onboard
Built for agents
One request grounds an agent completely: its exact scope, the live schema, and runnable examples drawn from real data. Productive in a single round-trip.
OliverDB is a fast columnar analytics database that ships as one small binary. Embed it in your application or run it as a service — same engine either way. Its cellular design scales from a tiny in-process instance to a coordinated cluster, and every query is scoped by the key that carries it, so it’s safe to hand to an untrusted agent.
One request grounds an agent completely: its exact scope, the live schema, and runnable examples drawn from real data. Productive in a single round-trip.
Every key carries a policy. Before a query runs, it is rewritten to fit that scope — narrowed, not just rejected — so a least-privilege key is safe to hand to an untrusted agent.
The engine ships as one small process. Run it in‑process or as a service while durable data remains in shared object storage — no heavyweight runtime to operate.
The unit of scale is a stateless cell over shared storage. Add or replace cells without moving durable data — no sharding project, no rebalancing, and no data loss when a cell disappears.
Vectorized, multi-threaded scans and pre-aggregated rollups return grouped answers in milliseconds — with standout single-core efficiency.
Query in standard SQL or a compact JSON DSL — same execution, same security path. Text search (terms, phrases, fuzzy, regex) is built in, right next to your aggregates.
A conductor poses a task; a swarm of small models queries the Engine directly and returns grounded, cited findings.
The same policy discipline — rewrite, don’t validate — in front of Postgres, Snowflake, and ClickHouse.