Agent frameworks & infrastructure
Agent sandboxes & runtimes
1,721 agents · most starred first
Agent sandboxes & runtimes, page 6 of 18
Internal agent platform on Amazon Bedrock AgentCore: interactive Claude Code cloud workspaces + headless agent kernels behind one portal, with publishing, scheduling, channels, memory, evals and governance
The operating system built for AI agents. Purpose-built, open-source, engineered from the kernel up to run AI agents at hardware speed.
`swarmd` is a multi-tenant runtime for running background Agents in a safe and secure manner. Agents are defined in YAML and run as goroutines in a multi-tenant server with a virtual shell and custom tools.
Ngent wraps command-line agents that speak the Agent Client Protocol (ACP) into a web service, making them accessible via HTTP API and Web UI.
Docker sandbox backend for deepagents
Durable, multi-tenant runtime for agents that act on real systems — credentials never enter the model's context
An easy-to-use macOS sandbox for Claude and other AI coding agents
TaskForge Platform runs AI agents in sandboxed Docker containers with capability-based security. Agents start with minimal permissions and must request new capabilities (packages, network access, tools) through a human-in-the-loop approval process. Every approval triggers a container image rebuild, and every LLM interaction is logged for audit.
macOS Seatbelt sandbox CLI for developers. Protect credentials (SSH, AWS, GPG) from malicious npm packages, supply chain attacks, and untrusted build scripts. Deny-by-default filesystem isolation. Perfect for Claude Code agentic workflows with --dangerously-skip-permissions.
A secure agent harness runtime and framework
🦞 OpenClaw Pool — multi-tenant AI agent platform on AWS EC2 Firecracker microVMs with per-tenant isolation, auto-scaling, and web management console.
ZimaOS Blue - A Local-First Agent Runtime for Bold Builders. Out-of-the-Box, Open-Source, Universal, Vendor-Neutral
Security-posture-first list of sandboxes for running AI coding agents' code, ranked by isolation, egress control and secrets handling
A self-hosted web interface for the pi coding agent: chat, files, documents and an integrated terminal, with the agent's shell confined by OS sandboxes.
An Agent used to demonstrate the operation of code based on a sandbox in a multi-user environment
OpenFang 中文本地化版:面向中文用户的开源 AI Agent OS
端到端智能体平台:把本机的 AI Agent 变成人人可用的 Web 产品(Vue 3 + Flask + Go WS 网关 + Docker 部署)
Run OpenClaw and OpenCode in the browser. Self-hosted, multi-user, Helm-deployable.
mkenv is a fast, secure, reproducible local isolated development environment generator.
🏛️ Hermes Gate — Terminal TUI for managing remote Hermes Agent sessions with auto-reconnect, detach support, and zero config
OpenRouter for agents and sandbox environments. Provides a control plane and configuration options for persistent agent workers. Route runtime + substrates, provision specialist presets, and supervise your fleet.
Local-first desktop workspace for AI agents: browser, terminal, memory, plugins, runtimes, and durable loops.
Lima Devbox Claude Skill to help you setup a custom sandbox VM for AI agent use
Generate VM's with kernel tracing, code sandboxing and security profiles for long running agents.
Embedded multi-session terminal panel for Agent Zero.
No Mac Mini? No problem. HermesFace is the best free way to deploy Hermes Agent on HuggingFace Spaces — 2 vCPU, 16GB RAM, 50GB storage, always online, zero cost.
Open-source local agent runtime for CTRL NODE — the CLI bridge that pairs your machine with Claude Code, Copilot, Gemini, Cursor and more, and executes tasks from the CTRL NODE dashboard.
A sandbox service for running untrusted, agent-authored code that reports the isolation boundary each run executed behind. Lessons, no clone needed: https://plimsollmark.github.io/plimsoll/trainers/
Agent — A next-generation sandbox for AI code agents. Environment management, CI/CD pipelines, and multi-agent collaboration.
Secure Dynamic Session Agent for AI generated code execution built using OpenAI, Langchain and Azure Container Apps
Human-in-the-loop execution for LLM agents
Deploy agents to Claude Managed Agents, AWS Bedrock AgentCore, and Google. One folder, one definition, one command.
"Run AI coding agents (Claude, Mistral, other providers coming soon) in parallel inside hardened, isolated containers with git worktree separation.
Open-source, self-hosted runtime and SDK for AI applications. One tenant/project deployment serves many users and apps with governed tools, scoped credentials, isolated generated code, and cost controls.
在網路受限的 Docker Dev Container 裡安全執行 Claude Code,透過 iptables/ipset 防火牆白名單限制 AI Agent 的對外連線,降低資料外洩與誤操作風險。改自 Anthropic 官方 .devcontainer 參考實作。
Self-hostable, multi-tenant AI agent service with sandboxed code execution(可自托管的多租户 AI Agent 服务,支持沙盒隔离执行)
AI-Native Operating Environment — Linux-based OS with Rust kernel modules and Python AI/ML runtime. Local LLMs, multi-agent orchestration, and intent-driven computing, all on-device.
Joblet is a micro-container runtime for running Linux jobs with: Process and filesystem isolation (PID namespace, chroot) Fine-grained CPU, memory, and IO throttling (cgroups v2) Secure job execution with mTLS and RBAC Built-in scheduler, SSE log streaming, and multi-core pinning Ideal for: Agentic AI Workloads (Untrusted code)
Desktop AI agent runtime with dual-brain safety review, GUI-first operation, local model support, and multi-channel delivery.
PawFlow (Platform for Agentic Workflows) Self-hosted agent runtime for real infrastructure. Run durable AI agents against your own files, tools, browsers, desktops, services, and workflows with relay-backed execution, shared context, multi-provider agents, and deterministic flows.
Agent libOS provides a runtime substrate for self-evolving LLM agents with process-like execution, tool management, and security boundaries implemented through explicit capabilities and runtime primitives.
An agent runtime that treats context, execution, and permissions as OS primitives — not LLM problems.
Harness is all you need. The operating system for autonomous AI agents.
The Open-Source Agent Operating Environment Built for Getting Work Done
Orpheus is an Agent Runtime that solves the 'Silent Failure' of CPU autoscaling. It scales on queue depth, persists memory, and recovers from crashes automatically.
Secure and lightweight Python code execution environment for LLMs
Instant hosting for AI agents — turn output into pages worth sharing
🤖 🚀 Agent Runtimes — Managed AI Agents.
Real virtual desktops for AI agents. MCP-native, self-hosted, fully isolated.
A lightweight AI agent for legacy hardware. Written in Zig, runs on Windows XP SP3 x86 with 64 MB RAM. Connects to Ollama for LLM-powered system diagnostics and management via function calling.
Simple code sandbox supporting jupyter notebook style code execution. Used for agent training
Local-first AI project OS that turns coding models into reliable software agents with memory, sandboxed execution, verification, recovery, GitOps, and deployment.
Open-source continuation of Daytona — secure, elastic sandbox infrastructure for AI code execution and agent workflows. Community fork of the last AGPL-3.0 release (v0.190.0).
Portable, self-contained sandboxes for AI coding agents - microVMs, containers, isolation harnesses.
Elixir SDK for Fly.io Sprites: computers for agents. Manage Sprites and run remote commands from Elixir, with APIs that match the language's conventions.
Give every agent session its own VM. One command up, zero burn when idle.
CLI for Local Isolated Environments for Coding Agents
Run OpenClaw, Hermes and your own agents in microsandbox microVMs. One reviewable TOML role sets image, allowed domains and host-bound secrets. No daemon.
Lightweight AI agent runtime targeting Raspberry Pi Zero 2W and low-end VPS. File-based memory, MCP tool support, Telegram integration — all in a single binary.
A sandboxed bash interpreter for AI Agents, built in Rust
Full-stack backend infrastructure for AI agents: Postgres, auth, storage, serverless functions and atomic deploys. TypeScript SDK, CLI and MCP server. Paid with x402/MPP. Open source.
Cognitive Loop Runtime for Kubernetes
MonoLight is a General Autonomous Agent runtime with IM interaction as its primary interaction modality, focusing on safe execution and human-machine collaboration.
BranchContext gives AI agents and automated workflows copy-on-write branching over filesystems and processes
Governed API and agent runtime for desktop, VS Code, and CLI. Secure API testing, mocking, automation, secret protection, and controlled AI tool access.
Run Claude Code (or another coding agent) unattended, inside a disposable microVM, with its work committed to git so you can review everything from your host.
Python SDK for Fly.io Sprites: computers for agents. Sprite management, remote command execution, filesystems, checkpoints, services, and network policy from Python.
Run claude with rootless podman
An agent / harness multiplexer that allows you to access your agents from any device and machine
A hidden Omarchy desktop for your AI agents: they launch, drive and screenshot apps and shell plugins there, never on yours.
Run a coding agent in a staged workspace, review its file changes, and apply only what you keep.
A cross-device control hub for OpenClaw with guided setup, connector onboarding, mobile pairing, and visible diagnostics.
A modern solution for running untrusted code in Node.js
Checkpoint everything git misses in AI agent workspaces. Save, fork, push to any OCI registry.
The Armored AI Agent — visual dashboard, sandboxed execution, multi-provider LLM support. See what your bots do.
A systems language designed so an AI agent can safely write and run backend code: proven no GC/races/overflow, GBNF-constrained generation, structured diagnostics, OS-process sandboxing.
PicoAide Harness:企业级 DeepSeek Harness 一体化平台。桌面客户端 + 本地智能体引擎 + 管理后台,支持私有化部署。
✅ Agentic OS-aware Intention Programming Technology
Run claude with rootless podman
Forkable code execution server for LLMs, agents, and devs
An operating system for companies: AI agents build and run software tailored to your company. Own your software instead of paying for services you barely use. Agents work in sandboxes, a person approves the important steps, and every run is recorded.
Open-source, self-hostable server and runtime core for Run402 — backend infrastructure for AI agents. Postgres/PostgREST with RLS, storage, routing, functions and portable project archives. Apache-2.0.
Computer for AI Agents
AegisTrace - verifiable execution for durable AI agents. Compile policy-bound workflows, record every effect, replay without the network, and prove individual events.
World's first agentic linux distro.
safely extend poke's capabilities to your machine with an isolated docker environment. poke pc is a secure mcp worker that lets your ai agent run terminal commands and manage sessions within a sandboxed container
AI agent sandbox prison runtime with a local-first desktop app + CLI. Chat with any model (Claude, OpenAI, Groq, Ollama, LM Studio); every action passes Intent → Proposal → Commit through signed capability writs, risk-gated approvals, and a replayable hash-chained ledger. Watch it think in the Mind graph. Cognition proposes; the runtime governs.
Declarative, local-first runtime for bounded AI agents — define an agent in one manifest, run it headless, and audit every effect.
Run any AI agent — Claude Code, GAP, LangGraph DeepAgents — in a hermetic sandbox. Reference implementation of the Harness Protocol.
One isolated Lima VM per GitHub project — sandboxed Claude Code/Codex, Docker, Node, signed git
No Termux required to use AGENTCODI. No WebView shell. No separate gateway setup. with its own native sandbox in protected mode - With its own customized app server.
A sandbox-first JavaScript engine for AI agents: host-set capabilities and limits, deterministic runs, TypeScript annotations, test262 on every pull request.
Give Pi one typed tool instead of a toolbox.
Open desktop runtime for AI agents, tools, and integrations — MCP servers, sandboxed WASM tools, and permissioned add-ons in one workspace.
A macOS app for running parallel AI agents in sandboxed local VMs
The Agentuity Platform (SDK, tools and runtime)
User-friendly agentic Linux distribution based on Arch Linux.
The runtime substrate for agentic systems — one Go binary, six LLM providers, MCP-native, configurable as a managed sandbox or full agentic dev environment. Where agents live, talk, and learn.
Self-hosted sandboxes for AI agents: Docker, Firecracker or Apple Virtualization microVMs. Your real API keys never enter the sandbox. Part of Directionally Local.