Model checkpoints
Fine-tunes of the same base model share most of their bytes. Peblob stores those bytes once.
S3-compatible storage proxy
S3 storage, stored once.
Peblob dedups your buckets by content, behind the S3 API you already use.
Your apps
aws-cli, boto3, rclone, your SDK
Peblob
splits objects into chunks, hashes each one
Storage
each chunk stored once
How it works
Peblob stores each unique chunk once, by content hash. Your apps keep using S3.
-
01
Point your clients at Peblob
Change the endpoint. Buckets, keys and requests stay the same.
-
02
Writes keep new chunks only
Each object is split into chunks and hashed. A chunk Peblob has already stored is not stored again.
-
03
Reads return the same bytes
Peblob puts each object back together from its chunks. Your apps get exactly what they wrote.
Where the bytes repeat
Backups
Nightly full backups, billed close to incrementals. Restores work exactly as before.
Build artifacts
Builds re-upload the same layers. Peblob keeps one copy and points to it.
How much you save
Savings depend on how much your data repeats. Model checkpoints and backups typically repeat the most. The scanner reports your exact ratio before you change anything.
Pricing
We charge a third of what we save you. You keep the other two-thirds.
Every invoice shows
- Baseline cost: what AWS would have charged
- Cost with Peblob
- Savings
- Our fee