End-user documentation

How Capsules works for teams and agents.

This is the high-level product model: what a capsule is, how an AI agent uses one, where the limits are, and how credits map to active work. Internal control-plane and host details are intentionally left out.

In this guide

What a capsule is First agent workflow Restrictions Credits and usage

What a capsule is

A capsule is a small Linux environment you create through an API, CLI, SDK, or console. It is useful when an AI agent needs somewhere safe to clone code, install dependencies, run tests, expose a preview URL, and keep working state between sessions.

Persistent filesystem

Command and console access

Automatic sleep and wake

Per-capsule usage trail

First agent workflow

Create a restricted agent token, choose an approved size, start a capsule with a task name, run commands, inspect logs, and terminate the capsule when the work is merged or abandoned.

Scoped tokens

Naming prefixes

Runtime and storage limits

Network policies

Restrictions

Capsules are meant for controlled development and automation tasks, not unrestricted public hosting. Beta defaults keep sizes, regions, public URLs, private-network access, and idle behavior intentionally bounded.

One EU region at launch

No GPU or Windows guests

Private ranges blocked by default

One public URL per capsule

Credits and usage

Usage is shown as a ledger: active compute minutes, persistent storage, lifecycle events, and preview traffic. The initial onboarding flow starts with $5 of credits so a team can test real agent tasks without a sales call.

Prepaid credits

Minute-metered compute

Storage visibility

Audit events

Quickstart shape

Exact command names may change before public release, but the end-user workflow should feel like this: create a restricted token, start a capsule, run the task, inspect the result, and terminate it.

Step 1

Create a token

capsule tokens create --scope agent:write --prefix agent-

Step 2

Start a workspace

capsule create agent-pr-482 --size 2x2gb --credit-limit 2.00

Step 3

Run the agent task

capsule exec agent-pr-482 -- npm test

Step 4

Terminate when done

capsule terminate agent-pr-482

Current beta framing

Capsules is aimed at controlled engineering and AI-agent automation. It is not positioned here as a general public cloud, global hosting platform, GPU environment, Windows environment, or unlimited free tier.

Start with $5 credits Open console mockup