About CyborgDB

CyborgDB is a Confidential Vector Database that enables you to search and analyze vectors in a privacy-preserving manner, without revealing the contents of the vectors themselves.

​

Why Confidential?

According to KPMG, 63% of enterprises say that confidentiality and data privacy are their top risk to AI adoption (1). In regulated sectors, this figure increases to 76%. Yet, when it comes to solutions which address these concerns, the market has yet to answer. This leaves a critical mass of companies unserved and unable to adopt AI in their workflows. They need Confidential AI.

​

Why Vector?

Vector Search is at the heart of the most popular AI applications - RAG, RecSys, Semantic Search, IR, etc. - and the market is overcrowded with Vector DBs. All of these share a key commonality: they need to store their indexed contents (vector embeddings) in plaintext to enable Approximate Nearest-Neighbor (ANN) search. This creates a significant attack vector for that data, making confidentiality a near-impossibility.

To solve this, we need a brand new approach to Vector Search - Confidential Vector Search. CyborgDB is the first solution to implement this. By leveraging cryptographic hashing and symmetric encryption, CyborgDB enables ANN search over encrypted space. This means that vector embeddings are not decrypted during search, and remain encrypted throughout their lifecycle. This greatly reduces the attack surface while guaranteeing the confidentiality of inference data.

​

How It Works

CyborgDB is built on a novel architecture that keeps confidential inference data secure through a zero-trust design. It leverages cryptographic hashing and forward-privacy to maintain vector embeddings end-to-end encrypted throughout their lifecycle, including at search time.

This makes it easy to integrate with existing AI workflows, providing high-performance indexing and retrieval that can be GPU-accelerated with CUDA. CyborgDB works with databases such as Postgres and Redis, transforming traditional databases into Confidential Vector Databases.

​

Backing Stores

CyborgDB supports a variety of backing stores, including:

• PostgreSQL
• Redis
• Memory (for testing and development)

In addition, CyborgDB supports managed services such as:

• AWS RDS
• AWS ElastiCache
• Azure Database for PostgreSQL
• Azure Cache for Redis
• Google Cloud SQL
• Google Cloud Memorystore

We are continuously adding support for more backing stores. If you have a specific backing store in mind, please let us know.

​

Deployment Models

CyborgDB supports a variety of deployment models, including:

1. Core Library

In this form, CyborgDB is packaged as a standalone module (Python or C++) which contains all of the encrypted indexing and querying logic. The module is run locally and can connect to any of the supported backing stores.

Recommended for evaluation, small-scale deployments, and custom integrations.

2. Microservice

Coming June 2025

The CyborgDB microservice is a self-contained Docker image that can be deployed locally or in a Kubernetes cluster. It enables scalable and repeatable deployments, and also connects to any of the supported backing stores. REST API as well as client SDKs (Python, JS/TS, C++ and Go) will be available.

Recommended for development and production deployments.

3. Microservice w/ DB

Coming Q3 2025

This evolved microservice integrates a backing store within the Docker image, enabling a self-sufficient Confidential Vector DB deployment without external backing stores. REST API as well as client SDKs (Python, JS/TS, C++ and Go) will be available.

Recommended for development and production deployments.

4. Serverless

Release Date TBD

The fully-managed CyborgDB offering will provide the lowest barrier to adoption - simply generate an API key and install a client SDK.

Recommended for everything from evaluation to large-scale deployments.

​

CyborgDB Lite

CyborgDB is a proprietary product, but we also offer a free version of the core library called cyborgdb_lite. This version is intended for evaluation and non-commercial use only, and comes with the following functional limitations: