Archil now supports mounting from just-bash

TL;DR: You can launch an sandboxed instance of just-bash with an Archil disk by running:

npx @archildata/just-bash <region> <disk>

Vercel's just-bash gives AI agents a sandboxed bash environment in TypeScript—shell access without the security risks of real bash. It's especially useful in environments like AWS Lambda or Fargate where you don't have root access to set up traditional sandboxing.

Today, we're launching @archildata/just-bash, an adapter that connects just-bash to Archil. Agents can now run bash commands against your cloud data, working with datasets far larger than what fits in local memory.

What is just-bash?

just-bash is a bash interpreter written in TypeScript by Malte Ubl at Vercel. It implements 60+ Unix commands—ls, cat, grep, find, sed, awk, and more—in a sandboxed environment that can't escape to the host system.

This matters for AI agents. When you give an LLM access to a shell, you're handing it a sharp tool. just-bash provides the utility of bash commands without the security risk of actual shell access. The agent can manipulate files, run pipelines, and explore data—all within a controlled environment.

How the integration works

Archil already turns S3 buckets into POSIX-compatible filesystems. Our caching layer provides sub-millisecond reads for hot data while maintaining full read-write access to the underlying bucket.

The just-bash integration brings this same capability to sandboxed agent environments:

import { ArchilClient } from '@archildata/client';
import { ArchilFs } from '@archildata/just-bash';
import { Bash } from 'just-bash';
 
// Connect to Archil
const client = await ArchilClient.connect({
  region: 'aws-us-east-1',
  diskName: 'myaccount/agent-workspace',
});
 
// Create filesystem adapter and shell
const fs = new ArchilFs(client);
const bash = new Bash({ fs });
 
// Agent now has full bash access to S3
await bash.exec('ls -la /data');
await bash.exec('cat /data/context.txt | grep "relevant"');
await bash.exec('echo "agent output" >> /data/results.log');

Every file operation flows through Archil's caching layer to S3. The agent sees a normal filesystem. Under the hood, it's backed by infinite, durable cloud storage.

Why this matters

Agents can finally persist state. An agent running on Lambda can write to /data/memory.json, terminate, and pick up exactly where it left off on the next invocation. No more losing context between runs.

Agents can work with massive datasets. We're working with partners building AI agents that need access to terabytes of data. Now agents can grep through gigabytes of logs or iterate through millions of files—all through familiar bash commands.

Agents can share data. Multiple agents can mount the same Archil filesystem. One agent writes a file; another reads it. No coordination layer needed. The filesystem handles consistency.

It works everywhere TypeScript runs. Mac development machines. Fargate tasks. Lambda functions. Edge workers. If you can import a package, you can give your agent S3-backed storage.

Getting started

Install the packages:

npm install @archildata/just-bash @archildata/client just-bash

Or use the CLI directly:

npx @archildata/just-bash aws-us-east-1 myaccount/mydisk

This drops you into an interactive bash session backed by your Archil disk. Useful for debugging what your agents will see.