CyborgDB
 | |
| Developer(s) | Cyborg Inc.[1] |
|---|---|
| Initial release | October 31, 2024 [2] |
| Stable release | 0.12.0 [2]
/ August 28, 2025 |
| Written in | C++, CUDA, Python |
| Engine | |
| Type | |
| License | Open-source / proprietary components |
| Website | www |
Search CyborgDB on Amazon.
CyborgDB is an encrypted vector database proxy developed by Cyborg Inc., a technology company based in New York City. The product provides encryption for vector embeddings used in artificial intelligence (AI) applications, particularly retrieval-augmented generation (RAG) systems.[3] CyborgDB functions as a proxy layer for existing databases, enabling similarity search on encrypted embeddings while maintaining performance with GPU acceleration.[4]
History
Cyborg Inc. was founded in 2017 by Nicolas Dupont, initially developing data compression technology before pivoting to security infrastructure.[5][6]
The company introduced CyborgDB in 2024 as a solution for encrypting vector embeddings in enterprise AI systems.[3] In May 2025, Cyborg received investment from iGan Partners to expand development of the platform.[7]
Cyborg’s work on confidential vector search has also been featured at technical conferences, including the OpenSSL Conference 2025[8] and the Confidential Computing Summit 2025.[9]
Technology
CyborgDB functions as a proxy layer between applications and existing databases such as PostgreSQL or Redis, adding encryption capabilities for vector embeddings.[3] The system encrypts vectors client-side before storage, preventing raw embeddings from being exposed at rest or in transit.
All sensitive data in CyborgDB (vector embeddings, document identifiers, document contents, and metadata) is encrypted end-to-end using the AES-256 algorithm in GCM mode, with per-record initialization vectors to prevent ciphertext reuse.[10] The design provides authenticated encryption, guaranteeing both confidentiality and tamper detection.
In addition to encrypting stored data, CyborgDB protects the **search index** used for approximate nearest neighbor (ANN) retrieval. Its patented scheme implements forward-secure cryptographic indexing, meaning that new insertions cannot be linked to past queries even if the index is later compromised.[10] Indexing relies on key derivation with HMAC and SHA-3, producing per-cluster seeds and binary-tree node keys. Search queries generate minimal, token-scoped keys that allow only the necessary portions of the encrypted index to be accessed, leaving irrelevant nodes opaque.[10]
The system executes similarity search directly over ciphertext nodes, a technique related to searchable encryption. Optional client-side decryption ensures that plaintext embeddings never need to reside on the server.
According to NVIDIA, GPU acceleration through NVIDIA cuVS achieves significant performance improvements, including "47× faster index building" and "9.8× faster batch retrieval" compared with CPU-only methods.[3] Cyborg reports that its optimizations, including AES-NI hardware acceleration, parallel key derivation, and lazy decryption, deliver query latencies under 10 milliseconds and throughput of over 10,000 queries per second on encrypted ANN search workloads.[10]
SDKs and deployment packages
CyborgDB is distributed both as client SDKs for application integration and as packaged services for deployment.
| Category | Platform / Language | Distribution | Repository |
|---|---|---|---|
| SDK | Python | cyborgdb (PyPI) | GitHub |
| SDK | JavaScript / TypeScript | cyborgdb (npm) | GitHub |
| SDK | Go | cyborgdb-go (pkg.go.dev) | GitHub |
| Service | Python service | cyborgdb-service (PyPI) | GitHub |
| Service | Docker image | CyborgDB Service (Docker Hub) | GitHub |
Reception
Coverage of CyborgDB has highlighted its role in bringing encryption to vector search. NVIDIA featured the system in its developer blog, citing GPU-accelerated performance improvements using cuVS.[3] Industry outlet theCUBE has also profiled Cyborg as part of the confidential AI market, with CEO Nicolas Dupont discussing enterprise adoption and security requirements.[5][6] Cyborg is also listed by VentureRadar as a notable company in database security and confidential computing.[11]
See also
- Vector database
- Retrieval-augmented generation
- Homomorphic encryption
- Confidential computing
- Database security
References
- ↑ "Cyborg Inc". Cyborg Inc. 2025.
- ↑ 2.0 2.1 "CyborgDB Changelog". Cyborg Documentation. 2025. Retrieved 3 October 2025.
- ↑ 3.0 3.1 3.2 3.3 3.4 Nolet, Corey; Dupont, Nicolas (July 2024). "Bringing Confidentiality to Vector Search with Cyborg and NVIDIA cuVS". NVIDIA Developer Blog. NVIDIA Corporation. Retrieved 3 October 2025.
- ↑ "CyborgDB Backing Stores". Cyborg Documentation. 2025. Retrieved 3 October 2025.
- ↑ 5.0 5.1 "Cyborg CEO Nicolas Dupont on Securing Enterprise AI". theCUBE. SiliconANGLE Media. May 2025. Retrieved 3 October 2025.
- ↑ 6.0 6.1 "Cyborg CEO Nicolas Dupont at NYSE Wired MOE Series". theCUBE. SiliconANGLE Media. September 2025. Retrieved 3 October 2025.
- ↑ "Cyborg Receives Investment from iGan Partners". iGan Partners. May 2025. Retrieved 3 October 2025.
- ↑ ["https://openssl-conference.org/speaker-sessions/detail-164_1292983#sectionLink" "OpenSSL Conference 2025: Confidential Vector Search"] Check
|url=value (help). OpenSSL Conference. 2025. Retrieved 3 October 2025. - ↑ "Confidential Computing Summit 2025: Confidential AI's Missing Piece: End-to-End Encrypted Vector Databases". YouTube. 2025. Retrieved 3 October 2025.
- ↑ 10.0 10.1 10.2 10.3 "CyborgDB Encryption Documentation". Cyborg Documentation. 2025. Retrieved 3 October 2025.
- ↑ "Cyborg on VentureRadar". VentureRadar. Retrieved 3 October 2025.
External links
- Official website
- Official documentation
- CyborgDB on GitHub
- CyborgDB on Docker Hub
- CyborgDB on PyPI
- CyborgDB on npm
- CyborgDB Go package
This article "CyborgDB" is from Wikipedia. The list of its authors can be seen in its historical and/or the page Edithistory:CyborgDB. Articles copied from Draft Namespace on Wikipedia could be seen on the Draft Namespace of Wikipedia and not main one.
