A filesystem that outlives the machine

Close your laptop.
Your agents keep working.

PortableFS puts agent workspaces in the network, not on a disk. Adopt a directory at the gate, board the flight, and land to a volume full of new commits — every write ordered, durable, and mountable from the next machine.

$ curl -fsSL https://portablefs.com/install | sh

Open source under Apache-2.0open betabyopensteer

laptop — SFO, gate B32
17:58
$portablefs adopt ~/project
vol_project main adopted — now a live network volume
$portablefs exec vol_project -- claude -p "fix the flaky auth tests"
running in sandbox sb_18f2 against vol_project@main
boarding call. close the lid.
workstation — home
21:04
$portablefs mount vol_project ~/project
mounted vol_project@main (live)
$portablefs history vol_project
20:42 c41f09 +212 files tests green
19:19 9c22e1 +36 files rewrote token refresh
17:58 55d0b7 adopted at SFO gate B32

How it works

01

Adopt a directory

portablefs adopt turns an existing project into a live network volume. The files stop belonging to one disk; every machine and sandbox mounts the same bytes.

02

One ordered writer

Each branch has a single authority that puts every write in order — no sync, no merge, no last-writer-wins. That is why git indexes and SQLite files stay correct across machines.

03

History for free

Every write lands durably in the shared journal as your agent works — no snapshot you forgot to take. Snapshots are optional lightweight points for inspecting, rewinding, publishing, or forking; they never pause writes.

Fork per agent

Give every attempt its own volume. Keep the one that wins.

A fork is a full, independent volume created from any snapshot in seconds — agents cannot trample each other, and abandoning a bad attempt costs nothing. Run three fixes in parallel, diff the results, promote the winner.

vol_project@main

├── vol_attempt-1 claude tests green in 12m

├── vol_attempt-2 codex stuck on migration

└── vol_attempt-3 claude rewrote too much

$portablefs fork vol_project --name attempt-4

Works with what you already run

Claude Code

Point it at a mounted volume; every edit is durable in the journal and visible from your laptop mid-run.

Codex CLI

Run it in a sandbox against the volume. Kill the sandbox; the workspace is untouched.

E2B and any sandbox

Sandboxes die by design. Mount PortableFS inside and the workspace outlives all of them.

No SDK required — if it reads a filesystem, it works on PortableFS.

Open source core

Apache-2.0. The exit is a config change.

The entire data plane — volume API, authority manager, mount client, CLI — is open source. Self-host it with Docker and point the same CLI at your own deployment; this cloud is the hosted control plane, not a lock-in.

Self-host from GitHub

Simple pricing

Flat plans. Mounts and history are never metered.

Free

$0

forever

  • 1 volume
  • 1 GB storage
  • Unlimited mounts and history
Start free

Pro

beta plan

$20

per month, flat

  • 10 volumes
  • 25 GB storage
  • Unlimited mounts and history
Upgrade in console

Team

Soon

shared workspaces

  • Shared org volumes
  • Member roles
  • Usage-based storage

Your agents deserve a filesystem that outlives the machine.

Create a workspace
$ curl -fsSL https://portablefs.com/install | sh

Then portablefs login connects the CLI to your workspace with a one-time device code.