S3, without changing your code
Buckets, uploads, downloads, listings, multipart and presigned links all speak the S3 API. Point your app at a new endpoint and it keeps working.
Open-source object storage
Your own S3.
In one binary.
Jay stores your files behind the same S3 API your tools already speak — on your own server, with no database and nothing else to run. Start it in a minute, keep it for years.
Free and open source (MIT) · No database to set up · Linux and macOS
terminal
$ export AWS_ENDPOINT_URL=http://localhost:9000
$ aws s3 mb s3://photos
make_bucket: photos
$ aws s3 cp cat.jpg s3://photos/
upload: ./cat.jpg to s3://photos/cat.jpgYour existing S3 tools work unchanged. Only the endpoint is new.
upload.py
import boto3
s3 = boto3.client("s3", endpoint_url="http://localhost:9000")
s3.create_bucket(Bucket="photos")
s3.upload_file("cat.jpg", "photos", "cat.jpg")boto3 and the AWS SDKs connect like they would to Amazon S3.
main.go
// The native protocol: no HTTP, no XML, no re-hashing.
c, err := client.Dial(ctx, "localhost:4444", tokenID, secret)
if err != nil { log.Fatal(err) }
defer c.Close()
c.CreateBucket(ctx, "photos")
c.PutObject(ctx, "photos", "cat.jpg", f, size, nil)From Go, a faster binary protocol on the same server and the same data.
main.go
// No server at all: the same engine, inside your program.
s, err := jay.Open("./data")
if err != nil { log.Fatal(err) }
defer s.Close()
s.CreateBucket(ctx, "photos")
s.Put(ctx, "photos", "cat.jpg", f, nil)Or skip the server: import Jay as a Go package and keep it in-process.
Works with the tools you already use
- AWS CLI
- boto3
- AWS SDKs
- Go
- Docker
How it works
From nothing to a stored file in about a minute.
No cluster to plan, no database to provision, no migrations. One container and a folder for your data.
-
1
Start one container
Give it a volume and two secrets. That is the whole setup.
docker run … ghcr.io/ivangsm/jay -
2
Create a key
Mint a key for your app and choose exactly what it is allowed to do.
POST /_jay/tokens -
3
Point your S3 client at it
Change the endpoint in your app or CLI. Everything else stays the same.
--endpoint-url http://localhost:9000
Features
Everything you need from object storage. Nothing you have to babysit.
Up to 5× faster from Go
Go services can use a lightweight binary protocol instead of HTTP. Same server, same files, a fraction of the overhead.
Or embed it as a library
Import Jay into your Go program and get the whole storage engine in-process. No network, no tokens, nothing to deploy.
Files that stay intact
Every object is checksummed on the way in and re-checked in the background. A crash mid-upload never leaves a half-written file behind.
Keys with exactly the access they need
Give each app or teammate its own key, limited to certain buckets, folders and actions. Share files with links that expire.
Nothing else to run
No database, no message queue, no cluster. Everything Jay knows lives in one folder you can copy, inspect and back up.
- Multipart uploads
- Range reads
- Server-side copy
- Hourly metadata backups
- Health checks
- Metrics and JSON logs
- TLS
- Linux and macOS binaries
Performance
Fast over S3. Faster over its own protocol.
Reads go straight from disk to the network without passing through the program. And when the caller is Go, the native protocol drops HTTP and XML altogether — the busier the server, the bigger the gap.
See every benchmark and how to reproduce it →Made for
Anywhere you would reach for S3.
Your app’s uploads
Avatars, attachments and documents, stored next to the app instead of on a third-party bill.
Build and CI caches
Fast, local artifact and cache storage that any S3-aware tool already knows how to use.
Media backends
Photos and video for a site or an image server, served with range reads straight from disk.
S3 for development and tests
A real S3 endpoint on your laptop and in CI, in one container that starts in seconds.
Why Jay
Small on purpose. Honest about it.
You own everything
Your files live on your disk, in a folder you can see. No per-request bill, no egress fees, no lock-in.
No silent surprises
If Jay does not support something, it says so with a clear error. It never pretends a request worked and quietly drops your data.
Know where it stops
One server, not a cluster: no replication or versioning. If you need those, Jay will tell you upfront instead of half-doing them.
Run your own S3 in about a minute.
One container, two secrets, a volume. Try it on your laptop today and move it to production when you are ready.