Most large-scale logging systems fall into two camps. Funnels like Kafka move high-volume, keyless streams—clickstreams, telemetry, bulk metrics—through a well-balanced pipe. They append data to immutable files with nearly 1× write and read amplification, so scanning the entire log in order is cheap. Routers, by contrast, need millions of small, addressable logs: message queues, feeds or microservice traces where each consumer only cares about one key’s subset of events. Running a router workload on Kafka means scanning huge partitions just to find a handful of records, driving read amplification to partition-size/record-size in the worst case. OpenData Log targets that routing use case. It’s a single Rust binary, MIT-licensed, backed by any object store. Instead of topic-partitions, it uses keys: you can spin up hundreds of thousands or millions of independent logs on one node. Under the hood it’s a segmented LSM tree over (key, sequence) pairs. New writes hit an append-only tree; older data compacts into sorted runs you can binary-search. When segments age out of retention you drop them wholesale. That architecture gives you efficient per-key scans and localizes failures. A “poison pill” in user-123’s stream won’t stall every other key. Resharding is simpler too—keys live in range partitions, so you split or merge without moving consumer offsets. Benchmarks on an m5n.xlarge against S3 show stable tail latencies even with millions of keys. In short, OpenData Log sacrifices none of Kafka’s durability or throughput but delivers the fine-grained access patterns that routers demand.