Use cases

Where AT-1 pays for itself

Logs & observability

Cut the retention bill, query the cold archive

8–37× on real logs, byte-exact. Keep one archive that answers “errors in this 5-minute window” reading only the relevant time buckets — no rehydration. See the logs page →

Financial market data

Tick archives you can actually query

18.7× on real exchange ticks; a time-and-symbol query reads ~1% of the file and returns the exact rows. Half the size of Parquet, and recoverable. See the finance page →

Databases

Queryable, byte-exact export archives

Real MariaDB/MySQL dump → 26× (~2× better than xz); Postgres CSV → 16×; MongoDB/Elasticsearch NDJSON → ~28× — all queryable. Query the cold copy without restoring the DB. See the database page →

Genomics & science

Beats the incumbent format (BCF)

209× vs raw and 2.48× smaller than native BCF (2.95× vs .vcf.gz) on real chr22 — byte-exact, a hard requirement in regulated science. Region queries via positional zone maps. See the genomics page →

Lakehouse / tabular

27% smaller than Parquet, on the same table

A drop-in cold tier for Snowflake / Databricks / Iceberg: −27% vs Parquet-zstd, lossless, and still block-addressable — the biggest corpus on Earth is already in Parquet. See the lakehouse page →

Event streams / CDC

Smaller than the gzip your pipeline already runs

Kafka / Debezium / Fluentd sinks compress per segment with gzip; AT-1 slots in with O(chunk) cost and is 1.5–1.6× smaller, verified per segment.

Telemetry / IoT

Sensor time-series, compressed and queryable

22× on real telemetry; per-device, per-window queries via decimal + time zone maps.

Medical imaging (DICOM)

Wins the uncompressed archives

Structure-aware split beats uncompressed / RLE / color DICOM 2–13×, byte-exact — common in legacy PACS. Honest boundary: on monochromealready under JPEG 2000 / JPEG-LS, the image codec wins, so we don't claim it.