Technology
ScramDB is a UTAP database: Unified Transactional Analytical Processing. One engine runs your transactions, your real-time analytics, and your AI and machine-learning work on one live copy of your data.
Most data stacks split those jobs across separate systems: a transactional database for writes, a separate warehouse for analytics, and often a third store for AI and search, all stitched together with pipelines that copy data between them. Every copy adds cost and delay, and gives the numbers a chance to disagree. ScramDB removes the copies. You write your data once and query it every way, on the spot.
This section explains, in plain terms, how ScramDB delivers that.
One engine, one copy
Because transactions and analytics share the same live data, there is no ETL step to build, no nightly batch to wait for, and no "the dashboard is a few hours behind" caveat. An order you just wrote is instantly part of the report you run a second later. One system to run, one copy to secure, one set of numbers everyone trusts.
Native-code speed
ScramDB compiles your SQL down to native machine code, the same kind of instructions a hand-written program runs, instead of interpreting the query one step at a time. This happens automatically in the background, with nothing to configure, and compiled queries are cached so later runs are fast from the start.
Read more in JIT compilation.
Storage that serves writes and scans together
The Tundra storage engine keeps your data in a single columnar copy that is fast to write to and fast to scan. Transactions update it and analytics read it at the same time, without blocking each other. Committed data survives crashes and power loss, every block is checksummed against silent corruption, and colder data is tiered out to low-cost cloud storage automatically.
Read more in Storage engine.
Every core, and the GPU
ScramDB spreads each query across every core in the machine. Work is split into small units and handed to workers pinned to individual CPU cores, so adding cores adds throughput almost in step. For heavy analytical queries, ScramDB can also offload work to the GPU, and it falls back to the CPU automatically when a GPU is not the right fit or not present.
Read more in GPU acceleration.
Consistent across a cluster
Run ScramDB as a distributed cluster and it stays one consistent database, not a scatter of stale copies. Transactions are fully serializable across the whole cluster, so every node agrees on the same up-to-date view of your data. You get the capacity of many machines with the correctness of one.
Analytics and AI in the database
Because your data never leaves ScramDB, analytical and AI work runs right where the data lives. Built-in packages cover geospatial indexing, full-text and relevance search, clustering, regression, anomaly detection, embeddings, approximate aggregates, and more. You can also add your own functions in familiar languages and call them from SQL. No exporting to a separate tool, no round trip.
PostgreSQL-compatible
ScramDB speaks the PostgreSQL wire protocol, so you connect with psql, JDBC, psycopg2, or any PostgreSQL driver or BI tool you already use. It is standard SQL, not a custom query language, so your applications, ORMs, and dashboards work without a rewrite.