{"slug": "agentenv-a-platform-for-running-agent-env-at-scale-kimi-k3-rl", "title": "AgentENV: A platform for running agent env at scale (Kimi K3 RL)", "summary": "AgentENV (AENV), a platform for running agent environments at scale powering agentic RL training for Kimi K3, launches with support for snapshot-backed environments that boot or resume in under 50 ms and pause in under 100 ms, incremental snapshots completing in under 100 ms, and an E2B-compatible HTTP API. The platform runs on Linux kernel 6.8+ and Ubuntu 24.04, using Firecracker microVMs with overlaybd for on-demand image loading and memory ballooning for high overcommit.", "body_md": "**Running agent environments at scale**\n\nAgentENV (AENV) is a platform for running agent environments at scale, powering agentic RL training for **Kimi K3**.\n\n**Scale across diverse environments**: AENV runs massive numbers of Firecracker environments across machines and diverse OCI-compatible images, loaded on demand via overlaybd. Local disk acts as a bounded cache, retaining hot data and evicting cold, so images can exceed disk capacity while startup stays fast cluster-wide, without pre-warming every host.**Make idle environments inexpensive**: Snapshot-backed environments boot or resume in under 50 ms and pause in under 100 ms. Idle environments can quickly release CPU and memory, then return when new work arrives.**Native snapshot and fork support**: AENV snapshots memory and filesystem changes incrementally, completing in under 100 ms even under heavy disk modification. A running environment can fork into multiple independent sandboxes for parallel agent workflows. Snapshots persist to S3-compatible object storage or a shared distributed filesystem to prevent data loss.**Preserve performance and density over time**: AENV delivers high-performance I/O via ublk while sharing the host page cache across storage and memory-snapshot data. Memory ballooning returns reclaimable guest memory to the host, sustaining high overcommit as environments run longer and diverge.\n\n**Linux kernel 6.8+**; the install script additionally requires** Ubuntu 24.04**(see*Quick Start*below for installation options)`/dev/kvm`\n\naccess for Firecracker microVM execution\n\n**1. Install and start the server**\n\n*Option A — install script (Ubuntu 24.04)*\n\nInstall both the server and the `aenv`\n\nCLI, then start the server as a\nsystemd service:\n\n```\ncurl -fsSL https://raw.githubusercontent.com/kvcache-ai/AgentENV/main/scripts/install.sh | sudo bash\nsudo systemctl start aenv\n```\n\n*Option B — Docker*\n\nSet up the server:\n\n```\ncurl -fsSL https://raw.githubusercontent.com/kvcache-ai/AgentENV/main/scripts/docker-setup.sh | sudo bash\ndocker pull ghcr.io/kvcache-ai/aenv-server:latest\ndocker run -d --privileged -v /dev:/dev -p 8000:8000 ghcr.io/kvcache-ai/aenv-server:latest\n```\n\nThe server is accessible at `http://127.0.0.1:8000`\n\nby default.\n\n**2. Install the aenv CLI** *(skip if you used Option A in step 1)*\n\nInstall separately if you used the Docker method above, or if you are running the CLI on a different machine from the server. Supports Linux and macOS on x86_64 and arm64:\n\n```\ncurl -fsSL https://raw.githubusercontent.com/kvcache-ai/AgentENV/main/scripts/install-cli.sh | bash\n```\n\n**3. Authenticate**\n\n```\naenv auth\n# AENV server URL [http://localhost:8000]: http://127.0.0.1:8000\n# API key: dummy\n```\n\n**4. Pull a template and run a sandbox**\n\n```\naenv pull ubuntu:22.04 --name ubuntu\naenv start ubuntu            # starts a sandbox and attaches an interactive shell\n```\n\nFor Docker Compose / Kubernetes cluster deployment and build-from-source instructions,\n\nsee 📖 [Deployment](https://kvcache-ai.github.io/AgentENV/deployment/manual-compile.html) .\n\nAgentENV exposes an E2B-compatible HTTP API. Point `E2B_API_URL`\n\nat your\nserver and use the standard E2B Python / TypeScript SDK without any code\nchanges. See 📖 [E2B integration](https://kvcache-ai.github.io/AgentENV/integration/e2b.html)\nfor setup details.\n\n```\n# Templates\naenv pull docker.io/library/ubuntu:latest --name ubuntu    # FROM <image> → template\naenv template list                      # alias: aenv template ls\n\n# Sandboxes\naenv start ubuntu                       # start + attach interactive shell\naenv start ubuntu --detach              # start, print sandbox ID, don't attach\naenv cn <sandbox-id>                    # reattach a shell\naenv exec <sandbox-id> ls -la /         # one-shot command\naenv ls\n\naenv pause   <sandbox-id>\naenv resume  <sandbox-id>\naenv timeout <sandbox-id> 600           # extend TTL to 600 s from now\naenv delete  <sandbox-id>               # alias: aenv rm\n```\n\n`aenv start`\n\naccepts a template UUID or human-readable name/alias. `aenv list`\n\noutputs a table on TTY and JSON when piped; override with `--output table|json`\n\n.", "url": "https://wpnews.pro/news/agentenv-a-platform-for-running-agent-env-at-scale-kimi-k3-rl", "canonical_source": "https://github.com/kvcache-ai/AgentENV", "published_at": "2026-07-27 17:36:25+00:00", "updated_at": "2026-07-27 17:53:05.425342+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-agents", "developer-tools"], "entities": ["AgentENV", "Kimi K3", "Firecracker", "Ubuntu", "E2B"], "alternates": {"html": "https://wpnews.pro/news/agentenv-a-platform-for-running-agent-env-at-scale-kimi-k3-rl", "markdown": "https://wpnews.pro/news/agentenv-a-platform-for-running-agent-env-at-scale-kimi-k3-rl.md", "text": "https://wpnews.pro/news/agentenv-a-platform-for-running-agent-env-at-scale-kimi-k3-rl.txt", "jsonld": "https://wpnews.pro/news/agentenv-a-platform-for-running-agent-env-at-scale-kimi-k3-rl.jsonld"}}