ClickHouse
The append-only analytical datastore used as the temporal telemetry intelligence layer. Stores raw metrics, logs, traces, cognitive events, and pattern outcomes. Optimized for aggregation-heavy time-series queries and intended to operate at 1,000+ service scale.
ClickHouse sits at the analytical layer of the memory substrate, handling the class of queries that OpenSearch is poorly suited for: aggregating over millions of time-series data points to identify trends, computing rolling statistics over telemetry windows, correlating cognitive event sequences with infrastructure metric patterns, and running bulk pattern-outcome analysis for the reinforcement feedback worker. Its columnar storage format and vectorized query execution make these aggregations extremely fast even over very large datasets - the design target is 1,000+ services each emitting metrics and logs at 15-second scrape intervals, producing tens of millions of rows per day. ClickHouse is append-only in practice: rows are written once by the ingestion pipeline and never updated, which aligns with the immutability principle that governs the episodic truth layer. The analytical results ClickHouse produces - trend signals, pattern confidence estimates, anomaly baselines - feed back into the associative memory layer's retrieval context and the grounding engine's prediction-error calculations.