AgentENV (AENV) is a distributed platform for running agent environments at scale. - kvcache-ai/AgentENV

AgentENV is an open-source execution environment for large-scale Agentic RL that combines strong microVM isolation with fast snapshots, copy-on-write forks, and efficient resource reuse.

AgentENV (AENV) is a distributed platform for running agent environments at scale. - kvcache-ai/AgentENV

Nvidia, SpaceX, Microsoft launch AI safety initiative as OpenAI cyberattack fallout continues.[1] Kimi AI and kvcache-ai Open Sources ‘AgentENV’: A Distributed System that Powers Agentic Reinforcement Learning (RL) Training for Kimi K3.[2] OpenAI CEO Sam Altman claims AI singularity has arrived.[3] Authors have mixed feelings about the $1.5B Anthropic copyright infringement ruling.[4] Sources: [1] [2] [3] [4]

Moonshot AI's much-watched Kimi K3 model has opened its weights. K3 is the world's first 3T-class open-weight model. Its specifications are striking. The model spans 2.8T total parameters and uses a mixture-of-experts (MoE) architecture, with 104B parameters actually activated. It supports native vision and a 1M context window, and it ranks near the top across many benchmarks. Moonshot AI also released K3's technical paper alongside the weights.

AgentENV 소개 코딩 에이전트나 코드 인터프리터를 운영해 본 팀은 실행 환경에서 같은 딜레마를 만납니다. 모델이 만든 코드를 그대로 돌리려면 격리가 필요한데, 컨테이너는 호스트 커널을 공유하니 격리 경계가 얕고, 가상 머신은 경계가 확실한 대신 부팅이 느려 태스크마다 새 환경을 띄우기엔 비쌉니다. 에이전트 강화학습처럼 환경 수가 수천 개 단위로 올라가면 문제가 한 단계 더 커집니다. 환경 대부분은 모델이 다음 행동을 고민하는 동안 아무것도 하지 않으면서 CPU와 메모리를 붙들고 있고, 롤아웃마다 필요한 이미지가 달라 노드마다 이미지를 미리 받아 두는 비용도 함께 늘어납니다. AgentENV(줄여서 AENV)는 이 지점을 겨냥한 셀프호스팅 샌드박스 런타임입니다. 샌드박스 하나가 Firecracker 마이크로VM 하나로 뜨기 때문에 격리는 커널 단위로 확보하고, 그 위에 스냅샷 기반 부팅을 얹어 시작과 재개를 50밀리초 안에 끝냅니다....

AgentENV (AENV) is a Rust-based distributed platform for running AI agent environments at scale, now powering Kimi K3's agentic RL training. **Why It Matters:** Most agent frameworks are single-process toys. AENV makes agents distributed with Firecracker micro-VMs: * **50ms cold start** — snapshot environments boot or resume in <50ms * **Incremental snapshot/fork <100ms** — fork a running environment into independent sandboxes * **Overlaybd OCI images** — no pre-pulling, cluster-wide cache, images can exceed disk capacity * **ublk I/O** — host page cache sharing across storage and filesystem, high density * **S3 persistence** — snapshots persist to S3-compatible storage **Quick Start:** git clone https://github.com/kvcache-ai/AgentENV cd AgentENV && cargo build --release ./target/release/agentenv --node-id node1 **vs Alternatives:** | AENV | Ray | Kubernetes ---|---|---|--- Isolation | micro-VM | process | pod Cold start | 50ms | ~1s | ~1s+ Snapshot/fork | ✅ <100ms | ❌ | ❌ Agent-native | ✅ | ❌ | ❌ **Stars:** 1411 · **Language:** Rust · **License:** MIT https://github.com/kvcache-ai/AgentENV

Introduction: AgentENV and the New Infrastructure of Agentic RL Kimi AI and kvcache-ai open-sourcing AgentENV isn’t just another GitHub release.

Kimi AI and kvcache-ai have open-sourced AgentENV under the MIT license to enhance environment throughput for agentic reinforcement learning (RL) training. Designed for Kimi K3, this distributed syste

AgentENV (AENV) is a distributed platform for running agent environments at scale. - kvcache-ai/AgentENV

CryptoMarket and Blockchain News