Portals
Anything that can listen on a port and speak HTTP can be a Portal: a unique URL routed to a service running in an Orb.
Every new thread spawns its own orb: a fresh machine with your repository cloned, tools installed, and the agent already inside. You set up nothing.
Orbs are infinite. Spawn one for a typo fix, a dep bump, a flaky test — no worktrees to juggle, no port conflicts, nothing to clean up when it’s done. And for the big stuff, orbs come in every size from tiny to xxlarge.
Orbs aren’t isolated boxes. Agents spawn other agents in fresh orbs, send messages, and pass files back and forth to hand off work.
An orb installs whatever the task needs — browsers, databases, toolchains. It grows around the work, but only the diff leaves the orb.
Rich diff review, live portal previews, multiplayer for teammates. And since nothing runs on your machine, you can pick the session up from any machine — even your phone.
Spawning, pausing, waking, billing by the minute. Paused orbs cost nothing. You stop thinking about machines and start thinking in threads.
Anything that can listen on a port and speak HTTP can be a Portal: a unique URL routed to a service running in an Orb.
Browse files, review diffs, inspect the agent’s evidence, and open a terminal in the same machine.
Share an Orb with your teammates so you, your agent, and your colleagues can work in the same thread and Orb.
Wake agents in Orbs at specific times or on regular intervals. They resume with the thread’s full context and history.
Mention @Amp in Slack to ask questions, fix bugs, and start work without opening Amp.

Give an Orb an externally available URL and send it events from CI, GitHub, Linear, or anything else that can make an HTTP request.

Agents can configure the software installed in an Orb, start other agents in fresh Orbs, and send messages and files back and forth.

Orbs are ephemeral remote machines where Amp agents can run without supervision. When you create a new thread on ampcode.com, you get a fresh Orb with your cloned repository, plugins, and the tools the agent needs.
We think they’re fantastic. They fundamentally changed how we develop software as a team, possibly more so than coding agents themselves.
While the combination of “remote machine” and “agent” sounds similar to many homemade setups for running agents remotely, using Orbs in Amp is different from remote-controlling agents on a remote machine.
Orbs are infinite.
You can spawn as many as you want without running into any resource issues. No port conflicts, no disk space shortage, no memory pressure.
This changes things tremendously: instead of treating machines on which your agents run as pets, giving them individual names, tending to their wants and needs, you just stop thinking of the machine altogether.
You go from “is this worktree dirty? Is the dev server running in this one?” to “what do I want to get done?”
Open and control the same Orb thread in your desktop browser, the Amp CLI, or on your phone. The conversation stays in sync.
People hear the term “sandbox” and think we’re talking about child’s play. But Orbs range from tiny to extra extra large. You can run small one-off scripts in them or have the agent run a full E2E verification with a headless browser on a multi-service application, handing you screenshots.
You can either choose a size for an individual thread, set a default for a project, or set a workspace default for new projects.
| Size | CPU | Memory | Disk | Price |
|---|---|---|---|---|
| a1.tiny | 1 | 2 GB | 60 GB | $0.08/hour |
| a1.small | 2 | 4 GB | 60 GB | $0.17/hour |
| a1.medium | 4 | 8 GB | 60 GB | $0.33/hour |
| a1.large | 8 | 16 GB | 60 GB | $0.66/hour |
| a1.xxlarge | 16 | 32 GB | 60 GB | $1.32/hour |
Orbs are billed by the minute. They pause automatically after five minutes of inactivity, and paused Orbs cost nothing. Amp handles pausing and waking them.