Keewano launches KeewanoDB, raises $12 million

Keewano's launch comes from its founders' experience developing agents for the gaming industry, where a single user generates a large and complicated history of interactions. They concluded that the tabular architecture underlying conventional relational databases and data warehouses was not built for AI agents doing analytics and reasoning, according to co-founder and Chief Executive Mark Kardashov.

KeewanoDB stores events grouped by entity and in their original order, avoiding traditional extract, transform and load pipelines. It supports Parquet and Apache Iceberg, can ingest event streams through systems like Apache Kafka, and can run alongside existing analytical databases such as Snowflake. The company charges by the number of active entities rather than events.

The database does not distinguish causation from correlation on its own; developers must specify those parameters. Keewano also offers a separate analytics layer with an agent called Signal that applies machine learning and statistical methods to test hypotheses and present event sequences as evidence. Whether KeewanoDB's stated performance and savings hold up under independent testing — the company says its performance figures have not been independently verified — will likely determine how quickly agent builders adopt it.