ZeptoDB is an in-memory columnar database designed for live time-series analytics and AI agent memory. It supports concurrent high-throughput ingestion and real-time analytical queries using lock-free ring buffers, LLVM JIT compilation, SIMD vectorization, and RDMA networking. The system provides HTTP and Python APIs, enabling direct data transfer to frameworks like NumPy and PyTorch without serialization.[01]
- Website
- https://zeptodb.com[01]
- Source Code
- https://github.com/zeptodb/zeptodb[02]
- Tech Docs
- https://zeptodb.com/docs/[03]
- Country of Origin
- KR
- Start Year
- 2026
- Project Types
- Hobby, Open Source
- Written in
- C++
- Operating System
- Linux
- License
- Business Source License
ZeptoDB is an in-memory columnar database designed for live time-series analytics and AI agent memory. It supports concurrent high-throughput ingestion and real-time analytical queries using lock-free ring buffers, LLVM JIT compilation, SIMD vectorization, and RDMA networking. The system provides HTTP and Python APIs, enabling direct data transfer to frameworks like NumPy and PyTorch without serialization.[01]
History
ZeptoDB was developed as an open-source project to address the latency and throughput requirements of AI agents operating on live event streams. It combines a time-series engine with an agent memory layer to store operational telemetry alongside agent-scoped context, embeddings, and prompt cache entries.