Operational dataqueried live
Stop replicating production data into a warehouse just to chart it. Push down predicates, cache what's hot, and serve dashboards directly from your operational systems — governed, audited, sub-second.
Three things that make it fast
Real-time on SOFI is not a marketing word — it's a query path.
Predicate pushdown
SQLGlot rewrites your SELECT into native dialect for each engine — Postgres, Oracle, ClickHouse — so filters run at the source, not in your client.
Hybrid cache strategy
Hot dimensions cached in Redis, cold facts go straight to source. Per-view TTLs, source-aware invalidation, no stampedes.
Governed at line-rate
Policy evaluation runs in the same pipeline as the query. Masking, RBAC and rate limits are enforced before the response leaves the engine.
Source to chart in four hops
No ingestion. No scheduler. Just a typed query path with policy in the middle.
Connect
postgres.tx · clickhouse.events · redis.cache
Model
sales_live view · 6 cols · windowed agg
Govern
rate.tenant · cache.30s · audit.aggregate
Publish
sql · rest · jdbc to bi
What real-time actually looks like
A windowed sales aggregate, federated across two engines, served with cache and audit.
CREATE VIEW analytics.sales_live AS
SELECT
date_trunc('minute', e.event_at) AS minute,
p.region,
p.sku,
SUM(e.amount_cents) / 100.0 AS revenue,
COUNT(*) AS orders
FROM clickhouse.events.orders e
JOIN postgres.catalog.products p USING (sku)
WHERE e.event_at > now() - interval '15 minutes'
AND e.status = 'paid'
GROUP BY 1, 2, 3
WITH POLICY rate_limit_tenant, cache_30s, audit_aggregate
PUBLISH ON sql, rest, jdbc;What real-time actually buys you
Measured on the federation engine + DuckDB, not on benchmarks.
<800 ms
p95 federated
End-to-end query path: parse → plan → push down → fetch → policy → return.
1.2k
qps sustained
Per view, with predicate pushdown to source. Burst capacity scales with cache.
94%
cache hit ratio
Hot dimension cache plus per-view TTL cuts source load by an order of magnitude.
0
etl pipelines
No CDC, no scheduler, no warehouse round-trip. Sources stay sources.
What infra teams ask first
The questions that decide whether SOFI replaces the warehouse, sits next to it, or is the wrong fit.
We don't ingest. SOFI federates your existing operational systems live — Materialize and Tinybird ingest via CDC and serve from their own store. SOFI is right when you can't (or won't) duplicate the data; warehouses win when you need long retention and complex transforms.
// real-time, governed
Ship a live dashboard from your operational DB this week.
Trial includes the real-time-analytics recipe. Wire up Postgres + ClickHouse, point Tableau or Looker at SOFI's pgwire endpoint.