ScramDB
A brand-new, closed-source database that speaks the PostgreSQL protocol and promises transactions, analytics and vector search on one copy of the data; first builds shipped in summer 2026.
- Type
- Commercial product
- Made by
- ScramDB Labs
- License
- Proprietary, closed source; free single-node Community edition capped at 64 GiB, Enterprise licence key required for clustering; no published licence agreement found
- Facts checked
- 6 October 2026
What it is
ScramDB ships as one binary for Linux (x86-64 with AVX2, or arm64), a Docker image and a Helm chart. A node listens on port 5432 for the PostgreSQL wire protocol, serves Prometheus metrics and a health check on 9090, and runs an MCP server for AI agents on 9191. Its storage engine, Tundra, keeps one columnar copy of each table: writes land in a write-optimized area and are visible immediately, then get folded into compressed column blocks with per-block min/max summaries so scans can skip data. A write-ahead log handles crash recovery, and MVCC gives isolation up to serializable. Queries are compiled to native code in the background and the compiled code is cached on disk; heavy queries over at least 50,000 rows can be offloaded to an NVIDIA, AMD or Apple GPU, falling back to the CPU. Functions run as sandboxed packages, and the install includes packages for BM25 text scoring, embeddings, H3 cells, HyperLogLog and other sketches. CREATE DATABASE ... CLONE makes an instant copy-on-write branch, and with WAL archiving to S3, GCS or Azure a branch can start from a past timestamp.
It is very new. Its author, Theo M. Bulut, announced the first version on 23 July 2026. The installer pulls automated prerelease builds named after git commits from a GitHub releases repository (the latest from 30 September 2026); Docker Hub has a 1.0.0 image tag from 18 September, but the installer’s index lists no stable version yet. The source is not public, and the site has no terms or licence page. The Community edition is free on a single node and turns read-only past 64 GiB of data. Enterprise, priced on request, lifts the cap and unlocks clustering: a node with a [cluster] section in its config refuses to start without an Enterprise or Trial key, and a key only covers binaries built before its end date. There is no managed service.
Why teams pick it
- Quick to try: a one-line installer or
docker run, then any PostgreSQL driver,psqlor BI tool connects - Branching and point-in-time recovery are built in: copy-on-write
CLONE, branches from a past timestamp, and WAL archiving switched on in the Docker image - Search, vector, geospatial and sketch functions come as bundled sandboxed packages, with no C extensions to compile
- Unusually candid docs: a limitations page lists what doesn’t work yet with workarounds, and the licensing page spells out exactly what is enforced
Watch out for
- Pre-release in practice, from a tiny vendor: first announced in July 2026, the installer still serves untagged prerelease builds, and every blog post is by its author; there are no published customers or independent reviews
- Closed source with no published licence terms; clustering, the core of the “distributed hyperscale” pitch, needs an Enterprise key at an unpublished price, and Community stops writes at 64 GiB
- Performance claims are self-reported: its benchmark page sets ScramDB’s own single-node run against other systems’ published results, and ClickBench’s maintainer removed the “UTAP” and “JIT” tags from its entry, calling them made up
- Gaps listed in its own docs: backups are cold (the server must be stopped), single-row write latency is still being tuned, and there are no range types, XML, locale collations, function overloading or incremental materialized view refresh
- The Docker image starts with authentication disabled on all interfaces, and the site contradicts itself in places, for example calling MCP “on the roadmap” while the docs describe an MCP server
Alternatives
What all relational databases have in common →
Alibaba Cloud PolarDBCommercialAlibaba Cloud’s cloud-native relational databases: MySQL- and PostgreSQL-compatible clusters whose compute nodes share one copy of the data on distributed storage, plus PolarDB-X, a distributed MySQL-compatible edition.
CedarDBCommercialA PostgreSQL-compatible database rebuilt from scratch at TU Munich, running transactions and analytics on one copy of the data on a single server.
Cloudflare D1CommercialCloudflare’s serverless SQL database that speaks SQLite, runs each database inside a Durable Object, and bills by rows read and written rather than by the hour.
CloudNativePGOpen sourceA Kubernetes operator that runs highly available PostgreSQL with no Patroni and no StatefulSets: the Kubernetes API itself holds the cluster’s state.
CockroachDBCommercialA distributed SQL database that speaks the PostgreSQL wire protocol, replicates every slice of data with Raft and keeps transactions serializable across nodes and regions.
MySQLOpen sourceThe long-standing open-source SQL database behind much of the web, built on the InnoDB engine.
NeonCommercialServerless Postgres that keeps data in a shared storage layer, so compute can scale to zero and a branch of the database is a pointer, not a copy.
PatroniOpen sourceA Python agent beside each Postgres server that elects one primary through etcd, Consul, ZooKeeper or Kubernetes, and fails over when it dies.
Percona OperatorsOpen sourceFour free Kubernetes operators from Percona that run its builds of MySQL, PostgreSQL and MongoDB, each using the engine’s own replication and backup tools.
PGO (Crunchy Postgres for Kubernetes)Open sourceCrunchy Data’s Postgres operator: Patroni in every pod for failover, pgBackRest for backups, and a declarative `PostgresCluster` resource on top.
PlanetScaleCommercialManaged MySQL on Vitess and managed Postgres, with schema changes shipped through branches and deploy requests.
PostgreSQLOpen sourceThe open-source relational database that most new projects start on, extensible enough to grow into many niches.
SQLiteOpen sourceA complete SQL database in a library and a single file, with no server to run.
TursoOpen sourceA from-scratch Rust rewrite of SQLite with concurrent writes, plus libSQL (an open-contribution SQLite fork) and Turso Cloud, which hosts many small databases on S3.
VitessOpen sourceA proxy layer, born at YouTube, that shards MySQL across many servers while the app still sees one database.How it fits with other technologies
Run it in Docker, query it, branch it · sql
-- docker run -d --name scramdb -p 5432:5432 -v scramdb-data:/var/lib/scramdb scramdb/scramdb:latest
-- (development only: the image's default command disables authentication)
-- psql -h localhost -p 5432 -U scramdb scramdb
CREATE TABLE sensors (
id BIGINT PRIMARY KEY,
device_id INTEGER NOT NULL,
temperature DOUBLE PRECISION,
recorded_at TIMESTAMP
);
INSERT INTO sensors VALUES
(1, 101, 22.5, '2026-10-06 10:00:00'),
(2, 101, 23.1, '2026-10-06 10:05:00'),
(3, 102, 19.8, '2026-10-06 10:00:00');
SELECT device_id, avg(temperature) AS avg_temp, count(*) AS readings
FROM sensors
GROUP BY device_id
ORDER BY avg_temp DESC;
-- an instant copy-on-write branch to test a migration against
CREATE DATABASE sensors_test CLONE scramdb;What ships and what is still a claim#
ScramDB’s site describes far more than a product this new usually has. This is what its own docs, releases and outside sources showed on the date checked.
| Checked 6 October 2026 | Status |
|---|---|
| PostgreSQL protocol, MVCC, serializable transactions | Ships and is documented; compatibility is self-tested |
| Single-node install | Ships for Linux x86-64 and arm64, Docker and Helm; no native macOS build |
| Stable release | Docker tag 1.0.0 (18 September 2026); the installer offers only prerelease builds |
| Multi-node and multi-region clusters | Documented, but needs an Enterprise or Trial key; no independent reports found |
| Backups and recovery | Cold base backups, WAL archiving, point-in-time recovery and branching; no online backup |
| GPU execution | Documented for NVIDIA, AMD and Apple Metal; GPU metrics not exposed yet |
| MCP server for agents | Documented on port 9191; the home page still calls MCP “on the roadmap” |
| Speed | Self-run numbers; ClickBench results submitted in September 2026 |
See it in a real stack
No architecture uses ScramDB yet. Drop it into the playground and wire it up yourself.Open playground →Benchmarks
How to read a benchmark →Where it came from2018–2026
ScramDB is the work of Theo M. Bulut. In his launch post he writes that he started eight years earlier, when analytical systems were still built around traditional execution models, and set out to build one system that would make Redis, PostgreSQL, CDC tools, Kafka, vector databases and search engines unnecessary for most teams. The project went through what he calls multiple burnouts on the way.
The ScramDB Labs GitHub organisation, which gives Germany as its location, dates from 2020 and holds earlier open-source pieces from the same work: Callysto, a stream processing framework (2021, Apache 2.0), and a white paper on ANODE, an on-disk approximate nearest-neighbour index (2024). The database itself stayed closed. A Docker Hub repository appeared in March 2026, and on 23 July 2026 Bulut announced the first version.
Automated builds have been published to GitHub since August 2026, ClickBench results were submitted in September, and a 1.0.0 image tag followed on Docker Hub. It is too early to say what, if anything, it has changed.
Why it’s built this way
- One columnar copy with a write-optimized area
- Running a row store for transactions and a column store for analytics means a pipeline between them and stale dashboards. Tundra takes writes into a fast area and folds them into compressed columns in the background, so both workloads read the same data.
- The PostgreSQL wire protocol instead of a new API
- So existing drivers, ORMs and BI tools connect as they are, with no ScramDB SDK to learn.
- Sealed, capability-declared function packages
- PostgreSQL loads untrusted C extensions into the server, where one bug can crash it. ScramDB functions run in a sandbox and get only the access their package declares.
- A free single node, a licence key for clustering
- Anyone can run the full engine on one machine, while running many nodes, which is what the company sells, needs an Enterprise licence enforced at startup.
Timeline
- 2018
Theo M. Bulut starts work on the engine, by his own account
- 2021
Callysto stream processing framework open-sourced under Apache 2.0
- 2024
ANODE white paper on on-disk vector search published
- 2026
First version announced on 23 July; prerelease builds from August
- 2026
ClickBench results submitted and a
1.0.0Docker tag in September
Sources18 · facts checked 6 October 2026
- ScramDB home page (claims, UTAP, MCP “on the roadmap”)
- ScramDB: Get started (Community and Enterprise editions)
- ScramDB docs: Cluster licensing (Enterprise gate, 64 GiB cap, build-date check)
- ScramDB docs: Quick start (Docker, installer, platforms, ports)
- ScramDB docs: Docker deployment (authentication disabled in the image)
- ScramDB docs: Kubernetes (Helm chart)
- ScramDB docs: Storage engine (Tundra, block summaries, WAL, MVCC)
- ScramDB docs: GPU acceleration
- ScramDB docs: Semantic AI (MCP server)
- ScramDB docs: Branching
- ScramDB docs: Backups (cold base backups, object storage destinations)
- ScramDB docs: Storage tiering (WAL archival to S3, GCS, Azure)
- ScramDB docs: PostgreSQL compatibility
- ScramDB docs: Current limitations
- ScramDB: Benchmarks (self-run, compared with published results)
- GitHub: scramdblabs/releases (installer index, prerelease builds)
- GitHub: ClickBench commit removing made-up UTAP and JIT tags (22 September 2026)
- Docker Hub: scramdb/scramdb tags (1.0.0 on 18 September 2026)