Skip to content

Long-running agents

An agent spends most of its time waiting: an approval, an email, a deadline. Suspend the machine in that gap. Memory and processes are kept, and resuming takes about 1 s.

from e2b import Sandbox
sandbox = Sandbox.create(timeout=2 * 60 * 60)
sandbox.commands.run("python3 agent.py", background=True)
sandbox_id = sandbox.sandbox_id # keep the id
# the agent asked for approval: suspend
sandbox.beta_pause()
# ... hours later, when the answer arrives
sandbox = Sandbox.connect(sandbox_id) # resumes where it stopped
sandbox.set_timeout(60 * 60) # renew the timeout
import { Sandbox } from "e2b";
let sandbox = await Sandbox.create({ timeoutMs: 2 * 60 * 60_000 });
await sandbox.commands.run("python3 agent.py", { background: true });
const id = sandbox.sandboxId; // keep the id
await sandbox.betaPause();
sandbox = await Sandbox.connect(id);
await sandbox.setTimeout(60 * 60_000);
  • Keep the sandbox_id outside the machine (database, file). It is what brings you back.
  • Suspended, the machine uses no CPU or memory on the host, but it still counts toward your machine limit. Remove what you will not use again.
  • After resuming, renew the timeout with set_timeout.
  • If resuming answers 503 (The beta is full right now), try again in a few minutes. The state stays saved.