SereneDB Review (2026): Pricing, Features & Honest Verdict

TLDR

SereneDB is an open-source database that puts Elasticsearch-style full-text search, vector search and ClickHouse-style analytics behind one Postgres-compatible SQL engine. Verdict: fast, free and ambitious, but single-node and very new. Best for: developers tired of running Postgres plus Elasticsearch. Price: free (Apache 2.0), no managed cloud yet. Rating: 7.0/10.

What is SereneDB?

We found SereneDB on Product Hunt on 22 September 2026, the day it launched Krummelanke, which it calls its first production-ready release. The tagline was “Ultra-Fast Search and Analytics Database, Agentic AI ready.” The pitch is simple: most teams run Postgres for their data, Elasticsearch or OpenSearch for search and something like ClickHouse for analytics, then glue them together with ETL jobs. SereneDB wants to be all three in one engine you query with SQL.

It is built by a Berlin team that raised a $2.1M first funding round in December 2025. The code is on GitHub under Apache 2.0, written in C++, with about 820 stars and 56 forks at the time we looked. Search runs on IResearch, an open-source C++ retrieval library that core members of the team have developed since 2016. The project promises that the complete single-node version “is and always will be” Apache 2.0.

What Are SereneDB’s Key Features?

Search and analytics in one SQL query

This is the core idea. Full-text search in SereneDB is an inverted index sitting beside columnar data, so you filter with the @@ operator, rank with BM25 and aggregate with GROUP BY in the same query. In Elasticsearch that is one request returning two result trees. Here it is one SQL statement, with real joins and full ACID transactions, which Elasticsearch does not offer across documents. Phrase, prefix, wildcard, regex, fuzzy and range queries all have SQL equivalents.

Vector and hybrid search

The same inverted index covers vector search with ANN indexes and geospatial search, and you can combine BM25 and vector scores in one query for hybrid retrieval. SereneDB says a single node can hold billions of vectors. There is a LangChain integration with a vector store, hybrid search and filtering, which makes it a candidate for a RAG retrieval layer without adding a separate vector database.

Postgres wire protocol and clients

SereneDB uses PostgreSQL’s SQL parser and speaks the Postgres wire protocol, version 3.0, so psql, DBeaver, DataGrip, Grafana and the usual drivers for Python, Java, JavaScript, Rust, C# and R connect without code changes. Its docs are frank that it is a different engine built from scratch: some Postgres features behave differently, some are not implemented yet, and the XML type is not planned. There is also SereneUI, a Postgres-compatible UI client for browsing and querying.

Zero-ETL search over files in object storage

You can point SereneDB at Parquet, Iceberg, CSV or JSON files on S3 or HDFS and run BM25 and vector search over them in place, without an ingestion pipeline. Tiered storage keeps hot data local and cold data in object storage, and the team claims more than 10x compression. For log search over a data lake, this is the feature most likely to save real money.

What Does Getting Started With SereneDB Look Like?

Installation is one line. On Linux, a shell script from serenedb.com installs the serened command for your user. On macOS or Windows, you run the official Docker image with a single docker run command that exposes port 7890. Tarballs and Debian packages exist for amd64 and arm64 if you prefer to manage the binary yourself.

From there it feels like Postgres. You connect with psql or DBeaver, create a table, and add an inverted index on the text columns you want to search, choosing an operator class per column to decide how each one is tokenised. Analyzers become text search dictionaries.

Then you write ordinary SQL: a WHERE clause with @@ for matching, an ORDER BY on a BM25 scorer for relevance, and a GROUP BY for counts. If your data already sits in Parquet on S3, you declare a remote table and search it without loading anything. The learning curve is short for anyone who knows SQL, and much shorter than learning the Elasticsearch Query DSL.

How Fast is SereneDB?

SereneDB publishes SearchBench, 92 search and analytics queries over 100 million, 1 billion and 10 billion OpenTelemetry log rows, run against Elasticsearch, OpenSearch, CrateDB, ClickHouse, ArangoDB and Postgres-based search. Its headline claims are 10x faster than ClickHouse on 10 billion logs and indexing 1 billion logs in under eight minutes with about a tenth of the disk. A second benchmark pits IResearch against Lucene and Tantivy on raw retrieval.

These are the vendor’s own numbers, written up by its CTO. The benchmark source and raw results are public on GitHub, which is more than most database launches offer, and you can rerun them. But treat them as a strong starting point, not a verdict. Run your own queries on your own data before you move production.

Where Does SereneDB Fall Short?

It is single-node. Scale-out today works through federation, and “full multi-node compute over an elastic storage layer” is on the public roadmap for next year. There is no managed cloud, so you run it yourself, on Linux for amd64 or arm64, or in Docker. There is no native macOS binary.

Coming from Elasticsearch is a rewrite, not a swap. The migration guide maps Query DSL features to SQL functions and tells you to model each index as a table. An Elasticsearch-style REST layer exists in the test suite for index management, but the documented query path is SQL. The project is also young: the repo was created in August 2025, it shipped four releases in the week of launch, and there are over 300 open issues. We found no published price for commercial support.

How Much Does SereneDB Cost?

SereneDB is free. The full single-node database is Apache 2.0, with no usage limits, no seats and no license key, and you install it with a one-line script on Linux or a single Docker command. Your only cost is the server and the storage it runs on. The site has a contact form for teams that want to talk, but no pricing page, support plan or enterprise tier.

Compare that with the tools it wants to replace. Elasticsearch is free to self-host, with Elastic Cloud as the paid path. ClickHouse Cloud offers a 30-day trial with $300 in credits before usage billing. Meilisearch Cloud starts around $30 a month. The difference is that SereneDB has no paid hosted option at all yet, so free comes with the full operations burden.

Who is SereneDB Best For?

Use SereneDB if you are a developer or a small data team running Postgres beside Elasticsearch or OpenSearch and you are tired of syncing them. It fits log search and observability over large volumes, product search that needs SQL joins, and RAG retrieval that wants BM25 and vectors in one place. It is a good fit if you are comfortable running a database yourself on Linux. Teams building AI agents get one more reason to look: an agent that can write SQL can search, filter and aggregate in a single tool call, instead of juggling a search API and a database connection.

Skip SereneDB if you need a managed service, multi-node high availability today, or a drop-in replacement for an existing Elasticsearch Query DSL codebase. Skip it too if you rely on specific Postgres extensions: it speaks the protocol but is not Postgres underneath.

Best SereneDB Alternatives

Elasticsearch and OpenSearch

The incumbents for full-text search, with mature clustering, huge ecosystems and managed options. Both are free to self-host. They win on scale and tooling today, and lose on joins, transactions and running a second system beside your database.

ClickHouse

The default choice for fast analytics over billions of rows, open source with a managed cloud that starts with a 30-day trial and $300 in credits. It can do text search, but relevance ranking is not its first job. Pick it if dashboards and aggregates matter more than search quality.

Meilisearch

A developer-friendly search engine for product and site search, free to self-host with a cloud plan from about $30 a month. It is far simpler to run for app search but does not try to be an analytics database. Pick it for search boxes, not log lakes.

Final Verdict: Is SereneDB Worth It?

SereneDB is one of the more interesting database launches this year. Putting BM25 search, vector search and columnar analytics behind one Postgres-compatible SQL engine, and letting you search Parquet on S3 in place, solves a problem a lot of teams pay for twice. It is free under Apache 2.0, and its benchmarks are public and reproducible.

It is also a first production release from a small team, single-node for now, with no managed cloud and a fast release cadence. Try it on a side workload, such as log search or a RAG index, and rerun its benchmarks on your own data. If multi-node lands as promised, the score goes up. Today it earns 7.0.