AgentBox

What is AgentBox?

AgentBox is a command-line tool for running several AI coding agents side by side, each inside its own isolated and checkpointed VM. One command copies your project into a box on local Docker, a machine you reach over SSH, or a cloud provider such as Vercel, Hetzner, DigitalOcean or Daytona, then launches Claude Code, Codex, OpenCode or Pi there. Boxes carry over your shell and editor settings, agent logins, skills and plugins, and each has a browser you can screen-share, a persistent shell and a VS Code or Cursor session your local editor attaches to. An orchestrating agent can spawn subagents in separate boxes, wait for them, read their plan questions and answer them, while you control git per box from the host. Git credentials stay on your machine, and pushes go through a permission request. The site shows no pricing: it says local Docker and your own hardware need no account, token or bill, while cloud providers bill you directly.

Use cases

• Run coding agents in isolated VMs
• Work on several branches in parallel
• Orchestrate subagents from one agent
• Preview a web app from a remote box
• Run agents on your own server over SSH
• Pause idle boxes to save cost

Who it's for

• Developers
• Engineering teams
• AI-assisted coders

Why use AgentBox

• Each agent works in its own VM, so parallel tasks and risky commands do not touch your main machine.
• A single command carries over your project, agent logins, skills and settings, which removes the setup work of a remote dev box.
• You choose where boxes run, from local Docker to a Mac mini on your desk or a cloud server.
• Checkpoints and auto-pause keep restarts fast and idle cost down.
• Git credentials stay on your laptop, and pushing to a remote needs an explicit permission.

AgentBox features

• One-command teleport of a project into a VM
• Providers: local Docker, your own hardware, Vercel, Hetzner, DigitalOcean and Daytona
• Launches Claude Code, Codex, OpenCode or Pi in a detachable tmux session
• Checkpoints with sub-second startup and pause or resume
• Auto-pause when a box is idle
• Live browser with VNC screen sharing and local preview tunnels
• VS Code or Cursor attached over Dev Containers
• Queue and wait-for commands to orchestrate many agents
• Native macOS menu bar app
• agentbox.yaml for repeatable setups
• Per-box git control with permission requests for pushes

AgentBox integrations

• Claude Code
• Codex
• OpenCode
• Pi
• Vercel
• Hetzner
• DigitalOcean
• Daytona
• Docker
• VS Code
• Cursor
• iTerm2
• tmux
• cmux
• Notion
• Linear

How to get started with AgentBox

1. Run the CLI with npx @madarco/agentbox or install it, then pick a provider with agentbox install.
2. Create a box with agentbox create, or start an agent directly with a command like agentbox claude or agentbox codex.
3. AgentBox seeds the VM from your repository, provisions packages and Docker services, and starts the agent in a tmux session.
4. Open the box in your IDE, a shell, the VNC viewer or the menu bar app, and follow your app through a local preview URL.
5. Pause, checkpoint or destroy the box, and use queue and wait-for commands to coordinate several agents.

AgentBox limitations

• Vercel boxes are limited to the iad1 region, a fixed 32 GB disk and at most 4 exposed ports
• Vercel session length is capped at 45 minutes on Hobby and 5 hours on Pro and above
• A DigitalOcean Droplet bills by the hour even while stopped, at roughly $24/mo for an s-2vcpu-4gb box
• The first base-snapshot bake takes about 15 to 20 minutes on DigitalOcean

AgentBox pricing

We couldn't confirm pricing on the official site. Check agent-box.sh for current prices.

Prices can change and may vary by country or exclude tax. Check the official pricing page.

AgentBox alternatives

The main AgentBox alternatives listed on Debuly are E2B, Conductor, TaskHandoff, AG2, LangChain Deep Agents and Letta. The table compares pricing and platforms; order is A–Z.

1. 

   AG2

   Open-source multi-agent framework, formerly AutoGen, plus an AgentOS platform for coordinating and deploying teams of AI agents.

   Developer platformsAutomation and agentsCoding
   Open sourceVisit AG2 (official site)
2. 

   Conductor

   Enterprise platform for AI search (AEO) and SEO: content creation, visibility reporting, site monitoring and agent tooling.

   Marketing and SEOWritingResearch and search
   Contact salesVisit Conductor (official site)
3. 

   E2B

   Cloud sandboxes where AI agents run code safely, with Python and JavaScript SDKs and per-second usage billing.

   Developer platformsAutomation and agentsCoding
   FreemiumVisit E2B (official site)
4. 

   LangChain Deep Agents

   Open-source LangChain agent harness: long-term memory, a virtual filesystem, subagents and planning included

   Developer platformsAutomation and agentsCoding
   Open sourceVisit LangChain Deep Agents (official site)
5. 

   Letta

   Platform, formerly MemGPT, for building stateful AI agents with persistent memory, from a terminal app, SDK or API.

   Developer platformsAutomation and agentsCoding
   FreemiumVisit Letta (official site)
6. 

   TaskHandoff

   Self-hosted workbench that runs Codex and OpenCode AI sessions in isolated workspaces across local and remote machines.

   Developer platformsCodingAutomation and agents
   Visit TaskHandoff (official site)

AgentBox FAQ

Sources & verification

Facts on this page come from the official sources below, each with the date we read it. Written by Debuly; the vendor did not review this page.

Image credits

• Logo: from agent-box.sh · fetched 23 Oct 2026 · Vendor trademark, used only to identify the product; removed on request
• Cover: none collected yet, so a generated tile is shown.

Spot an error? Report an issue with this listing. We also remove a logo or image on the owner's request.