// the find
hyperdxio/hyperdx
Resolve production issues, fast. An open source observability platform unifying session replays, logs, metrics, traces and errors powered by ClickHouse and OpenTelemetry.
HyperDX is a self-hosted observability platform that puts logs, traces, metrics, and session replays into one UI, all backed by ClickHouse. It's now the open-source front-end for ClickHouse's ClickStack offering, which is both a strength (ClickHouse's query speed is real) and a complication (it's not just a standalone tool anymore). Teams drowning in Datadog bills who already have or are willing to run ClickHouse are the target.
ClickHouse as the storage backend is a genuinely good technical call — columnar storage makes time-series queries fast even at high cardinality, and you can run this on data volumes that would bankrupt you on Datadog or Honeycomb. The schema-agnostic approach means you can point it at an existing ClickHouse cluster rather than being forced into their data model. Full OpenTelemetry support means you can instrument once and switch backends without re-instrumenting. The MCP server integration (packages/api/src/mcp/) is a practical addition — letting an AI assistant query your observability data directly is actually useful, not just a demo.
MongoDB is a hard dependency for storing configuration (dashboards, alerts, saved searches) which makes the operational footprint bigger than the Docker one-liner implies — you're running ClickHouse + MongoDB + the OTel collector + the app itself. The project is now effectively a ClickHouse Cloud acquisition target (they rebranded it as part of ClickStack), which raises real questions about long-term open-source commitment — the README already points you to ClickHouse Cloud sign-up as the first hosting option. Anonymous telemetry is on by default and you have to opt out, which is a bad default for a self-hosted tool where privacy is often the reason you're self-hosting. Test coverage is reasonable for API routes but the ClickHouse query-rendering layer has snapshot tests rather than behavioral tests, so breaking changes in query generation are easy to miss.