Your logs, folded.
Lossless compression for your logs. In our tests, 1 GB came down to about 20 MB and unpacked to the original, byte for byte. Where only the facts matter, we build a condenser for your format that goes smaller still.
Send a sampleRequest a pilotHow it works
Logs repeat themselves.
Services write the same kinds of lines over and over, all day. You pay to store every one of them, and pay again to move them.
General-purpose compression finds repeated bytes. It can’t tell the facts you use from everything around them.
Built for one format. Yours.
General tools have to work on any data. A condenser is compression written for your format. It keeps exactly the facts you need and packs them into a compact record, checked on your own logs.
Discovery
Under NDA, you send us sample logs. We measure them and report back with the size you can expect, what is kept and what is dropped. You see a number on your own data before you decide on a build.
Build
We write your condenser, a strict reader for wherever your data lands, and a harness that checks, field by field on your real logs, that the facts going in come back out. If you need logs rebuilt from it, we build that too.
Run
Formats change. On a monthly retainer we keep your condenser and its checks current, and send you a report each month.
Case study: BattleBytes Deck Tracker
66.7 MB in. 17.7 KB out.
BattleBytes Deck Tracker, our own app for Hearthstone Battlegrounds players, folds a game’s log into a record on the player’s PC. One 66.7 MB log became a 17.7 KB record, about 1/3,760 of the size, with every board, shop, fight and result still in it.
Game log 66.7 MB
Record 17.7 KB
We rebuilt the game from its record and compared it with the original.
- 15 of 15boards
- 15 of 15shops
- 14 of 14fight results
- Matchedfinal placement
The log spends 213 lines on a single minion. The record says it once.
Player names, BattleTags and account IDs never enter the record.
General-purpose compression, which keeps every byte, took the same log to 1.81 MB. The record uses it too, on the facts alone.
Next: a million games a month.
BattleBytes expects more than a million game logs a month and keeps them to train its models.
- 30–60 TBof raw logs a month
- 10–20 GBof records a month
$73,000–$145,000could be saved on storage and transfer in the first year
By the end of that year, the raw logs would fill 360 to 720 TB and cost about $7,400 to $14,600 a month to store. The records would fill 120 to 240 GB and cost about $3 to $5 a month. The gap grows every month the data is kept.
Sending data into the cloud costs nothing in transfer fees at list price. Taking it out does: downloading each month’s raw logs once for training would cost about $2,400 to $4,700. Downloading a month of records costs under $2.
A year of records fits on one laptop drive, so a training run reads gigabytes, not terabytes.
Our estimate for BattleBytes, against keeping the logs uncompressed and downloading each month once for training. Storage at 2.1 to 2.3 cents per GB a month and transfer out at 7 to 9 cents, a major cloud’s standard list prices, before request fees. Discovery measures your volumes.
Lossless log compression
1 GB of logs. About 20 MB stored.
Unpack it and you get the original back, byte for byte.
- 50×smaller, across 1.6 GB of logs
- 42%smaller than xz -9e, the best of the general-purpose tools we tested
- 2×smaller than a leading log-specific compressor, on six complete logs (2.9 GB)
| Compressor | Smaller by | 1 GB becomes |
|---|---|---|
| gzip | 12.5× | 82 MB |
| zstd -19 | 26.0× | 39 MB |
| Brotli 11 | 27.5× | 37 MB |
| xz -9e | 29.4× | 35 MB |
| FoldBytes, fast | 44.3× | 23 MB |
| FoldBytes, max | 50.7× | 20 MB |
Measured 18 Nov 2026 on public research log datasets and on logs from our own server, Windows PC and Mac. The log-specific comparison is size only, with that tool at its strongest setting; its archives can be searched without unpacking. Ratios depend on the data; non-log data compresses like standard tools. Sizes here are binary: 1 GB = 1,024 MB.
By log type, on complete logs
- 155×Windows servicing log (CBS)
- 119×979 Hadoop container logs, packed together
- 59×Android
- 56×Supercomputer system log
- 30×HDFS, one 1.5 GB file
- 25×Web access log
- Checked file by file. We unpacked all 1,133 outputs and compared each with its original. All matched.
- Unpack speed. It unpacks about 440 MB a second on four CPU cores.
- Where it runs. On your servers or in your cloud, so in production your logs never leave.
Tell us which logs you keep and how much. We sign an NDA, run FoldBytes on a sample and send you the numbers.
Don’t attach data to the first email; anything that arrives before the NDA is deleted unread.
Wherever logs pile up.
- Game studios. Match logs, telemetry and replays from every session.
- Devices and fleets. Event streams from IoT devices, vehicles and kiosks.
- Security and audit. Trails you keep for years. If you need the original bytes too, we plan for that.
- SaaS platforms. Log storage and egress bills that grow with every customer.
- Remote links. Data sent over slow or metered connections.
How we handle your data.
- NDA first. We sign before you send a single line.
- Kept apart. Your sample lives in an access-limited place of its own, never in our code.
- Deleted after. When the work ends, your data is deleted, unless your contract says otherwise. Anything sent before an NDA is deleted unread.
- Privacy rules built in. You name the fields that identify a person in your data, and your condenser drops or masks them. Our reports hold sizes and counts, never your log lines.
Questions.
- Is it lossless?
- Our lossless compression is: each file unpacks to the original, byte for byte. A condenser is not. It keeps the facts you need, exactly, and the discovery report lists what it drops before you decide. Logs rebuilt from a condenser work the same in your tools.
- What do you need from us?
- A representative sample of your logs, what reads them today and what decisions come from them. Your daily volume, how long you keep logs and what you pay per gigabyte help us show the savings.
- Where does it run?
- Wherever your data starts or lands: in an agent on a device, as a library call in your service, or as a batch job in your pipeline.
Send us a sample.
Start with an email: what your logs are, roughly how much you keep and what you use them for. We sign an NDA, then agree how you send the sample.
Please don’t attach any data to your first email.