ObsessionDB is managed ClickHouse hosting or BYOC on AWS, Google Cloud and Azure. Free dev tier; example 3-node cluster with 10 TB estimated at $4,090/mo.
Researched and fact-checked by Bartu Cavusoglu · Updated October 26, 2026
ObsessionDB is a managed hosting service for ClickHouse software. It is built on the ClickHouse engine with compute-storage separation and is backed by Y Combinator. The company positions it as a data warehouse for AI workloads and as a lower-cost alternative to other managed ClickHouse services.
It can be used in two ways. In the fully managed mode, ObsessionDB hosts the cluster on AWS, Google Cloud or Azure in any public region. In bring-your-own-cloud mode, the same setup runs in the customer's own AWS or Google Cloud account or on their own hardware, and the ObsessionDB team operates it in both cases.
The core pitch is cost and latency at scale. The company says other managed ClickHouse services get more expensive as data grows, while ObsessionDB keeps p99 latency flat and lowers cost per query.
ObsessionDB uses a cache mesh as its query layer. Reads loop between the client and the nodes and do not leave the cache, so queries do not make a round trip to object storage. S3 holds the data for durability only, and writes persist to it asynchronously.
Because data lives in S3, a failed compute node can be replaced in seconds with nothing lost, and point-in-time recovery can roll a database back to any moment. Compute-compute separation lets separate workloads, such as an AI agent, run on isolated read-only compute over the same data.
The software is native ClickHouse, with the same wire protocol, SQL dialect and client libraries. The vendor says existing queries, integrations and tooling should work without modification, and that a dedicated migration team benchmarks and shadows production traffic before cutover.
For agent workloads, ObsessionDB provides isolated compute, full logging of every agent query down to the column, and no per-query charges. CHKit supplies typed commands for agents to describe, validate, test and deploy schema and pipelines as code. The database connects over MCP to Claude, Cursor, Codex, VS Code, GitHub Copilot, Zed and other MCP clients, and a per-agent sandbox is in early access.
Enterprise features include SSO through your identity provider, roles scoped to table and column, private networking, dedicated compute without noisy neighbors, and encryption in transit and at rest. Contracts include SLAs with service credits and monthly reports on uptime and query volume.
Operations are handled by the vendor: upgrades, scaling and 24/7 on-call, plus a shared Slack channel and monthly performance reviews. A free cost audit is offered, with $3K in credits if the customer converts.
ObsessionDB is a managed hosting service for ClickHouse software. It runs fully managed on AWS, Google Cloud and Azure, or in a customer's own cloud account or hardware as BYOC. It is Y Combinator-backed.
Managed pricing depends on compute and storage. In the site's example, 3 × 64 GB nodes cost $3,840/mo and 10 TB of storage costs $250/mo, for an estimated $4,090/mo. BYOC is one annual fee that is quoted on request.
Yes. A free developer tier offers up to 4 vCPU and 8 GB RAM with no credit card required. The company also offers a trial on larger instances if you contact them.
The company says it runs native ClickHouse software with the same wire protocol, SQL dialect and client libraries. Existing queries, integrations and tooling are said to work without modification.
It lists 46 integrations, including language drivers for JavaScript, Go, .NET and Rust, BI tools such as Looker, Grafana and Sisense, ingestion tools such as Kafka, PeerDB, Debezium and Spark, and MCP clients such as Claude, Cursor and VS Code.
Yes. It deploys into your own AWS or Google Cloud account, or on your own hardware, with an air-gapped option. Your cloud provider bills the infrastructure and ObsessionDB charges one annual fee for license, operations and support.
The company states that ClickBench results on identical workloads show comparable or better performance at 40 to 60 percent lower cost across AWS, GCP and Azure. These figures are vendor-reported.
The product is aimed at teams with large analytical workloads and technical staff. Small teams can try the free developer tier, but the paid clusters start at multi-node configurations priced in the thousands of dollars per month.