Skip to main content
Archil Hero Light Archil is cloud storage that runs your code. An Archil disk is a POSIX file system backed by your object storage. You can mount it from any server or sandbox, or run bash commands directly on it with disk.exec — no compute to provision. Disks spin up in milliseconds, sync bidirectionally with an S3 bucket, and have no capacity provisioning or egress fees.
Mount it
sudo archil mount aws-us-east-1/agent-workspace /mnt/data
Run code on it
import { Archil } from "@archildata/client/api";

const { disk } = await new Archil().disks.create({
  name: "agent-workspace",
  mounts: [{ type: "s3", bucketName: "my-bucket" }],
});

const { stdout, exitCode, timing } = await disk.exec(
  "grep -r ERROR logs",
);
Mount the disk when you need local POSIX latency, a long-lived process, or any tool that expects a real file system (training loops, Jupyter, model servers). The disk appears as a regular directory backed by a managed NVMe cache. Use disk.exec when you want a result instead of the raw data, when you’re fanning work out across one disk, or when you’re giving an agent a bash tool. Each call runs inside a container with the disk as the working directory and returns stdout, stderr, exit code, and timing. Billing is based on command wall-clock time.

Quickstart

Create a disk and mount it in under 5 minutes

Serverless Execution

Run bash commands on a disk with disk.exec

Archil for macOS

Mount your disks as native volumes from a menu bar app

CLI Reference

Learn about Archil’s command line interface

​
How it works

Archil runs a managed NVMe cache in front of your bucket, so reads are fast, writes are durable, and you get full POSIX semantics without the latency or consistency tradeoffs of S3 FUSE mounts. Mounted clients and disk.exec calls are both Archil clients and share the same read-after-write consistency model. A disk can point at one or more data sources — S3, GCS, R2, or any S3-compatible bucket — which Archil keeps in sync with the file system in both directions.

​
Built for AI workloads

​
Also great for