Agent frameworks & infrastructure
Agent sandboxes & runtimes
1,721 agents · most starred first
Agent sandboxes & runtimes, page 10 of 18
Core CLI engine for AI agent workspace sandboxing, universal IDE binding, and GitOps configuration governance. Local-first. Zero-dependency.
Live visualizer for Agent Substrate: watch AI agents restore from snapshots, queue, run in gVisor sandboxes, and checkpoint back to object storage
DSH sandbox tester: process-isolated testing ground for DeepSeek Harness
Copier template: isolated local Docker dev container per Orca ADE workspace, with Claude Code, git, gh/glab pre-authenticated via repo-scoped tokens. GitHub and GitLab (incl. self-managed).
Docker Sandboxes kit for running Meta's Muse Code CLI in an isolated sandbox
DSHSH — a harness that harnesses your harness
MHOS (Multi-Harness Operational System) — open-source, self-hosted multi-harness runtime.
Kernel for building agentic operating systems
Deep Rust reimplementation of DeepSeek Elastic Compute (DSec) — sandbox infrastructure for agentic RL training at scale, with a PufferLib core port
Duso is a complete server runtime built from the ground up for your next project. Free and open source. Clean script language powered by Go. No npm or dependency hell, just one 12MB binary. Efficient on everything from cheap virtual servers to multi-core powerhouses.
边缘智能体管理平台 — 让工业网关拥有 AI 大脑 | ONNX 推理 · LLM · RAG · 多智能体编排 · 断网自治 · 全协议接入
Agent-first display server: a small trusted core speaking a capability-native protocol, with every legacy app confined to its own per-app shim. Humans and AI agents operate the same GUIs under revocable, capability-scoped authorization.
Open-source Agent Environment & Host Runtime (macOS Apple Container / Docker / WSL 2). Stream native GUI windows (Wayland to Metal), MCP server, and instant snapshots for AI Agents.
The Open Agent OS & Sandbox Environment. Arch/kata-based Linux with git-like snapshot trees, CRIU memory restore, and per-window GUI streaming for Computer Use agents.
Give Claude Code, Codex and any MCP agent hands on every device you own: run commands, move files, use desktops and browsers. Peer to peer over iroh, end-to-end encrypted, no account, per-device approvals.
A Rust OS for AI agents on a real phone — Linux kernel for drivers, Rust userspace for the system. Pixel 5: voice-first agent, camera as peer input, screen as HTML canvas. 给 Agent 的操作系统。
AI-native Linux desktop. Built with AI.
Self-hosted AI agent runtime in a single Go binary: local models via Ollama, persistent SQLite state, a web UI and REST API, and tools from plugins, MCP servers and sandboxed JS/Wasm.
MicroVM lifecycle engine with snapshot, restore, guest control, and on-demand data loading.
Hardened Docker sandbox for OpenAI Codex CLI with workspace-only mounts, per-project auth, and allowlisted network egress
E2B cloud sandbox integration for pi — redirects all tool execution to a secure remote E2B sandbox
Opinionated lightweight sandbox for daily dev tasks with coding agents
Secure, self-hosted browser-streamed Linux desktops with disposable malware-detonation sandboxes.
Open-source, self-hosted control plane for sandboxes: one API, tenancy, quotas and policy-gated placement across OpenSandbox, 15 more provider adapters (E2B, Daytona, Vercel, AWS, Azure, Google Cloud…) or your own backend.
Autonomous AI agents inside a Qubes-isolated sandbox - tag-scoped Admin API access with dom0-mediated trust boundary.
The orchestrator for AI agents: runs Claude and OpenAI-compatible agents as sandboxed, budgeted, traced workloads with human approval and MCP, on a small Kubernetes-like orchestrator written from scratch in Go (Raft HA, WireGuard, Docker/Podman, dashboard).
MIAO.my · Agent Native App Runtime
Fast Firecracker microVMs for AI agents' builds and tests, on your own Linux host.
Virtual machines on Apple Silicon that you and your AI agents can both drive: macOS, Linux and Windows guests, seamless windows, MCP server
Open-source artifact publishing for agents and teams
The missing execution model for AI agents. Turn every machine you can reach into capabilities your agents can safely use: files, commands, services. Every action policy-gated, traced, and audited. Works over Tailscale, WireGuard, LAN, Kubernetes, or Operon's own peer-to-peer links.
Universal code & logic calculator for AI models: 51 MCP tools across 31 languages. Rust-sandboxed execution with verdicts, sessions, artifacts, exact arithmetic, and verified translation/optimization. No LLM, no gateway, no telemetry: the caller is the model. Apache-2.0.
Gabby is an agent runtime and governance framework that helps developers configure agent behavior, control tool access, and run tasks within explicit policies and limits.
Open-source, self-hosted AI coding agent runner. Run Claude Code, Cursor CLI, OpenCode & Pi in isolated microVMs or gVisor sandboxes with 2.8s warm snapshots.
A fresh Linux VM on your Mac in half a second. Throwaway sandboxes for coding agents on Apple Silicon, with credentials brokered from the Mac.
AI Agent + Data Canvas + Linux Sandbox on your phone.
Deploy your agents on Cloud Run!
Ultra-lightweight Go AI Agent 鈥?<10MB RAM, <1s startup, runs on $10 hardware. Edge intelligence for IoT, monitoring, and automation.
# 🐢🚀 Node.js Sandbox MCP ServerThis repository hosts a Node.js server that implements the Model Context Protocol (MCP) for running JavaScript in isolated Docker containers. It allows for on-the-fly npm dependency installation, making it easy to execute code safely and efficiently.
One Docker. Zero Worries. A containerized server-side agent platform with white-label branding, multi-user RBAC, and per-session Docker sandbox.
claw-family agentic distribution — OS, CLI, installer.
Agent Gateway with sandboxed, code-able tool calls 💥
Secure Docker sandbox for running AI coding agents in isolated containers — config-driven, single command, no GUI
An isolated environment for AI agents to execute code.
Runtime for agents operating on Canopy: wake on real work, route addressed coordination, and leave evidence.
一个基于WebContainer API的React代码沙箱组件库,让你能够在浏览器中运行Node.js环境。
OpenClaw plugin for AgentSandbox: execute Python and Bash code in secure cloud sandboxes with persistent sessions, file management, and execution history.
VM for AI
sandbox for coding agent
Secure Node.js execution sandbox for AI. Allows coding agents & LLMs to dynamically run JavaScript, install NPM packages, and retrieve results, facilitating code generation, testing, and interactive assistance. MCP-compatible.
Enterprise control plane for ephemeral AI agents via Firecracker MicroVMs
Lightweight, secure sandboxed command execution for AI agents
AMOF Runtime Authority - Governed execution for AI workers.
LeanClaw is a high-efficiency, security-first AI assistant runtime built for fast local execution, low resource usage, and reliable automation through constrained tool access.
Self-hosted AI assistant with web interface, integrated autonomous agents and sandboxed execution
Run Claude Code with --dangerously-skip-permissions safely. Per-project Docker sandbox with capability-drop, docker-socket proxy whitelist, git-RCE hardening, and shared OAuth token across projects.
Self-hostable AI agent runtime — single Rust binary, Tailscale cluster, runs across Mac/Linux/Windows/Android/iOS without a cloud account. Apache 2.0.
Local-first Agent Harness Kernel with Git-like control for safe, auditable AI agents.
A decentralized mini Linux distro for P2P private domain hosting (not onion domains). These domains are freely available for decentralized multi-stack app development and hosting with a new concept of protocol stack in cyber ecosystem.
Live demo of per-user/per-chat session isolation with Microsoft Foundry Hosted Agents. One agent, many sandboxes — visualised step by step.
Arbiter — Agent Runtime · Agent 运行时。上下文管理、配额调度、熔断器、审计轨迹。开源免费。pip install 即用。
strangeClaw is a minimal, self-hosted autonomous AI agent with full Firecracker sandbox isolation support.
Agentic workflow sandbox tailored for research tasks
A small, clear, and extensible multi-user Web Agent — chat, call tools, operate a workspace, browse the web, search for information, with each conversation running in an isolated sandbox.
A self-hosted cognitive operating system. AI agents as kernel processes. One Rust binary, no runtime dependencies, multi-tenant.
👷 Docker-for-agents: build agents into runnable container images
static region:us
A reference implementation for LLM Agent runtime engineering — native ReAct loop, sandbox isolation, flat-file memory, SSE streaming API
AOS — Agent Operating System: 给 AI 一个操作系统形态的运行框架,含可视化桌面、跨会话记忆、项目隔离与技能复用,纯文件系统零依赖
Run AI coding agents headlessly in Docker with a predictable contract. Pluggable runtime.
Easy, fast, isolated microVM sandboxes for AI agents and untrusted code on AWS Lambda MicroVMs — Python SDK, asb CLI, and MCP server.
Self-hostable AI agent runtime — single Rust binary, Tailscale cluster, runs across Mac/Linux/Windows/Android/iOS without a cloud account. Apache 2.0.
Scripts to configure a development Windows Hyper-V Gen2 VM, IPSW-based MacOS VM or Parallels MacOS Windows VM
A statically analyzable execution platform for nondeterministic programs, with reviewable authority and reproducible deployment state.
Run coding agents in isolated sandboxes, stream their work, and get reviewable pull requests with attached evidence. Self-hostable.
run https://pi.dev in a container
Self-hosted execution infrastructure for AI agents — your clouds, your machines, your rules.
A lightweight Docker sandbox for AI agents. Use it as a LangChain / Deep Agents backend, interact with it directly in Python, or expose sandbox tools through an optional MCP server.
A bastille for your bots
An AI agent that runs on Windows XP and has exactly one tool.
A local control plane for AI agents: crash recovery, human approval, cost caps and retry-safe effects, driven by an append-only event log
AGI OS is an Arch-based Linux distribution with AI agents built in — from installation to everyday use.
Open Agent Engine for long-running agents. More agents per GPU. No compaction. Every request rebuilt from a typed state frame so prompts stay within a fixed budget no matter how long the task runs. Built and measured on real software engineering work. AGPL-3.0 or commercial license.
🚀 Production-ready WebAssembly plugin examples for the Noorle platform - showcasing HTTP APIs, WASI 2.0, and the Component Model across Rust, Go, Python, JavaScript, and TypeScript
Ultra-minimal opencode agent runner for Cloudflare. Run AI agents indefinitely on your repos. Deploy in 5 minutes. ~/month. 300 lines of code.
LLM Agent with LangGraph and Docker for executing the agent's code in a containerized environment.
Welcome! This guide will walk you through running your Swarms Agent in a Trusted Execution Environment (TEE) using Phala Cloud. This setup ensures your agent runs in a secure, isolated environment.
The runtime your AI agents deserve. Process management, memory bus, tool routing, message passing, and scheduling for autonomous agents.
AI code agent sandbox for Mac Users
Agents 🧟 awake on trigger and provide an outcome
Python & JS/TS SDKs for embedding code execution capabilities in AI apps
Secure, isolated container for Claude Code with strict network firewall controls and optimized development tools (fd, ripgrep, fzf, jq)
Minimal AI agent kernel for a distributed future. <10MB binary, <50ms boot, Rust sandbox, dual-mode runtime. Pico Mode or full mesh — same binary.
Manage stateful, persistent agent workloads with Cloudflare Durable Objects for efficient, scalable real-time communication and scheduling.
A hardened, isolated development environment for VS Code with GitHub Copilot.
Secure command sandbox with ACL-based filtering preventing AI agents executing destructive operations like file deletions during panic states.
Lua-scriptable runtime for portable AI agents
Hot-reloaded agentic UI harness in a dev container
A local-first runtime for running many distinct agents — each with its own memory, secrets, skills, workspace, and autonomy — from one install.