Why AgentPod: The Story Behind Portable AI Sandboxes

Drift note — 12 September 2026. This post is kept as the record of why AgentPod started, and it is nine months out of date about what AgentPod is. The argument below is still the honest origin; the product it describes is not the one you would install today.

What changed: there is no desktop app any more, and no mobile one — the client is a web console. OpenCode is not the foundation; it is one of six harnesses AgentPod detects, alongside Claude Code, Codex, Hermes, OpenClaw and Pi. And the thesis moved outward. This post argues for getting agents off your laptop and into a sandbox. The product became a console for the machines agents already live on — attach to what you are running, wherever it runs, rather than move it somewhere new.

Nothing below has been edited to look current. See docs.agentpod.dev for what is true now, and the newer post for how the one became the other.

The motivation behind AgentPod came from a clear need: portability with coding agents and a platform for experimenting with new UI paradigms.

The TUI Revolution

With the emergence of TUI-based coding agents like Claude Code, Gemini CLI, and more recently OpenCode (which powers this very project), we’re witnessing a revolution in how developers interact with AI. These terminal-based tools are incredibly powerful.

But here’s the problem: TUIs are difficult to work with on screens other than desktop.

Try using Claude Code on a tablet. Or running Gemini CLI from your phone. It’s nearly impossible. These powerful tools are trapped in desktop environments.

The Security Problem

There was another pressing issue: sandboxing.

AI agents that can run arbitrary commands on your primary laptop? That’s a security nightmare waiting to happen. We needed isolation. We needed containers. We needed a way to let agents do their work without risking our main system.

OpenCode Server: The Godsend

This is where OpenCode became a godsend.

OpenCode’s server mode allows running the agent in sandboxed containers while the client runs on your primary machine to interact with users. This separation of concerns is exactly what we needed:

• Server: Runs in Docker containers, isolated and sandboxed
• Client: Runs on any device — desktop, tablet, or phone
• Connection: Secure communication between client and server

What AgentPod Does

AgentPod takes this architecture and makes it accessible:

1. Spin up sandboxes — Docker containers or Cloudflare workers
2. Connect from anywhere — Desktop app now, mobile coming soon
3. Stay safe — Your laptop stays clean, agents stay isolated
4. Use any LLM — 75+ providers supported through OpenCode

The Vision

We’re building the “Portable Command Center” — your personal AI-powered development environment that travels with you.

Desktop support is complete. Mobile is in progress. The future is portable.

---

Thanks to the OpenCode team for building an incredible open-source agent platform. AgentPod wouldn’t exist without their work.