BytePlus VikingDB Vector Database

BytePlus is the enterprise technology arm born from ByteDance's technology. It traces its origins to 2018, hired its first team in Singapore and London in 2019 and officially launched in March 2021, offering ByteDance-derived AI, data and cloud services to businesses outside China.

VikingDB is BytePlus's managed vector database, built on the same vector technology that serves more than 50 ByteDance business lines such as Douyin, Toutiao and Lark. It separates storage from compute and index building from serving, supports search over tens of billions of vectors with inner product, Euclidean and cosine metrics, and combines dense plus sparse vector, scalar and keyword retrieval. Built-in vectorization supports text, images and video with Doubao and open-source embedding models, accessed through a console, OpenAPI and Go, Java, Python and Node.js SDKs. It targets RAG, search, recommendation, deduplication and customer service use cases.

• Hybrid retrieval: dense plus sparse vectors combined with scalar filters and keyword compatibility.
• Multimodal vectorization: built-in embedding for text, images and video with configurable frame rates.
• Proprietary indexes: HNSW, IVF and DiskANN variants with IP, L2 and cosine similarity.
• Decoupled architecture: separate storage, compute, index building and serving on different hardware.
• Model integration: Doubao embedding models plus open-source models such as BGE-M3.
• Developer access: web console, signed OpenAPI and SDKs for Go, Java, Python and Node.js.

• Vietnamese e-commerce and short-video platforms building image and video similarity search and recommendations.
• Banks and telecoms deploying RAG-based customer service assistants.
• Media companies deduplicating and tagging large content libraries.

• Millisecond retrieval across billions of vectors, scaling toward trillions of entries.
• About 3x better performance and cost-effectiveness than open-source engines on HNSW, IVF and DiskANN, and filtering up to 10x faster than open-source Elasticsearch, per the vendor.
• Multi-site disaster recovery at minimal extra cost.
• Hosted in the Johor (Malaysia) region close to Southeast Asian users.

Enterprises across Southeast Asia are moving generative AI from pilots to production chatbots, search and recommendation, which requires scalable vector retrieval close to users. Regional cloud regions such as Johor help Vietnamese and ASEAN companies meet latency and data governance needs while using hyperscale AI infrastructure.

BytePlus Pte Ltd, born from ByteDance technology; first hires in Singapore and London in 2019, official launch in March 2021. VikingDB first released on 1 April 2025 with Johor public beta; target beta markets include Vietnam, Indonesia, Thailand, Malaysia and the Philippines.