Callra's architecture draws a clean boundary.
What the analytics store is responsible for
The analytics store is the hot analytical layer for:
- website analytics events and rollups
- control-plane tables that support the product surface
- call and AI observability data after batches have been flushed
It is where dashboards, reporting, and high-cardinality analysis should feel fast.
What the analytics store is not responsible for
It is not the immediate durability buffer for the ingest API.
If the analytics database restarts or batch loads fail temporarily, Callra still wants ingest nodes to keep accepting events until their local queue capacity is exhausted. That responsibility belongs to the ingest layer, not the query layer.
Why the split is useful
This separation gives the system two clear operating modes:
- ingest durability mode on the node
- hot analytical query mode in the analytics store
That is simpler to reason about than making the same component solve both problems at once.