022) — a single-server PIR scheme fast enough to be practical. The database is a square matrix over a finite field; the client encrypts a one-hot query vector so the server computes an answer without seeing which row was requested.
• Tiptoe (Henzinger et al., 2023) — a hybrid retrieval pipeline: BM25-style ranking on a small embedding index, then a single PIR query for the chosen document.

The pipeline

1. Pre-index. A curated corpus (Wikipedia, etc.) is embedded with a sentence transformer. Two databases are built: an embedding DB for ranking, and an encoding DB for retrieval.
2. Rank. Your browser sends the query text. The server returns ranked results. (This step is not private — the server sees what you're searching for.)
3. Retrieve privately. Your browser builds a one-hot vector, encrypts it with the public SimplePIR parameters, and sends it to the server. The server returns an encrypted answer. Your browser decrypts it locally and reconstructs the document.

What this protects

If the server's logs are subpoenaed, seized, or breached, an observer sees the ranking queries but not which documents were opened. Combined with the fact that ranking queries are noisy (you don't necessarily open the top result), this raises the bar on traffic analysis significantly.

What this doesn't protect

PIR hides which document was retrieved, not that you made a query. It doesn't anonymize your IP — pair it with a VPN or Tor for that. And it only works on curated indexes (a corpus we've pre-embedded); it can't make an arbitrary live web search private.

Try it

Switch to 🔐 PIR Vault in the search bar and pick a curated collection. The badge on each result tells you whether it was retrieved with PIR.

Open the search →