Remote workspaces for coding agents

Give your agent
a computer.

Barad runs your coding agent in a remote Linux workspace, with root access and a browser. Leave a task running and check back from your laptop or phone.

Create account Watch the film 1:07

Bring a Claude, ChatGPT or Grok subscription, or an API key. Workspaces cost $0.20/hour while awake. Sleeping is free. Pricing

Isolated Linux workspaceTerminal and browser accessFiles saved between visits

Made in Barad

An agent made our launch film. This is its session.

One prompt, one Barad session, one Claude subscription. The agent read the code, rendered the film, found a bug in its own work, and went to sleep until the next message.

21:29Prompt sent. The session starts on its own Debian machine with root.film 0:00
21:31Reads the 523 files in the repository.film 0:18
21:40Searches the web for competitors.film 0:20
21:50Generates the voice and score with an API key from the session environment.film 0:24
22:54Renders 4,560 frames with Chromium and ffmpeg.film 0:26
23:15Watches its own render, finds a layering bug, fixes it.film 0:28
23:52Publishes a private report. The phone shows “ready for review”.film 0:32
00:02Ten quiet minutes. The machine sleeps at $0.00 an hour.film 0:48
09:35A new message wakes it. The files are still there.film 0:56

See an example

Every session runs on its own Linux machine with root access, a terminal and a browser the agent can see. When the work goes quiet, the machine sleeps and keeps your files, packages and conversation for next time.

Example sessionTwo repositories

Barad

Fix duplicate checkout orders

Ready for review

gpt-6-astra · High

Retrying checkout can create two orders. Reproduce it, fix the cause, and add a regression test.

10:42 AM

Worked5 steps · 6m 00s

thinkTracing retries from the storefront to the orders API.

bashnpx playwright test checkout-retryCompleted

readplatform/api/orders.tsCompleted

editplatform/api/orders.ts+11 −4Completed

bashnpm test -- checkout-retryCompleted

Retried checkouts now create one order

The server now reuses the original order for a repeated checkout key. The regression test sends two requests and verifies that only one order exists.

− await orders.create(cart)
+ await orders.getOrCreate(checkoutKey, cart)

checkout-retry.test.ts · 1 test passed

Completed · 10:48 AM · 6m 00s

Message the agent…

01 / How it works

Set up your first session.

Choose your repositories

Connect GitHub and clone multiple repositories into a workspace, or start with an empty directory. Git credentials are limited to the repositories you select.

Give the agent a task

Describe the change you want and how to check it. You can attach screenshots or ask for a test that reproduces a bug.

Run the code

The agent can install dependencies and run your project. It has a terminal and a browser for testing changes and taking screenshots.

Check the changes

Read the conversation, open the files in Files, or open the terminal yourself. Your files and conversation are still there when you return.

02 / Compare

Where Barad fits

Coding agents usually run in a terminal on your laptop or in one company’s cloud. Barad gives each session a Linux machine that stays yours between tasks, and it works with the model subscription you already pay for.

Barad compared with running an agent on your laptop and with single-provider cloud agents
Question	Agent on your laptop	Single-provider cloud agent	Barad
Keeps working after you close the laptop	No	Yes	Yes
Models	Any agent you install	That provider’s models and plan	Claude, ChatGPT or Grok subscriptions, or OpenAI and xAI API keys
The machine between tasks	Your own computer	Usually a fresh sandbox for each task	The same machine. It sleeps and wakes with files and packages in place
Browser and terminal	Whatever you set up	Varies by product	Chromium with screenshots for the agent, and your own end-to-end encrypted terminal
Check in from your phone	No	Often	Yes, with a notification when a turn completes
What you pay for the machine	Your hardware	Included in the plan or usage pricing	$0.20 per hour awake. Sleeping is free

03 / Pricing

Workspace and model costs

Workspace compute

$0.20 / hour

You pay for each started minute the workspace is awake, including idle time until it sleeps. Sleeping workspaces are free.

$10 in early-adopter credits covers 50 hours of workspace compute at this rate. Applicable tax is added when you buy credits.

Model usage

Claude subscription
ChatGPT subscription
Grok subscription
OpenAI API key
xAI API key

Connect one in Settings, then choose a model for the session. Your provider sets subscription fees, usage limits and any API or extra-usage charges. Barad credits do not pay for inference.

04 / FAQ

Workspace details

What happens when a workspace sleeps?

Your files, installed packages and conversation are saved. Running processes stop. Wake the workspace to continue, then restart any servers or other programs you need.

Can I use my Claude subscription?

Yes. Add a Claude subscription in Settings → Providers and complete Anthropic’s official sign-in. New Claude sessions default to Opus 5.5. Model access depends on your plan. Fable may bill usage credits without another confirmation.

Do I need to install anything?

No. Open Barad in your browser. Dependencies and builds run in the remote workspace.

Can I use the terminal myself?

Yes. Open the Terminal tab to run commands in the same workspace as the agent.

Can I share a preview?

Ask the agent to publish a static file or directory. Choose a public link or keep it private to your account. Published files stay available while the workspace sleeps.

Get started

Start your
first session.

A Linux machine for every session: root, terminal, browser.
Your Claude, ChatGPT or Grok subscription, or an API key.
$0.20 an hour while awake. Sleeping is free.

Create account