kronotop/kronotop

Distributed, transactional document database backed by FoundationDB.

What it solves

Kronotop is a distributed, transactional document database designed to provide high isolation and scalability. It specifically addresses the need for AI agents and multi-tenant applications to have their own logical databases (namespaces) within a single cluster, allowing for millions of isolated environments without significant overhead.

How it works

Built on top of FoundationDB, Kronotop uses FoundationDB's directory layer to implement namespaces as key prefixes. It offers two primary data models: "Bucket" for BSON documents with a query language (BQL) and vector search capabilities, and "ZMap" for ordered key-value pairs. It uses the RESP2 and RESP3 protocols, making it compatible with existing Redis-like clients. Document bodies are stored in a segment-based storage engine called Volume, while metadata and ZMap data reside directly in FoundationDB.

Who it’s for

It is intended for developers building AI agent context stores, horizontally scalable transactional document stores, and distributed coordination systems requiring lease-based locks or atomic mutations.

Highlights

  • Massive Multi-tenancy: Supports millions of isolated namespaces in a single cluster.
  • ACID Transactions: Inherits strictly serializable transactions from FoundationDB, supporting atomic commits across multiple namespaces.
  • RESP Compatibility: Works with redis-cli and valkey-cli without requiring new drivers.
  • Vector Search: Integrated approximate nearest neighbor (ANN) search using HNSW and Product Quantization for AI embeddings.
  • Hybrid Storage: Combines a document store (Bucket) with an ordered key-value store (ZMap).