DuckDB / DuckLake Server (With Arrow Flight SQL) for iOS
TPC-H 1GB in 2 seconds on iPhoneβArrow Flight SQL running locally.
Free, open-source OpenTelemetry backend powered by DuckDB π¦ and SQLite
Replaces ClickHouse clusters with a single DuckDB container handling 140K events/sec on an M1 Max.
DevOps engineers and homelab enthusiasts tired of complex observability stacks
Honeycomb Β· Grafana Tempo Β· ClickHouse
I started this as a project to build an observability stack for my homelab. After years of using the established OSS tools, they felt fragmented, had a steep learning curve, and often inflexible. I wanted to build a Honeycomb-like experience that unifies logs, metrics, and traces in a single tool that can be as easy to deploy as possible and free.
Specifically, I wanted it to - work well with the wide events pattern. - have sufficiently high throughput on a single host. - treat both human and agent interfaces as first-class citizens.
I initially planned to build on ClickHouse, but switched to DuckDB to keep distribution dead simple. I was also inspired by Motherduck's [Perf is not enough](https://motherduck.com/blog/perf-is-not-enough/) post. When DuckLake came out around the same time last year, it felt like the right foundation had just arrived. How good would it be if your OpenTelemetry backend just operate on a directory of Parquet files? So here it is: a small, open-source tool in an era where LLMs dominate the headlines. Hope some people find it useful.
TPC-H 1GB in 2 seconds on iPhoneβArrow Flight SQL running locally.
LLM-powered incident grouping with MCP server when Datadog and Honeycomb already dominate.
Sits between logs and Datadogβeliminates retry noise, saves 60β90% ingestion volume.
Honest learning project in a solved category β Grafana Loki and Datadog already dominate.
ByteBuddy injects trace context into Spark tasks; sees executor-level details no competitor offers.
DuckDB WASM and OPFS keep your CSV data local during AI analysis.