Keewano has launched KeewanoDB, a managed database built around a simple complaint from the agent era: AI systems are often asked to explain what happened after the useful sequence has been flattened into warehouse tables and aggregates. The company also announced $12m in seed funding led by Hetz Ventures, VentureBeat reports.
Watch Desk analysis
What happened
KeewanoDB stores events in sequence for each entity, such as a user, device or AI agent, rather than scattering the history across tables that need joining later. Developers connect agents through an SDK and the Model Context Protocol. The service is generally available on Google Cloud, with other cloud options and self-managed deployment planned for early 2027.
Keewano also describes a semantic layer that uses large language models during ingestion to add context as events arrive. Its separate analytics layer includes a natural-language query agent and Signal, which the company says performs anomaly detection and causal inference roughly once an hour. Those are product claims in the supplied VentureBeat report, not independent performance results.
Key findings
- Event history stays attached to the entity
A user, device or agent gets a continuous sequence, preserving changes such as active, inactive and fraud-flagged states. - Agents can query the database directly
The SDK and MCP connection are intended to remove a data-moving step between an agent and the history it needs. - The commercial launch has runway
KeewanoDB is generally available on Google Cloud and arrives with $12m in reported seed funding. - The semantic layer records changing context
Labels and context are added to the event stream rather than simply replacing earlier states.
Why it matters
An agent that can see only the latest row, summary or search result may know what happened without understanding why. Keewano’s answer is to preserve the sequence and make that context directly queryable, which could matter for support systems, fraud workflows and other applications where the trail is part of the answer.
The trade-off is that a clever storage model does not automatically produce reliable reasoning. The report does not establish independent accuracy, cost or latency results, and the company’s hourly anomaly and causal-analysis claims still need testing against real workloads.
Our read
This is a credible infrastructure bet on a problem that agent demos routinely hide: memory is only useful when its provenance and chronology survive. KeewanoDB is worth a look for teams building long-running agents, but the buying question is not whether it sounds unlike Postgres. It is whether preserving event history improves answers enough to justify changing the data stack.
What to watch
- Independent benchmarks for query latency, cost and answer quality.
- How the LLM-powered ingestion layer handles incorrect or ambiguous context.
- Whether self-managed and multi-cloud versions arrive as planned in early 2027.
- Which production customers adopt the system for consequential workflows.
Discussion spark: Would preserving an entity’s full event sequence improve your agents enough to warrant a new database, or should existing warehouses handle the reconstruction?
Sources and evidence
- Event-series database keeps agent history – Venturebeat (15 September 2026, 12:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.