Database startup Keewano today launched KeewanoDB, an event-oriented database designed to give artificial intelligence agents real-time context for analytics and decision-making.
The startup also announced $12 million in funding from Hetz Ventures, Andreessen Horowitz LLC’s a16z Speedrun and other investors.
The Tel Aviv-based company, whose formal name is Sandstorm Ltd., argues that conventional relational databases and data warehouses were built for structured queries by people, not for artificial intelligence agents that must examine long sequences of behavior and determine why an outcome occurred. Keewano’s founders encountered that problem while developing agents for the gaming industry, where a single user can generate a large and complicated history of interactions.
“We realized that the current tabular architecture wasn’t built for AI agents or machines, when it comes to analytics specifically, and reasoning in general,” said co-founder and Chief Executive Mark Kardashov (pictured, second from left, with Chief Product Officer Vitaly Bukhovsky, Chief Technology Officer Pavel Bibergal and Chief Operating Officer Dima Kargersaid).
KeewanoDB groups events around an entity such as a customer, device or transaction and stores them in the order they occurred. That arrangement lets an agent inspect an entity’s history without first flattening the information into tables or waiting for a new extract, transform and load pipeline.
“We’re always running on live data,” Kardashov said. “We don’t have ETLs. We don’t have pipelines. We don’t have lakes.”
The company says each event occupies about four bytes and is organized on solid-state drives for efficient reading. A distributed, multi-sharded design runs in Docker containers orchestrated by Kubernetes and uses vectorized instructions on standard central processing units instead of graphics processing units. Keewano claims the system can query about 250 million events in less than half a second, though those performance figures have not been independently verified.
Keewano supports Parquet and Apache Iceberg and can ingest event streams sent through systems such as Apache Kafka. Customers can send the same events to KeewanoDB for agents and to analytical databases like Snowflake for conventional business intelligence.
“We can live alongside your existing database,” Kardashov said.
Keewano charges according to the number of active entities rather than individual events. Kardashov said customers ingesting information from Snowflake Inc.’s data cloud or Google LLC’s BigQuery have reported token savings of about 84% compared with querying their existing systems directly.
Those savings partly come from an in-database acceleration engine that executes Lua scripts sent by agents through the Model Context Protocol. Lua is a lightweight, fast scripting language designed primarily to be embedded into other larger programs written in C or C++. Processing and filtering occur where the data resides, so the model receives a smaller window of relevant context instead of the complete event history.
“Part of the reasoning happens inside the database and not by the engine,” Kardashov said. “So the database becomes an extension of an agent.”
Keewano also provides a semantic engine that automatically constructs a business layer from incoming events, although customers can add mappings and business context manually. The company says examining surrounding events helps the software interpret ambiguous actions and preserve their sequence.
Kardashov acknowledged that the database alone does not distinguish causation from correlation; developers building agents on it must specify those parameters themselves. Keewano’s separate analytics layer includes an agent called Signal that applies machine learning and statistical methods, tests its hypotheses and presents supporting event sequences as evidence.
The company is launching two products: the general-purpose KeewanoDB for agentic analytics and reasoning and an analytics layer for product analytics and real-time operational decisions.
