Inferensys

Glossary

Ferret OT

A state-of-the-art OT extension protocol that achieves extremely high speed by leveraging vector oblivious linear evaluation (VOLE) and quasi-cyclic codes for correlation-robust hashing.
Engineer reviewing vector database search results on laptop, embeddings visualization on screen, home office coding session.
HIGH-SPEED OT EXTENSION

What is Ferret OT?

Ferret OT is a state-of-the-art oblivious transfer extension protocol that achieves extremely high speed by leveraging vector oblivious linear evaluation (VOLE) and quasi-cyclic codes for correlation-robust hashing.

Ferret OT is a high-performance oblivious transfer (OT) extension protocol that generates a massive number of OTs from a small base set using only fast symmetric-key operations. It uniquely combines vector oblivious linear evaluation (VOLE) with quasi-cyclic codes to dramatically reduce communication overhead and computational latency compared to earlier protocols like IKNP.

By employing correlation-robust hashing based on quasi-cyclic codes, Ferret OT minimizes the size of the correlated randomness required for OT extension. This design makes it a foundational building block for modern private set intersection (PSI) protocols, particularly those requiring malicious security, enabling practical deployment in bandwidth-constrained collaborative analytics.

HIGH-SPEED OT EXTENSION

Key Features of Ferret OT

Ferret OT is a state-of-the-art oblivious transfer extension protocol that achieves unprecedented throughput by combining vector oblivious linear evaluation (VOLE) with quasi-cyclic codes for correlation-robust hashing.

01

VOLE-Based Foundation

Ferret OT leverages Vector Oblivious Linear Evaluation (VOLE) as its core primitive, replacing traditional OT extension matrix operations with a more efficient algebraic structure. This shift dramatically reduces communication overhead by generating long vectors of correlated randomness from a small, constant-size seed. The protocol uses quasi-cyclic codes to instantiate the VOLE, enabling linear-time encoding and decoding operations that scale efficiently with batch size.

10M+
OTs per second
02

Correlation-Robust Hashing

The protocol employs a specialized correlation-robust hash function to compress the VOLE outputs into standard oblivious transfer messages. This cryptographic construction ensures that the hash outputs remain pseudorandom even when correlated inputs are hashed under the same key, a critical property for maintaining security in the random oracle model. Ferret OT's hashing approach eliminates the need for expensive public-key operations during the extension phase, relying entirely on fast symmetric-key primitives.

03

Quasi-Cyclic Code Construction

Ferret OT instantiates its VOLE using quasi-cyclic low-density parity-check (QC-LDPC) codes, which provide a highly efficient linear encoding structure. These codes are defined by circulant matrices that enable fast polynomial multiplication operations. The quasi-cyclic structure allows the protocol to achieve:

04

Malicious Security with Low Overhead

Ferret OT achieves malicious security—protecting against adversaries that arbitrarily deviate from the protocol—with minimal performance penalty compared to semi-honest variants. The protocol incorporates efficient consistency checks that verify the correctness of the VOLE correlations without revealing the underlying values. This is accomplished through batch verification techniques that amortize the cost of checking across millions of OTs, making it practical for real-world deployments where strong adversarial guarantees are required.

< 2x
Overhead vs semi-honest
05

Silent OT Extension Mode

Ferret OT supports a silent OT extension mode where the receiver can non-interactively derive its OT outputs after a one-time setup phase. This is particularly valuable for unbalanced PSI and contact discovery applications where a client with a small set queries a server with a massive database. The silent mode eliminates the need for the server to remain online during the extension phase, reducing latency and enabling pre-computation of OT instances before the receiver's inputs are known.

FERROT OT PROTOCOL

Frequently Asked Questions

Common questions about the Ferret OT extension protocol, its performance characteristics, and its role in high-speed private set intersection.

Ferret OT is a state-of-the-art oblivious transfer extension protocol that achieves extremely high throughput by leveraging vector oblivious linear evaluation (VOLE) and quasi-cyclic codes for correlation-robust hashing. Unlike traditional OT extension protocols based on the IKNP framework, Ferret generates a large number of OTs from a small number of base OTs using VOLE as an intermediate primitive. The protocol operates by first generating a long VOLE correlation between the sender and receiver using fast symmetric-key operations and quasi-cyclic low-density parity-check codes. This VOLE correlation is then converted into a large batch of oblivious transfers through a lightweight hashing step. The use of quasi-cyclic codes allows Ferret to achieve amortized communication costs that are sub-linear in the number of OTs, making it one of the fastest OT extension protocols available for high-volume private set intersection applications.

PERFORMANCE COMPARISON

Ferret OT vs. Other OT Extension Protocols

A technical comparison of Ferret OT against the foundational IKNP protocol and the state-of-the-art SoftSpokenOT, highlighting key differences in cryptographic assumptions, computational cost, and communication complexity.

FeatureFerret OTSoftSpokenOTIKNP Protocol

Core Primitive

Vector OLE (VOLE)

Vector OLE (VOLE)

Symmetric-Key OT

Code Structure

Quasi-Cyclic Codes

Pseudorandom Correlation Generators

Random Oracle + Matrix Transposition

Base OTs Required

~128

~128

~128

Correlation Robustness

Yes (Quasi-Cyclic)

Yes (PCG-based)

Yes (Random Oracle)

Computation Type

Fast Bitwise XOR + Linear Algebra

Fast Bitwise XOR + PCG Expansion

Heavy Symmetric-Key Operations

Communication per OT (amortized)

< 1 bit

~1 bit

~1 bit

Malicious Security Support

Relative Speed (LAN)

10-30x faster than IKNP

Comparable to Ferret

Baseline (1x)

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.