TECH Signal 402
Amp agents can now run remotely in orbs billed by the minute with local sync
↗
Amp introduced orbs, remote machines where agents run unsupervised with your code and tools, billed by the minute, with changes syncable back to a local machine.
Running agents on remote hosts was already possible through worktrees or SSH, but integrating remote agents into the same interface as local ones removes the friction that kept it rare. Multiple agents can now run in parallel without local resource contention, which shifts bug triage and long-running optimization from manual supervision to delegated agent work. Only one feed carried this story, so the claims rest on a single source.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Orbs are remote machines containing your code, plugins, and tools, billed by the minute, that start quickly and sleep when idle.
You can sync orb changes locally with `amp sync <thread>` and spawn orb agents with `amp -ox` or from within the Amp TUI.
Remote agents were previously possible via separate checkouts, worktrees, or SSH, but orbs integrate them into the same interface and controls as local agents.
THE READ
What the cluster adds up to.
Amp now lets you launch agents in orbs, which are remote machines provisioned with your code, plugins, and tools. They are billed by the minute, start up quickly, and go to sleep when neither you nor the agent needs them. You can spawn an orb-based agent with `amp -ox` followed by a prompt, or from within the Amp TUI without leaving it. The orb agent appears alongside your local agents with the same controls for reviewing changes, browsing files, and using a terminal.
To pull an orb agent's work onto your local machine, you run `amp sync <thread>` where the thread argument is a URL or ID. The material references orb sizes and pricing but does not include specific figures, so the cost of adoption beyond per-minute billing is not documented here. The sync mechanism is the bridge between remote and local work, letting you iterate alongside an agent that is running elsewhere.
Running agents on remote machines was not new: the article acknowledges using different checkouts, worktrees, or manual SSH sessions to remote hosts. What changed is that orbs sit inside the same interface already used for local agents, with the same controls, removing the operational overhead that made parallel agent work uncommon. The article frames this as a threshold crossing rather than an incremental feature, arguing that requiring agents to work on a single machine held them back.
The article is promotional in tone and describes how the Amp team's own workflow shifted rather than documenting edge cases or failure modes. It does not specify what orbs cannot do, what happens on disconnect, or how isolation and security are handled beyond the statement that each new thread gets a fresh orb. Because only one feed carried this story, there is no independent corroboration of the claims, and the practical limits of the feature remain untested in the material provided.
Written by elseif from the cluster below · checked for specifics the sources never containedTHE CLUSTER