ObsessionDB is a fully managed ClickHouse® service that delivers sub-second analytical query performance by keeping hot data in a distributed cache mesh while using S3 for durable storage. It eliminates the cost and latency of warehouse-style architectures, enabling teams to serve real-time data directly from their analytical database. The platform also provides CHKit, an open-source toolkit for schema management and migrations in TypeScript or Python.
Funding
Funding not disclosed
AVTYFounders
Product
Problem
Traditional managed ClickHouse® solutions and cloud data warehouses often impose high costs and latency penalties as data volumes grow. Queries that scan large datasets can result in expensive per-byte billing, and architectures that separate storage from serving layers introduce pipeline complexity and data staleness.
Solution
ObsessionDB provides a fully managed ClickHouse® service designed for low-latency analytical workloads at scale. It uses a distributed cache mesh as the query layer, ensuring that reads never hit S3 and writes persist asynchronously, keeping hot data always accessible. The platform supports compute-compute separation, allowing agentic workloads to run on isolated, read-only compute without impacting production traffic. It also offers CHKit, a schema management and migration toolkit for TypeScript and Python, enabling teams to manage ClickHouse schemas as code.
Target Audience
Primary customers are data engineering and product teams serving real-time analytical workloads, including blockchain analytics platforms, DeFi API providers, and organizations with large-scale, low-latency serving requirements.
Features
- Distributed cache mesh that keeps hot data in memory, eliminating S3 round-trips for queries
- Compute-compute separation for isolated, read-only agent workloads that don't slow production
- Full query logging for every agent action, enabling replay and auditability down to the column
- CHKit open-source toolkit for schema management and migrations in TypeScript or Python
- MCP connectivity for Claude, Cursor, Codex, and other AI agents to query databases directly
- Incremental materialized views that update as data arrives, avoiding rescanning large source tables