SereneDB

Search and analytics usually live in two systems: a search cluster on one side, an analytical warehouse on the other, and a synchronisation job between them. SereneDB, a database from a Berlin company founded in 2025, is built to collapse that pair into one engine.

One engine for text, vectors and columns

SereneDB combines full-text, fuzzy, geospatial, vector and hybrid search with SQL execution over columnar storage. It presents a PostgreSQL-compatible front end, so existing clients and tools can connect through the PostgreSQL protocol. Analytical queries such as joins, unions and large aggregations run on the Velox execution engine, while durable updates are backed by RocksDB. Its storage mixes search indexes, columnar data and remote tables.

Underneath sits IResearch, a retrieval library written in C++ that people on the SereneDB team have worked on since 2016. The company says the engine outperformed two well-known search libraries in a public benchmark it ran itself, which is a vendor reading of a third-party test and worth reproducing.

Krummelanke and the open single-node build

The first production-ready release, named Krummelanke, was published in September 2026. The complete single-node version is open under the Apache 2.0 licence and published to a public code repository host; the company commits to keeping that build open. Installation is a one-line shell command, with downloadable builds and a getting-started guide as alternatives. For AI agents, the site also offers a markdown homepage, a sitemap and an llms.txt file.

What to weigh before adopting it

The pitch is consolidation: one backup routine, one access-control model and one upgrade path instead of two. That also means one system to be wrong about. The company does not publish a price list for anything beyond the open single-node build, so a team planning a cluster has to ask. The about page describes the team as having a background in databases, distributed systems, search engines and large-scale analytics.

SereneDB is a real-time search analytics database from Berlin. Alexander Malandin, Andrey Abramov and Valery Mironov founded the company in 2025. Its first production release, Krummelanke, was published on 22 September 2026.

What SereneDB does

• Search: full-text, fuzzy, geospatial, vector and hybrid queries over analytical data.
• SQL analytics: joins, unions and aggregations over columnar storage, powered by Velox.
• Storage: indexes for search, column-oriented tables, remote tables, and durable updates kept in RocksDB.
• PostgreSQL protocol: connects from familiar Postgres tools and clients.

Clients

Teams that run a search cluster beside an analytical warehouse and want one engine for both.

Key facts

• Founded: 2025.
• Licence: the single-node build is under Apache 2.0 and published on a public code host, and the company says that build always will be.
• Search core: IResearch, a C++ information-retrieval library the team has worked on since 2016.
• Benchmarks: the company says it outperformed two established search libraries on one public benchmark, which is the vendor's own reading.
• Larger deployments: a quote from the company is needed.