Independent reproduction will determine whether Kimi K3’s AgentENV can fork dirty-memory microVMs in roughly 100 milliseconds and use that capability for practical scalable agent isolation.
state: expiredheat: lowuncertainty: highconvergesscott: highagent-harnesses microvms agent-sandboxing kimi-k3KimiGensee AI
What is this?
Moonshot’s Kimi team and kvcache-ai reportedly open-sourced AgentENV, a distributed sandbox platform said to have powered agentic reinforcement-learning training for Kimi K3. It runs agent environments in Firecracker microVMs, supports snapshot-backed boot or resume in under 50 ms, incremental snapshots and pauses in under 100 ms, and can fork a running sandbox into as many as 16 children on one node. The supplied snippets do not establish that the fork operation itself completes in roughly 100 ms, that the performance has been independently reproduced, or that Gensee AI is behind the system; those remain claims to investigate.
Why it matters to Scott
AgentENV independently implements the fast, disposable microVM execution substrate underlying Scott’s Sandboxed Execution, Goal-World Isolation and active SiloOS work. Reproduction of stateful forks at the reported latency would materially inform SiloOS implementation and sandbox economics, while also testing the boundary between mission-local cloned state and Scott’s preference for stateless workers with durable external continuity.
ip:concept.sandboxed-executionip:framework.siloosip:concept.goal-world-isolationip:concept.stateless-executiondev:project.silo-osradar:concept.agent-sandboxingradar:replit-snapshot-agent-isolationradar:kubernetes-agent-sandbox-adoptionradar:prime-intellect-agentic-rl-365k-envs
queries asked of Scott's wikis
- microVM snapshot forking for agent sandboxes
- dirty-memory cloning and stateful agent environments
- Firecracker isolation in coding-agent harnesses
- scalable sandbox orchestration for agent RL
- sandbox startup latency and agent workload economics
- E2B-compatible agent execution infrastructure
Measured heat
no measured readings yet — the hourly heat pass fills this in
How the heat travelled
no chain yet — the hourly chain pass fills this in
Evidence (2) — ⭐ canonical anchor
Interpretation history
2026-08-09T19:42:26Z
The reproduction window has faded without an independent benchmark, implementation report, or first-party clarification of the claimed dirty-memory fork latency. The claim remains potentially relevant to SiloOS, but this episode no longer merits active tracking unless substantive validation appears.
2026-08-07T19:34:36Z
No independent benchmark, implementation report, or technical clarification has appeared; the consequential 100 ms dirty-memory fork claim remains single-source and unvalidated. Staleness alone does not change the hypothesis, but there is no reason for near-term attention.
2026-08-04T23:26:43Z
grounded: converges/high — AgentENV independently implements the fast, disposable microVM execution substrate underlying Scott’s Sandboxed Execution, Goal-World Isolation and active SiloO
2026-08-04T23:24:38Z
case created — The claimed fork latency is a specific and consequential agent-runtime result, but currently rests on a single low-attention technical analysis.
Decision trace
- 08-10 05:42expireThe reproduction window has faded without an independent benchmark, implementation report, or first-party clarification of the claimed dirty-memory fork latency. The claim remains potentially relevant
- 08-10 05:42alert_silentOnly the staleness threshold fired; no new evidence changes the technical claim or creates a time-sensitive decision for Scott.
- 08-10 05:42alert_routeOnly the staleness threshold fired; no new evidence changes the technical claim or creates a time-sensitive decision for Scott.
- 08-08 05:34repriceNo independent benchmark, implementation report, or technical clarification has appeared; the consequential 100 ms dirty-memory fork claim remains single-source and unvalidated. Staleness alone does n
- 08-08 05:34alert_silentThe only trigger is elapsed time, with no new evidence or consequential delta; this can wait for an independent reproduction or first-party benchmark details.
- 08-08 05:34alert_routeThe only trigger is elapsed time, with no new evidence or consequential delta; this can wait for an independent reproduction or first-party benchmark details.
- 08-05 09:26groundAgentENV independently implements the fast, disposable microVM execution substrate underlying Scott’s Sandboxed Execution, Goal-World Isolation and active SiloOS work. Reproduction of stateful forks a
- 08-05 09:24createThe claimed fork latency is a specific and consequential agent-runtime result, but currently rests on a single low-attention technical analysis.