Snapshots & forks

A warm base at deploy time

alchemy.run.ts

const warm = yield* E2B.Snapshot("Warm", {
  template: Agent,
  run: [
    "git clone https://github.com/my-org/app /home/user/app",
    "cd /home/user/app && npm ci",
  ],
});

Alchemy boots the template, runs the commands, snapshots the sandbox with its memory, and kills it. A command that exits non-zero fails the deploy with SnapshotSetupFailed. A deploy runs the setup again only when the template or the commands change.

Create sandboxes from it

const sandboxes = yield* E2B.CreateSandbox(Warm);
const sandbox = yield* sandboxes.create({ timeout: 600 });
yield* sandbox.exec("cd app && npm test");

Snapshot a running sandbox

const snapshot = yield* sandbox.snapshot({ name: `session-${id}` });
const restored = yield* sandboxes.create({ templateID: snapshot.snapshotID });

Pass memory: false to keep only the filesystem; sandboxes from it boot fresh.

Fork a sandbox

const results = yield* sandbox.fork({ count: 3 });
for (const result of results) {
  if (result._tag === "Forked") yield* result.sandbox.exec("npm test -- --shard");
}

E2B checkpoints the sandbox (it keeps running) and starts copies of it. Each fork succeeds or fails on its own.

Snapshots at runtime

const snapshots = yield* E2B.CreateSnapshot();
const info = yield* snapshots.create(`warm-${slug}`, { template: "node-22", run: ["npm ci"] });
yield* snapshots.delete(`warm-${slug}`);

Runtime snapshots are not deleted when you destroy the Worker.