ObsessionDB

Teams running ClickHouse in production find that performance comes at an escalating cost as data grows: on other managed ClickHouse solutions, query latency rises and bills balloon with data volume, forcing teams to pay much more just to keep query times flat. This is especially painful for AI agent workloads, which probe databases relentlessly and in parallel rather than querying like humans, and which typically get charged per query even though they ask far more questions than a human would. Teams also lack visibility and control: no record of what an agent actually read, no isolation to stop an agent from slowing production traffic, and no safe way to let an agent experiment on real data without risking it.

ObsessionDB is a fully managed hosting layer for the ClickHouse engine, built on compute-storage separation, available on AWS, Google Cloud, and Azure in every public region. It uses a cache mesh of NVMe nodes that serves nearly all reads from hot cache while persisting writes asynchronously to S3 for durability, so queries never round-trip to object storage; the company reports p99 latency stays flat and cost per query falls as data scales from 1TB to 1PB. For AI agents specifically, it offers compute-compute separation so agents run on isolated, read-only compute without affecting production traffic, full query logging for auditability, compute-hour (not per-query) pricing, and CHKit, a set of structured commands letting agents describe, validate, test, and deploy schema and pipelines as code. It connects over MCP to tools like Claude, Cursor, Codex, VS Code, GitHub Copilot, and Zed, and an early-access feature lets each agent run operate on its own branch of production data. For enterprise customers it offers dedicated migration engineers, contractual SLAs with service credits, point-in-time recovery, SSO and column-level access control, private networking, dedicated compute, an