Introduction

Kronotop is a distributed multi-model database built on FoundationDB.

It has two data models behind one RESP interface: Bucket, a document model with secondary indexes and vector search, and ZMap, an ordered key-value model. Both live inside namespaces and share the same transaction model. Document bodies are stored on the local disk by Volume, a segment-based storage engine with primary-standby replication.

Data Models

Bucket stores BSON documents and provides a query language (BQL) with comparison, logical, and array operators. Buckets support single-field, compound, and vector indexes. Vector indexes are powered by JVector, use HNSW with automatic Product Quantization, and support cosine, euclidean, and dot_product distance functions. Results are ranked by similarity first, then filtered with BQL predicates (post-filtering).

ZMap is a RESP-compatible proxy over FoundationDB’s ordered key-value API. Keys and values are opaque byte sequences. Keys are stored in lexicographic order. ZMap provides typed numeric operations (int64, float64, decimal128), conflict-free atomic mutations through FoundationDB’s atomic primitives, and range operations over the ordered key space.

Namespaces

Namespaces are lightweight logical databases built on FoundationDB’s directory layer with hierarchical, dot-separated paths. Each namespace has its own keyspace; buckets, indexes, and ZMap keys in one namespace are invisible to another.

Transactions

Transactions are strictly serializable, inherited from FoundationDB. Each command runs in auto-commit mode by default. BEGIN and COMMIT group commands into a single atomic unit. A single transaction can atomically span multiple namespaces. Snapshot reads are available for read-heavy workloads where strict serializability is not required.

Wire Protocol

Kronotop speaks RESP2 and RESP3 and works with existing RESP-compatible clients. kronotop-cli or valkey-cli can connect directly.

Getting Started

The Quickstart starts a minimal cluster with Docker Compose and walks through inserting and querying a document.