Agent frameworks & infrastructure

Agent sandboxes & runtimes

1,721 agents · most starred first

Agent sandboxes & runtimes, page 9 of 18

io-harness★ 7

An embeddable agent runtime for Rust. Any task, any provider, in your process — with a permission boundary, a sandbox, and a durable trace you own.

Celestea-Harness-Studio★ 7

Self-hosted AI agent harness: web UI + HTTP API, sandboxed tool execution, persistent workspace memory, selection quoting, browser and computer use. TypeScript, zero runtime deps, one command to start.

genpark-lmax-disruptor-ring-buffer-skill★ 7

Production-grade genpark-lmax-disruptor-ring-buffer-skill skill for AI agents

genpark-michael-scott-lockfree-queue-skill★ 7

Production-grade genpark-michael-scott-lockfree-queue-skill skill for AI agents

helix★ 7

Helix — reference runtime for the Agent Execution Protocol (AEP). Source-available under the Fair Source License.

iram-os★ 7

🌌 IRAM OS — Turnkey agentic Hyprland desktop operating system with autonomous AI toolchain, 22 themes, Waybar, and live ISO installer.

agentkit★ 7

👷 Docker-for-agents: build agents into runnable container images

PAODO-Workspaces★ 7

Open-source platform for building your agentic org chart: each agent gets its own sandboxed Docker workspace, tools, HTTP API, and MCP endpoint.

harnis★ 7

Run AI agents safely across your company: human approvals, sandboxed code, tenant isolation.

genpark-jemalloc-slab-arena-allocator-skill★ 7

jemalloc-style high concurrency slab allocator with multi-size classes, bitmap run tracking, arena partitioning, and dirty page decay management.

algen-agent-runtime★ 7

Provider-neutral Python runtime for governed AI agents: approvals, policies, budgets, tenant isolation, audit and OpenTelemetry in the execution path. Works alongside LangGraph and the OpenAI Agents SDK.

openclaw-firecracker★ 6

🦞 OpenClaw Pool — multi-tenant AI agent platform on AWS EC2 Firecracker microVMs with per-tenant isolation, auto-scaling, and web management console.

pandora_app★ 6

Streamlit interface for the pandora_ai GPT4 python console package

silent.worm★ 6

ai; with a little sandbox for them to play.

projectdavid-core★ 6

Self-hostable LLM agent runtime

zeptopm★ 6

⚡ Process Manager for AI agents — like PM2, but for LLMs. Erlang-inspired process supervision, real-time agent channels, orchestrated multi-agent runs. Single binary, ~7 MB per agent.

harbour★ 6

🤖 Run agents across your repositories in a controlled, shareable environment

saifctl★ 6

Safety harness for autonomous AI agents: Spec-driven AI factory. Use with any agentic CLI. Language-agnostic. Safe by design.

AlfredOS★ 6

An open-source, self-hostable, multi-user, multi-persona, security-hardened agentic OS.

jean-dockerized★ 6

A full AI coding environment: Jean Claude Van Docker 💪

agent-os★ 6

DesireCore Agent OS. An open, delegation-based AgentOS (Agent Operating System) — a teachable, file-based AI companion desktop app you can grow.

the-hood★ 6

A local, provider-neutral runtime for governed software agent loops.

agentvfs★ 6

Workspace runtime and execution boundary for AI agents

praxovela-open★ 6

PRAXOVELA (formerly AxisRobo Agent): a governable, local-first desktop agent runtime for AI execution. Deny by default, sandbox-first, everything audited. This public repo hosts the Apache-2.0 integration surface: API docs, protocols, examples & build tooling.

agent-kit★ 6

Production Ready AI agents: curated memory, human-gated learning, and sandboxed multi-tenant execution.

agentplaza★ 6

Self-hosted AI agents, each in its own microVM, with rules, budgets and a kill switch you control.

microclaw★ 5

Sensor-level micro AI Agent in C/Rust 鈥?<1MB RAM, runs on $2-5 MCU hardware like ESP32. The smallest member of the Claw family.

nemoclaw-openclaw-sandbox★ 5

Operational documentation and repository scaffolding for deploying a `NemoClaw` sandbox with `OpenClaw` on a cloud service or VPS

openclaw-vps-deploy★ 5

Automated OpenClaw deployment to Ubuntu VPS. One-command setup with optional pre-configuration for Telegram + auth, or install-only mode for self-setup.

aetherion★ 5

Run all of your tools in containers (Linux/Mac).

manor★ 5

Local agent harness and Docker environment. Butler keeps Codex work supervised.

sand★ 5

Sand is a Swift CLI for Apple silicon Macs. It creates small named Linux environments. Each one has its own tools, shell state, package cache, and Pi login. The Host Mac only exposes folders you explicitly allow.

ag2-ext-yepcode★ 5

Enables AG2 (formerly Autogen) agents to securely execute code in isolated remote sandboxes using YepCode’s serverless runtime.

amber★ 5

Capability-based compiler/runner for reproducible agent scenarios

another-ai-sandbox★ 5

Rootless Podman containers to run AI coding agents (Claude Code, Codex, Cursor Agent) in hardened, read-only sandboxes.

aiOS★ 5

Kernel contract for the Life Agent OS — canonical types, runtime ports, event taxonomy, policy gates. The genome that all subsystems implement.

drydock★ 5

A defense-in-depth sandbox for AI coding agents. Containerized workspace, host Docker socket access, isolated memory and config from the host. Multi-project, single-image.

Hive★ 5

Local-first agent OS. Spawn persistent AI agents that collaborate, write code, and use tools autonomously. Multi-model (Claude, Codex, LM Studio). Config-driven with pre-built agent presets.

harness-kernel★ 5

A small TypeScript kernel for app-owned AI agent runtimes.

EmbedClaw★ 5

Run a modular agent runtime on ESP32-S3 that manages LLMs, tools, memory, and channels for efficient message processing.

arli★ 5

Production harness between LLMs and the real world. Kernel sandbox that cannot escape. Verification pipeline before delivery. On-chain attestation via ENSO/ICP. Fault-tolerant execution with rollback. Self-healing memory. 36 providers. 20 platforms. Live trading. Rust.

mission-control★ 5

Your own self-hosted AI agent OS

core-backend★ 5

Heavy-lifting brain: REST API, PostgreSQL-backed job queue, and sandbox pod orchestration (Rust).

falcone★ 5

Self-hosted, multi-tenant Backend-as-a-Service for the AI era — Postgres, document & object storage, identity, serverless functions, event streaming, realtime, durable workflows, and MCP server hosting, all from a single Helm chart for Kubernetes/OpenShift.

zeptocapsule★ 5

Isolation sandbox for AI agents — process, namespace, and Firecracker capsules

ContextSDK★ 5

Portable encrypted context state for AI sandboxes and VMs

helmdeck★ 5

A self-hosted, containerized platform for AI agents, exposed as Capability Packs — schema-validated, one-shot JSON tools — and native MCP. The defining metric is ≥90% pack success on 7B–30B-class open-weight models, something no frontier-targeting competitor is optimizing for.

executor-cloudflare★ 5

Deploy Executor (open-source agent integration layer) to your own Cloudflare account in one command, private behind Cloudflare Access. Built with Alchemy.

candor-ai★ 5

Lawful Good Rust Agentic Operating System — production-grade agent harness with WASM sandboxing, heterogeneous inference, SurrealDB memory, sentinel guardrails.

contAIn★ 5

Container based opencode setup for strict agent isolation

devcell★ 5

Local-first sandbox for AI coding agents. Run Claude Code, Codex, or Gemini in an isolated container or VM with zero permission prompts. No account required.

mirabilis★ 5

Run Claude Code fully autonomously in an isolated Docker sandbox — cross-platform dev container (macOS / Linux / Windows WSL2) with the neuro-matrix harness, persistent memory, and open egress.

Qyvaria★ 5

Software still in Open Beta, Internet browser fully working, Most thing work now and still patching it up. Please give stars and share.

flynn★ 5

A sandboxed, full-featured agent operating system in a single Go binary. Bring any model, manage local models, point it at a goal, and grant it real authority: every action is sandboxed, governed, and sealed into a verifiable, tamper-evident record an independent party can check. Runs interactive or 24/7, or embed it in your own system.

cash-claw★ 5

Local-first OpenClaw agent runtime with REST/WebSocket APIs, operator controls, 58 tests, and secure reproducible installs.

leash★ 5

Sandbox to run AI coding agents (Claude Code, Codex, Gemini, Grok) with full TLS inspection, domain allowlist enforcement, and Falco syscall monitoring

tank-agent-os★ 5

Run an autonomous AI coding agent — opencode, claw-code, or Claude Code — under controls it cannot disable. Fedora bootc + rootless Podman appliance with audited egress proxy, kernel-enforced nftables isolation, root-owned prompt-injection-hardened instructions, and service-gator-mediated external access.

furrow★ 5

Byte-exact, copy-on-write forks of your whole workspace, so coding agents run in parallel on one machine or on any other, with zero merge conflicts.

clawseed★ 5

Rust runtime for mobile and edge AI agents with remote tool execution over WebSocket.

littrs★ 5

🏖️🛡️ Keep your LLM's 💩 code where it belongs — in a sandbox.

runspace-agent★ 5

Run AI agents fully autonomously on a filesystem directory — MCP servers and skills enabled, with zero human approvals or permission prompts — safely isolated in a hardened Docker container (or locally).

ag2b★ 5

Client-side runtime for AI agents on the web.

agent-sandbox★ 5

Docker-based sandbox that runs AI coding agents with no git, no SSH, and can only access project folder. Built because permission prompts aren't real protection — you end up accepting everything by reflex, that's not a decision, it's fatigue. pi and opencode built-in. Extendable to use other agents running a single register command.

hostel★ 5

An agent-native sandbox runtime: run many isolated sandboxes (beds) from one process, with an OpenSandbox-compatible HTTP API.

crewship★ 5

Self-hosted runtime for AI coding agents. Real Linux containers, your hardware, your keys, your data.

software-factory★ 5

A secure, autonomous software factory: humans write specs, sandboxed untrusted agents build and verify, the trusted layer merges.

blueclaw★ 5

A POSIX-isolated agent host: a Go daemon that runs an AI agent harness on behalf of the person who asked, executes every tool call as that person's own unprivileged Linux user, holds side-effecting calls at an approval gate, and writes every step to a durable event ledger.

claude-win-sandbox★ 5

Isolation for Claude Code on native Windows with MSVC, no docker/WSL.

devkit★ 5

The project allows you to manage isolated containers with AI agents

vonzio★ 5

The runtime for production agents — bring your own model. Open-source, self-hostable, isolated Docker workspaces per session.

dsh-tensorlake-sandbox★ 5

A deepseek harness plugin for tensorlake sandbox

clum★ 5

Secure infrastructure for AI agents operating Linux hosts — persistent terminal sessions powered by rmux, full-chain audit logging, MCP-native interface for all major AI clients, with file transfer and multi-host orchestration.

keel★ 5

A local-first, governed agent harness: high autonomy inside structurally-enforced boundaries (Autopilot, not YOLO).

cma-blaxel-sandbox★ 5

Self-host Claude Managed Agents tool execution on Blaxel sandboxes.

Sluice★ 5

Run any project or coding agent in a sandboxed, egress-firewalled, non-root container. Drop in a config, run sluice.

agentweaver★ 5

Run an AI agent on a task inside a sandboxed git worktree, stream every step live, and gate merges behind human review — .NET + React, AKS-ready, with an MCP server.

AgentOS★ 5

Runtime infrastructure for AI agents — lifecycle, supervision, secrets, and deterministic time-travel replay. Built in Rust.

sovereign-agentic-os★ 5

Sovereign Agentic OS — self-hostable, EU-sovereign agentic data platform on Kubernetes (pre-beta). One Helm chart, runs in your own cluster.

skills-rce★ 5

Code execution service for AI agent skills. Runs in isolated subprocesses with skill caching.

agentos-docker★ 5

AgentOS: FastAPI for Agents

semaframe★ 5

A semantic spatial workspace for AI agents.

vmette★ 5

A hardware-isolated microVM sandbox for running untrusted local AI agents on macOS.

horsie★ 5

An open-source managed agent harness: bring your own key and runtime.

aether-core★ 5

Embeddable agent runtime and workbench for teams that need sandboxed execution, host integration, platform governance, and reusable AI infrastructure.

tortugalinux★ 5

Fedora Kinoite 45 based Linux distro with agentic AI features.

hermes-mobile★ 5

Hermes Agent (Android) — 通过 proot + Ubuntu rootfs 在 Android 上运行 Hermes Agent,内置 WebUI 仪表盘。无需 root。

OpenSquilla-mobile★ 5

OpenSquilla Android: Chaquopy + FastAPI gateway packaged as one APK with embedded Vue WebView UI

clawdaemons★ 5

The agent OS. Host AI agents 24/7 across Hermes and Deer Flow. 🦌🪽

asf★ 5

ASF. A lightweight framework for running AI agents in constrained, policy-controlled environments. The agent gets only what you explicitly allow.

shellclaw★ 5

🦞 The first physical agentic harness that participates in a global marketplace, using ASAP protocol.

microagent★ 5

Toolkit and libraries for running AI agents inside microVMs

dsh-launcher★ 5

deepseek-harness 启动器与运行环境管理器(Tauri 2 + Rust + React)

pi-sprites★ 5

Pi coding agent extension for Fly.io Sprites: computers for agents. Routes Pi's native filesystem and shell tools into persistent Sprites, with checkpoints, services, worker pools, and RPC hosting.

wardyn★ 5

Open-source governed-sandbox control plane for any workload — per-run identity, brokered credentials, layered egress, approvals, and an append-only audit. Coding agents are the flagship use. Apache-2.0, self-hosted.

code-os★ 5

Secure command center and gateway for agent-ready development environments

rust-nano-vm★ 5

Single-binary Rust microVM for AI-agent code execution. ~12 ms cold start, ~0.5 MiB Pss per fork.

agenhood★ 5

Self-hosted, open-source platform for running autonomous AI agents 24/7 with sandboxed Docker containers, persistent memory & workspaces, task scheduling, workflows, and live streaming. Pluggable brains: Claude Code, Codex, opencode. Your server, your data.

OpenIntentOS★ 5

Open-source Intent Operating System

DuDuClaw-OS★ 5

A Yocto-built Linux desktop OS where DuDuClaw AI agents and people share one x86-64 machine: A/B updates with rollback, 10 coding CLIs built in, offline llama.cpp. Pre-GA, tested in QEMU.

TAP-Runtime★ 5

Run your coding agent's repeated work as small sandboxed programs that borrow its tools and ask before changing anything. Includes tap discover, which finds those procedures in your own agent history, locally.

More in Agent frameworks & infrastructure