How OpenRouter observes, understands, and acts on 740 million LLM requests a day with ClickHouse Cloud

OpenRouter uses ClickHouse Cloud as a central analytics and operational data platform for its gateway serving more than 500 AI models from over 80 providers, supporting customer reporting, public rankings, internal agents, and real-time routing decisions. Since adopting ClickHouse in late 2024, OpenRouter has grown from roughly 600 million to 19.7 billion monthly requests, with its raw generation logs reaching about 134 billion rows and handling more than 110 trillion tokens weekly. Telemetry from each request is ingested through several pipelines, transformed through materialized views and a dbt-based warehouse, and reused for customer activity charts, provider health dashboards, model rankings, and a unified query interface. The company also enables staff to query documented warehouse data through Slack-based AI agents that can investigate customer issues and trace errors using bounded ClickHouse APIs. Routing has evolved from selecting providers based on price, speed, and availability to continuously evaluating reliability, quality, latency, and cost across modalities, while the Jev Router can now classify prompts, rank models according to recent real-world usage, apply customer constraints, and choose both a model and provider. Outcomes from each routing decision are stored back in ClickHouse, creating a feedback loop in which usage data improves subsequent recommendations and platform operations.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.