Open source · Rust · Alpha

The same mount.
Anywhere your code runs.

A POSIX filesystem with a multi-tier cache over S3-compatible storage.

See how it works Talk to us

— Before · After

Stop building the same cache.

Fig. 1 — Today

Per company. Privately. Forever.

Fig. 2 — With TierFS

One mount. Same everywhere.

— Use cases

Built for these workloads.

Model weightswarm replicas

$ tierfs mount models /mnt/models
$ vllm serve /mnt/models/llama-3-70b
# next replicas: warm from peers, ~seconds

Training datamulti-epoch reads

for epoch in range(10):
    for shard in dl("/mnt/data/c4/"):
        train(shard)
# epoch 2+: served from NVMe, not S3

Dev → prodsame path everywhere

# laptop / sandbox / GPU node / k8s
$ tierfs mount data /mnt/data
$ python train.py
# identical command, every environment

Agent sandboxesfiles, not connectors

$ tierfs mount ctx /mnt/ctx
$ ls /mnt/ctx
drive/  local/  s3/
# agent reads files. no SDK per source.

— Read path

Three tiers. All optional.

Fig. 3 — Cache hierarchy

Filesystem

POSIX subset

Platforms

Linux, macOS

Cache

Local + peer + tier

Backends

Pluggable

— Sources

What you can mount.

Available

Coming next

Roadmap

AWS S3

MinIO

Cloudflare R2

GCS

Wasabi

Backblaze B2

Ceph RGW

S3-compatible

Local FS

Google Drive

Azure Blob

Postgres

SFTP

— Quickstart

Install. Mount. Read.

$ cargo install tierfs

# ~/.tierfs/buckets/data.toml
[[source]]
type     = "s3"
bucket   = "my-data"
mount_as = "s3"

$ tierfs mount data /mnt/data
$ cat /mnt/data/s3/file.txt
# served from local cache

— Writing

Notes from the build.

Essay · 9 min · June 25, 2026

Every Team is Building the Same Cache

→

all writing →

— Stay in the loop

Following along, or about to be?

Drop a line. Early conversations shape what gets built next. Email goes straight to us.

Or just write to hello@tierfs.com

The same mount.Anywhere your code runs.