Topics / #agent-sandbox

Agent-sandbox projects on GitHub

Isolated execution environments that let AI agents run code and tools safely.

As of , the most-starred Agent-sandbox projects on GitHub are AstrBot (42K stars), NemoClaw (22.8K stars) and teable (21.9K stars). GitHub.Locker tracks 16 Agent-sandbox repositories with 139.7K stars combined. The fastest-growing on October 16 was SPACE, with +529 stars in a day.

16Projects
139.7KStars in total
+705Stars in a day
Browse in the catalog →github.com/topics/agent-sandbox ↗

Hot right now

Most stars gained in a day · Oct 16

+529

SPACE

perplexityai/SPACE · ★ 1.1K

SPACE (Sandboxed Platform for Agentic Code Execution) is a sandbox platform that runs agent and user code inside isolated Firecracker microVMs, controlled through an HTTP API.

+62

AstrBot

AstrBotDevs/AstrBot · ★ 42K

AstrBot is an open-source all-in-one AI agent chatbot platform that connects LLMs to messaging apps like QQ, Telegram, Discord, Slack, WeChat and more, so you can build a personal AI companion, customer service bot, or enterprise knowledge-base assistant inside your existing chat workflows.

+27

crm

trycompai/crm · ★ 11.4K

Comp AI CRM is an open source, agentic-first CRM: instead of a database with a chat box bolted on, the AI agent is a standalone deployment that runs its own schedule and research queue, and the CRM is just where it writes down what it finds.

+26

nono

nolabs-ai/nono · ★ 4.5K

nono is a zero-setup, zero-latency sandbox that runs AI coding agents (Claude Code, Codex, OpenCode, etc.) with least-privilege access, so they can't touch your SSH keys, cloud credentials, or the rest of your disk by default.

The all-time top

16 projects · by stars

  1. #1 · AstrBotDevs/AstrBot

    AstrBot

    AstrBot is an open-source all-in-one AI agent chatbot platform that connects LLMs to messaging apps like QQ, Telegram, Discord, Slack, WeChat and more, so you can build a personal AI companion, customer service bot, or enterprise knowledge-base assistant inside your existing chat workflows.

    42K
  2. #2 · NVIDIA/NemoClaw

    NemoClaw

    An open source reference stack from NVIDIA for running AI agents like OpenClaw, Hermes and LangChain Deep Agents more safely inside NVIDIA OpenShell sandboxes, with managed inference, network policy and lifecycle operations via its CLI.

    22.8K
  3. #3 · teableio/teable

    teable

    An AI spreadsheet for business — spreadsheet-simple on the surface, a real PostgreSQL database underneath, with AI chat, an app builder, and automations on top of your live data.

    21.9K
  1. 4

    crm

    11.4Ktrycompai/crm · TypeScript

    Comp AI CRM is an open source, agentic-first CRM: instead of a database with a chat box bolted on, the AI agent is a standalone deployment that runs its own schedule and research queue, and the CRM is just where it writes down what it finds.

  2. 5

    computer

    9.6Kcloudflare/computer · TypeScript

    Cloudflare Computer gives an AI agent a persistent computer: a virtual filesystem backed by a Durable Object's SQLite storage, with pluggable backends that actually execute shell commands or JavaScript against that filesystem.

  3. 6

    capsule-fs

    8.4Kunicity-aos/capsule-fs · Rust

    capsule-fs is a filesystem toolkit for AI agents inside Unicity AOS (also called Astrid OS): it gives agents readfile, writefile, replaceinfile, listdirectory, grepsearch, createdirectory, deletefile and movefile tools, all mediated by the OS kernel's VFS airlock.

  4. 7

    smolvm

    6.8Ksmol-machines/smolvm · Rust

    smolvm is a CLI tool for running embeddable, portable, branchable virtual machines so you can safely run AI agents and untrusted code locally, with sub-second cold start on macOS, Linux, and Windows.

  5. 8

    sandbox-runtime

    5.6Kanthropics/sandbox-runtime · TypeScript

    srt (Anthropic Sandbox Runtime) is a lightweight sandboxing tool that enforces filesystem and network restrictions on arbitrary processes at the OS level without needing a container, built to safely sandbox AI agents, local MCP servers, and bash commands.

  6. 9

    nono

    4.5Knolabs-ai/nono · Rust

    nono is a zero-setup, zero-latency sandbox that runs AI coding agents (Claude Code, Codex, OpenCode, etc.) with least-privilege access, so they can't touch your SSH keys, cloud credentials, or the rest of your disk by default.

  7. 10

    mirage

    3.7Kstrukto-ai/mirage · TypeScript

    Mirage is a virtual terminal for AI agents that mounts services like S3, Google Drive, Slack, Gmail and Redis into one unified virtual filesystem, so agents access everything with familiar unix tools (ls, grep, find, jq), run virtual CLIs, route code to different runtimes, and stay inside a policy-controlled shell.

  8. 11

    SPACE

    1.1Kperplexityai/SPACE · Rust

    SPACE (Sandboxed Platform for Agentic Code Execution) is a sandbox platform that runs agent and user code inside isolated Firecracker microVMs, controlled through an HTTP API.

  9. 12

    ai-newtab

    824ThariqS/ai-newtab · TypeScript

    AI Homepage is a Chrome extension that replaces your new-tab page with one a Claude agent writes for you, based on sites it finds in your own browsing history.

  10. 13

    agentbox

    538madarco/agentbox · TypeScript

    AgentBox is a CLI that runs multiple coding agents in parallel, each in its own sandboxed VM, with a single command — on your PC, a self-hosted server, or in the cloud.

  11. 14

    quicksand

    218microsoft/quicksand · Python

    Quicksand is an async Python API for launching, controlling, and snapshotting QEMU virtual machines, built specifically for sandboxing AI agents.

  12. 15

    ApexAgents-SkyRL-Recipe

    217Mercor-Intelligence/ApexAgents-SkyRL-Recipe · Python

    It's a complete, public recipe for RL-training knowledge-work agents on Mercor's APEX-Agents data, built on the open-source SkyRL training library and the open-source harbor sandbox/grading toolkit. It's the companion code for Mercor's 'Training Frontier Knowledge Work Agents' blogpost.

  13. 16

    arc-code

    143jerber/arc-code · Python

    arc-code is a research harness that lets Claude Code (or Codex) play the ARC-AGI-3 benchmark games through a shell and filesystem, writing its own parsers, simulators and search code for each game instead of relying on built-in ARC-specific tooling.

Often tagged together