What each agent gets
- Its own virtual machine, never shared with another agent or another customer.
- A real browser for sites that need logins, clicks and forms.
- Its own email inbox, for sign-ups and confirmation links.
- The run's shared folder, so later agents can build on earlier work.
- A model token that belongs to that one agent and expires with the run. Upstream API keys never enter a sandbox.
Why agents need their own sandbox
An agent with tools can delete files, install packages, follow a bad link or be tricked by a web page into doing something odd. Running it in a disposable machine contains that. Giving every agent its own machine also stops one agent's mess from breaking another's work, and makes each agent's actions easy to trace.
Lifecycle
- 01
Created on demand
Sandboxes start as the swarm starts agents, ten new ones at a time.
- 02
Kept while paused
If a run pauses at its budget or deadline, sandboxes are stopped with their disks kept for 7 days so the run can resume.
- 03
Deleted at the end
When a run completes or is cancelled, its sandboxes are deleted. The result and files remain.
Raw sandboxes or swarms?
If you want to build your own agent and only need a place to run its code, a raw sandbox provider is the right tool. Swarm Computers is for when you want the work done: the sandboxes, browsers, inboxes, coordination and budget control come together, billed per agent-minute.
Questions
- Can I rent a sandbox on its own?
- Not today. Sandboxes come with swarm runs and are billed in the agent-minute rate.
- Can agents reach the open internet?
- Agents can reach the domains you allow. Open web access is enabled per workspace after review.
- What happens to data in a sandbox?
- It is deleted with the sandbox when the run ends. Results and files the agents saved to the run folder are kept with the run.
Related
- A swarm of AI agents behind one API callSend one task and a budget; a swarm of AI agents works it in parallel, each with its own VM, browser and inbox. One merged result, never billed above your cap.Read more
- Multi-agent orchestration, hostedMulti-agent orchestration without the infrastructure: planning, parallel agents in their own sandboxes, budget limits and a merged result, behind one API call.Read more
- Agent swarms, explained and on tapWhat an AI agent swarm is, when a swarm beats a single agent, and how to run one through an API: many agents in parallel, each with its own computer, one merged result.Read more
- Two APIs. Two simple prices.Two prices: the Model API per token, and the Swarm API at one all-inclusive rate per agent-minute with a hard budget cap. Prices provisional.Read more