About
Smaller data, same science.
Compress is built by Flowershift, an AI and machine learning studio working on the physical world.
Our story
Built because our own data got too big.
Placeholder: Flowershift’s weather and Earth-system work produces petabytes of model output and imagery. Keeping it and moving it cost more than computing it. Compress started as the tool we built for ourselves.
Placeholder: why learned codecs beat general-purpose compression on scientific data, and why we insisted on hard error bounds from day one.
What we value
How we work.
Bounds, not vibes
Every lossy file comes with the largest error we measured. If we can’t verify it, we don’t ship it.
Files that last
Codecs are versioned, so a file written today still opens in ten years.
Real support
The engineers who build Compress help you fit it into your pipeline.
The team
Research scientists and engineers.
A small team who have spent years with very large datasets.
A Flowershift product
Part of something bigger.
Flowershift builds machine intelligence for the physical world: weather, acoustics, satellites and Earth-system models. Compress keeps all of that data affordable to store and move.
Latest articles
- News
Flowershift launches Compress
· 2 min read
- Scientific data
What an error bound actually guarantees
· 6 min read
- Case studies
Shrinking a national climate archive
· 12-page PDF
- Codecs
Learned codecs against zstd, SZ and ZFP
· 9 min read
- News
GRIB2 and Zarr v3 support in the Compress SDK
· 2 min read
- Edge & space
Compressing on orbit: more scenes through the same pass
· 7 min read
- Case studies
More scenes through the same pass
· 10-page PDF
- Storage & cost
The real cost of keeping a petabyte
· 5 min read
- News
Research partnership on compressing climate model output
· 2 min read
- Research
Will your files still decode in ten years?
· 6 min read
- Case studies
Checkpoints that don’t fill the cluster
· 8-page PDF
- Edge & space
Sensor data over thin links
· 4 min read
See it on your own data.
Every demo is run by the people who build Compress, on a sample of your own files.