tastyeffectco/sandboxd

Open-source, self-hosted AI app builder — an agent builds real apps in isolated sandboxes on your own server, each live at a preview URL. Self-host in one command. MIT.

sandboxd is an open-source platform for building apps by prompting an AI coding agent. When you send it a request, it spins up an isolated Docker container on your machine, runs a coding agent (OpenCode or Claude Code) inside that container against your prompt, and exposes the result at a live preview URL. Because it's self-hosted, your code, infra, and data stay on your server rather than on a hosted service like Bolt or Lovable.

It's designed to be lightweight: one Go binary drives Docker for isolation, Traefik handles URLs, and SQLite keeps state — no Kubernetes, external database, or message queue. Idle sandboxes sleep and wake on demand, so a single ordinary server can host many apps. It supports bring-your-own-models via OpenAI-compatible and Anthropic-compatible endpoints (including local servers, LiteLLM gateways, or AWS Bedrock routes),

Two interfaces are offered: a full REST API, or an optional web console where you can describe an app, watch an agent build it, edit files, and review git diffs in the browser. You can launch a ready-made open-source app like Ghost, n8n, or Gitea in one click, start from a scaffold, import a public Git repo, or build from scratch — then package the result for deployment. Installation requires Docker + git, and it runs on amd64 and arm64 (including Apple Silicon).

Related

  • Project
    opensandbox-group/OpenSandbox

    OpenSandbox provides isolated, secure environments on your own infrastructure for AI agents to execute code, run commands, and operate browsers or desktops.

  • Project
    docker/sbx-releases

    Docker Sandboxes (sbx) is a Docker‑based tool that creates isolated containers for AI agents, giving them controlled file‑system and network access so they can run autonomously without endangering the host. It ships as a native installer (Homebrew, WinGet, APT, etc.), supports pre‑release (RC/nightly) builds, and works with popular LLM CLIs like Claude Code, Codex, and Gemini.

  • Project
    newsnowlabs/dockside

    A self-hosted platform for launching isolated, browser-accessible development containers for every branch or AI agent session on your own infrastructure.

  • Dispatch
    Securing AI Agents with Isolated Docker Sandboxes

    An exploration of agent-sandbox, a tool designed to run AI coding agents within isolated Docker containers to prevent unauthorized system access and ensure execution safety.

  • Project
    finbarr/yolobox

    A secure containerized sandbox that lets AI coding agents run commands with full permissions while protecting your host home directory from accidental deletion or modification.