ClickHouse Builds Agent AI Stack
$250M/year Databricks for AI agents
ClickHouse is no longer selling only a fast database, it is assembling the operating layer around AI agents. HyperDX gives it the screen where engineers inspect logs, traces, and session replay. LibreChat gives it the chat workspace where employees and developers talk to models and data. Langfuse gives it the system for tracing prompts, evaluating outputs, and tracking cost and quality over time. PeerDB extends that stack upstream by pulling operational Postgres data into ClickHouse in near real time.
-
Each acquisition fills a concrete workflow gap. HyperDX turned ClickHouse into a full observability product with a native UI and later became the base for ClickStack in ClickHouse Cloud. LibreChat added a multi model chat front end for internal agents and data assistants. Langfuse added the debugging and eval layer that teams need once agents are live in production.
-
This mirrors the broader warehouse playbook. Databricks paired its core platform with Neon for serverless Postgres and Agent Bricks for building and managing agents. Snowflake paired Crunchy Data for Postgres with Observe for AI powered observability. The pattern is converging on one vendor owning storage, app data, agent runtime, and monitoring together.
-
The deeper point is distribution. HyperDX and Langfuse were already built on ClickHouse, so buying them turns existing workload pull into product pull. A team that starts with LLM tracing or infra debugging can land on ClickHouse naturally, then expand cloud spend as telemetry, traces, prompts, and eval data accumulate inside the same system.
The next step is tighter packaging into a single default stack for agent builders. As Databricks and Snowflake push the same bundle logic from the warehouse side, the winners will be the platforms that make it simplest to ingest application data, run agent workflows, inspect failures, and improve model behavior without stitching together five separate tools.