Today for AI
HOT RADAR
agentHEAT 13.7°

AI CURATED HOT STORY · 11/8/2026

Author Rebuilds Google's Agent Sandbox on Firecracker and Benchmarks It

1 reports archived1 independent sourcesupdated 11/8/2026, 00:01:40
Synthesis & Latest Updates
1 Sources Cross-Validated

Google described its Agent Compute Engine on October 22 without latency or density figures, so the author rebuilt the same pattern on Firecracker in roughly 2,200 lines of Go: one microVM per agent session, checkpointed while the model thinks, cloned copy-on-write from warm snapshots, and no network card. Measurements show 64 idle forks of a 512 MiB VM cost about 3 MiB of host memory each; on bare metal a snapshot restore answers in 25 ms, 26x faster than a cold boot, and 64 concurrent forks are ready in about 150 ms, while twelve concurrent forks under nested virtualization on a Mac mini never became ready.

LATEST/Google described its Agent Compute Engine on October 22 without latency or density figures, so the author rebuilt the same pattern on Firecracker in roughly 2,200 lines of Go: one microVM per agent session, checkpointed while the model thinks, cloned copy-on-write from warm snapshots, and no network card. Measurements show 64 idle forks of a 512 MiB VM cost about 3 MiB of host memory each; on bare metal a snapshot restore answers in 25 ms, 26x faster than a cold boot, and 64 concurrent forks are ready in about 150 ms, while twelve concurrent forks under nested virtualization on a Mac mini never became ready.

Hourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first
  1. Hacker News AIT2·72 pts

    Author Rebuilds Google's Agent Sandbox on Firecracker and Benchmarks It

    • One microVM per agent session, checkpointing during model thinking, and copy-on-write cloning from warm snapshots form the prevailing isolation pattern for agent sandboxes
    • On bare metal a snapshot restore takes 25 ms, 26x faster than a cold boot, 64 concurrent forks are ready in about 150 ms, and each fork adds roughly 3 MiB of host memory
    • Under nested virtualization concurrent forks never became ready, so the virtualization layer must be validated before deployment