Keewano Launches KeewanoDB with $12M to Power AI Agents with Real-Time Event Data

KeewanoDB Launches as Purpose-Built Database for AI Agents

Keewano made KeewanoDB generally available on Tuesday, introducing what the Tel Aviv-based startup calls an event-series database designed specifically for AI agents. The company, formally known as Sandstorm Ltd., announced the launch alongside $12 million in seed funding led by Hetz Ventures, with participation from Andreessen Horowitz's a16z Speedrun and other investors.1

1

2

The agent-focused database addresses a fundamental problem: traditional databases flatten events into rows and precomputed aggregates, discarding the sequence and context AI agents need to reason about why something happened. This forces agents to reconstruct history from a warehouse or search index at query time, creating bottlenecks in agentic analytics workflows.1

1

How KeewanoDB Differs from Traditional Time-Series Databases

Co-founder and CEO Mark Kardashov pushes back on categorizing KeewanoDB as a graph database, document database, or straightforward time series system. Instead, he describes it as an event-series database that treats each entity—a user, device, or agent—as a primary key, with every event attaching to it in sequence.1

1

The architecture creates one large table per entity rather than requiring joins across many tables. Events are stored as four-byte units on disk, sequenced by entity rather than split across relational tables. This approach lets agents inspect an entity's complete history without flattening information into tables or waiting for extract, transform and load pipelines.1

1

2

Real-Time Context Through Native Machine Learning Integration

KeewanoDB runs as a fully managed service on Google Cloud, with other clouds and self-managed deployment planned for early 2027. An agent connects directly to KeewanoDB and reads sequences where they already sit, wired through an SDK and the Model Context Protocol (MCP).1

1

The system includes a semantic layer built during ingestion using large language models that attaches additional context to each event as it lands. If a user transitions from active to inactive to flagged for fraud, all three labels persist in sequence—agents see current status without losing earlier context. An acceleration engine enables in-database agent queries by executing Lua scripting directly through MCP, eliminating the need to move data to external processes.1

1

2

Related Stories

SurrealDB 3.0 raises $23M to replace five-database AI agent stacks with one unified system

Couchbase launches AI Data Plane to deliver persistent agent memory from cloud to edge

Google Cloud rebuilds enterprise data infrastructure for AI agents with Agentic Data Cloud

E2B Raises $21M to Build Cloud Infrastructure for AI Agents, Attracting 88% of Fortune 100 Companies

Performance Claims and Cost Savings for Enterprise Deployments

Keewano claims the distributed, multi-sharded design running in Docker containers orchestrated by Kubernetes can query approximately 250 million events in less than half a second using vectorized instructions on standard CPUs instead of GPUs. Customers ingesting data from Snowflake or Google's BigQuery have reported token savings of approximately 84% compared with querying existing systems directly.2

2

The database supports Apache Kafka for event stream ingestion and works with Parquet and Apache Iceberg formats. Keewano charges based on the number of active entities rather than individual events, positioning the system to run alongside existing databases like Snowflake for conventional business intelligence while handling agent workloads separately.2

2

Origins in Gaming Industry Pain Points Drive Product Vision

Co-founder and CTO Pavel Bibergal previously ran data infrastructure for cross-platform gaming vendor Plarium, where his team started building with agents and large language models roughly two years ago. Despite a large budget and full analyst team, they could not get simple questions answered in real time about why outcomes occurred at sustainable costs. That gap led Bibergal to build KeewanoDB's first prototype independently before forming Keewano.1

1

A separate analytics layer includes an agent called Signal that runs anomaly detection and causal inference approximately once an hour, applying machine learning and statistical methods to test hypotheses and present 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.1

1

2