Skip to content

Back to the catalog

Mímir: A Stateless Private Search Engine
eprint.iacr.orgpaper

Mímir: A Stateless Private Search Engine

Mímir is a cryptographic protocol for stateless semantic private search over highly dynamic public databases. It uses homomorphic ciphertext switching, relies on (Ring) Learning with Errors, and needs no offline communication. Against Tiptoe it matches search quality with 84x less per-query communication (0.24 MiB). Against PACMANN it raises quality from 76% to 81% of non-private search, with a 9.54 MiB one-time cost instead of 2.68 GiB, but is 2.1x slower.

TL;DR Presents a lattice-based private search protocol that hides text queries from the server with far lower communication than Tiptoe and PACMANN.
Open original ↗
Document

What it is

Mímir is a cryptographic protocol for private semantic search over a public database that can change between queries. The authors are Gökberk Erdoğan, Emma Munisamy and Kamil Kluczniak, and the paper is an ePrint preprint (2026/2605, received 22 October 2026). The authors describe it as the first stateless semantic private search engine for highly dynamic databases. Its security relies on (Ring) Learning with Errors, and it needs no offline communication.

How it works

The abstract names only homomorphic ciphertext switching as the core technique. The design trades higher server computation for low per-query communication, and the authors say the computation is highly parallelizable. The abstract gives no further detail on the construction.

Operating points reported

Quality is measured as a percentage of non-private search quality. All timings below are single multithreaded machine unless noted.

ComparisonResult
Tiptoe, matched quality84x less per-query communication (0.24 MiB)
Tiptoe, matched per-query communication64% quality instead of 36%, and 4.2x faster (0.87 s)
PACMANNQuality 81% instead of 76%, with 72% of its per-query communication
PACMANN, one-time cost9.54 MiB instead of 2.68 GiB
PACMANN, speedMímir is 2.1x slower
Mímir alone59% quality at 8.13 MiB per query and 0.77 s total latency

At 59% quality, Mímir matches PACMANN's latency with 7.6x smaller per-query communication. The authors say it outperforms Tiptoe on all three metrics (quality, communication, latency).

Preconditions / Caveats

The Tiptoe comparison is stated for the case where the database may change between queries. Mímir is slower than PACMANN, so it suits deployments where a 2.68 GiB client download or stored state is the larger cost. The abstract does not give the database size, embedding model or hardware, so check those in the full paper before comparing the numbers to a real workload. The abstract does not mention code availability.

Organisations named here

  • PACMANN
  • Ring Learning with Errors
  • Tiptoe

Techniques named here

  • Learning With Errors

Related in other