Startup Keewano launches agent-focused database and pulls in $12M in funding
SiliconANGLE Paul Gillin
Keewano launched a database built for AI agents, plus $12M in funding. It’s meant to let software reason over live event histories without ETL baggage.
Based on reporting by SiliconANGLE, Paul Gillin — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Keewano, the Tel Aviv startup legally known as Sandstorm Ltd., has launched KeewanoDB and brought in $12 million from Hetz Ventures, Andreessen Horowitz’s a16z Speedrun and other investors. The pitch is simple enough: most databases were designed for people asking structured questions, not for AI agents trying to make sense of long, messy chains of events.
That idea came out of the company’s work on gaming agents, where one user can leave behind a sprawling trail of interactions. KeewanoDB keeps events grouped by entity — a customer, device or transaction — and stores them in the order they happened. The company says that lets an agent inspect history directly, without flattening data into tables or waiting for extract, transform and load jobs. “We’re always running on live data,” Chief Executive Mark Kardashov said. “We don’t have ETLs. We don’t have pipelines. We don’t have lakes.”
The system is built to be lean on infrastructure, at least by AI standards. Keewano says each event takes about four bytes, the data sits on solid-state drives, and the software runs in Docker containers orchestrated by Kubernetes. It uses vectorized instructions on standard CPUs rather than GPUs, and the company claims it can query about 250 million events in under half a second. That figure has not been independently verified.
KeewanoDB supports Parquet, Apache Iceberg and event streams from systems such as Apache Kafka. The company says customers can feed the same events both to KeewanoDB for agent use and to systems like Snowflake for standard business intelligence. Pricing is based on the number of active entities, not individual events. Keewano also says customers using data from Snowflake or Google BigQuery have reported token savings of about 84% versus querying those systems directly.
The other piece of the pitch is that some reasoning can happen inside the database itself. Keewano has an in-database acceleration engine that runs Lua scripts sent by agents through the Model Context Protocol, plus a semantic engine that builds a business layer from incoming events. But the company is careful on one point: the database does not magically tell causation from correlation. Developers still have to define that themselves. That’s the unglamorous part, and also the honest one.
My take — AI-written commentary, not fact-checked reporting
This is the right direction: make the database do less theater and more work. AI vendors love pretending context is a magic cloud feature, when most of the pain is just organizing events so a model doesn’t trip over itself. The sober bit here is the causation warning; that should be printed on the box of every “reasoning” product.
Read more about this at: SiliconANGLE