Boxes.dev logo

Boxes.dev

Run Claude Code and Codex in your own cloud environment

2026-06-04

Product Introduction

  1. Definition: Boxes.dev is a cloud development environment (ADE) and remote workstation platform specifically engineered for running AI coding agents like Claude Code and Codex. It functions as a managed service that provides each agent with a dedicated, persistent Linux virtual machine (cloud devbox) in the cloud, complete with a full developer environment, filesystem, and networking stack.
  2. Core Value Proposition: The core value proposition of Boxes.dev is to unleash the potential of parallel AI agents by eliminating local hardware constraints and ensuring continuous operation. It solves the critical limitation of local development—the single machine bottleneck—by providing isolated, always-on cloud computers for every coding thread, enabling developers to build with coding agents 24/7 from any device, drastically accelerating development cycles.

Main Features

  1. Parallel Agent Isolation via Forked Devboxes: Boxes.dev uses a VM snapshot and fork system. After an initial dvb init scans and migrates a local project's dev environment to a primary cloud devbox, any new agent task (e.g., a Codex or Claude Code chat) spawns a "fork"—a complete, independent copy of the primary VM's filesystem, database state, and running services. This provides true isolation; each agent operates on its own full computer, not a shared worktree or container, preventing conflicts over ports, dependencies, or system state. This is the key technology enabling scalable parallel coding agents.
  2. Continuous Cloud Execution & Mobile Control: Devboxes and their agent forks persist in the cloud independently of the user's client device. Developers can close their laptop, disconnect, or sleep, and the agents continue processing tasks. The platform offers native applications for desktop (macOS), mobile (iOS, Android), and a CLI, allowing users to monitor progress, review diffs, answer agent questions, approve requests, and commit code directly from a phone or tablet. This enables remote devbox management and breaks the tether to a physical workstation.
  3. Environment & Port Forwarding with Localhost Integration: Each forked devbox retains the full development stack (services, local databases, APIs). Boxes.dev provides secure port forwarding from the cloud devbox to localhost on the client machine. This allows developers to manually test an agent's changes by accessing localhost:3000 (or any port) while reviewing the specific fork, creating a seamless bridge between cloud agent work and local verification. The system supports persistent terminal sessions that survive disconnects within the web and desktop workbench.

Problems Solved

  1. Pain Point: The Single-Machine Bottleneck & Resource Contention. Traditional development, even with worktrees, limits concurrent AI agents due to shared local CPU, memory, RAM, and I/O. Running multiple agents simultaneously can cause system slowdowns, crashes, and port conflicts, severely hampering productivity and making parallel AI coding impractical.
  2. Pain Point: Context Switching & Workflow Fragmentation. Developers constantly switch between coding, reviewing agent output, and managing their environment. The lack of a unified interface for launching agents, inspecting changes, and communicating with AI from any device leads to workflow fragmentation and wasted time on tool-switching rather than productive work.
  3. Pain Point: Idle Agent Downtime & Device Dependency. AI coding agents often require long runtimes for complex tasks. If tied to a local machine, the process halts when the laptop is closed for travel, meetings, or sleep, leading to significant lost development time and an inability to leverage overnight or background processing.
  4. Target Audience: The primary target audience includes AI-assisted software developers, full-stack engineers, tech leads, and small engineering teams who are actively using or experimenting with LLM-based coding assistants (Claude Code, Codex) and need to scale their usage effectively. It also appeals to independent developers and founders running side projects who need to maximize limited time with powerful automation.
  5. Use Cases:
    • Parallel Feature Development: A team spins up 10 forks from a main devbox to simultaneously work on new features, bug fixes, and refactors, each in an isolated environment.
    • Overnight Batch Processing: A developer queues multiple complex coding tasks (e.g., "refactor authentication module," "upgrade all dependencies") before bed, letting agents run unattended until morning.
    • Safe Testing & Experimentation: Each agent operates in a sandboxed environment. Breaking changes, dependency conflicts, or database modifications in one fork do not affect the main development line or other active forks.
    • Agile Code Review & Iteration: A tech lead uses the mobile app to review an agent-generated pull request diff, leave a comment on a specific line, and request modifications while away from their desk.

Unique Advantages

  1. Differentiation vs. Traditional Methods & Competitors: Boxes.dev fundamentally differs from Git worktrees by providing a complete computer (CPU, memory, filesystem, network) per task, not just a checkout. Unlike generic cloud IDEs (e.g., Gitpod, Codespaces) focused on the human-developer loop, Boxes.dev is architected from the ground up for the AI agent workflow—prioritizing isolation, parallelism, and background execution. Compared to other agent orchestration tools, it provides the actual execution environment (the full cloud VM) rather than just managing API calls.
  2. Key Innovation: The "Fork" Model for Agent Workloads. The cornerstone innovation is the VM-level snapshot and forking mechanism. This allows for the instantaneous creation of a perfect, stateful clone of a complex development environment, which is essential for agents that need to interact with services, run tests, and modify databases. This model ensures that each agent's work is both isolated and reflective of a real deployment environment, leading to higher quality, reviewable output and enabling true, unconstrained parallel development.
  3. Integrated Multi-Surface Control Plane: The unified experience across CLI, Desktop App, and Mobile Apps provides a seamless control plane for agent work. The ability to initiate, monitor, and refine agent tasks from a phone without compromise is a significant advantage for developer mobility and responsiveness.

Frequently Asked Questions (FAQ)

  1. How is Boxes.dev different from using git worktrees for parallel development? A git worktree provides an additional checkout of a repository on the same machine, sharing system resources and potentially conflicting on ports or services. A Boxes.dev fork is a full, independent cloud computer (a Linux VM) with its own dedicated CPU, memory, isolated filesystem, ports, and database state. This allows true parallel execution without resource contention or interference.
  2. Does Boxes.dev sell me AI tokens or an LLM subscription? No. Boxes.dev does not sell AI tokens. It provides the cloud execution environment for agents. You use your existing Codex or Claude Code subscription or API key. Boxes.dev charges only for the compute time (box-hours) the cloud devboxes and forks are awake and running.
  3. Can I use my own editor like VS Code or JetBrains IDE with Boxes.dev? Yes. While Boxes.dev provides a powerful integrated web and desktop workbench for reviewing diffs and managing agents, it supports connecting to the devbox's port-forwarded localhost from any local editor or IDE. You can attach a local VS Code instance via Remote - SSH or simply open the forwarded port in your browser.
  4. What happens to my running agents if I close my laptop? The agents and devboxes keep working uninterrupted in the cloud. Boxes.dev is designed for this continuous operation. You can close your laptop, go offline, or sleep, and the agents will complete their tasks. You can check the progress and results later from the desktop app, web dashboard, or mobile device.
  5. How does the pricing model work? Pricing is based on a simple box-hour metric. One box-hour equals one 4 vCPU / 8 GiB devbox being awake for one hour. Idle forks automatically sleep to conserve usage. Plans include a pool of free or included box-hours per month (e.g., 40 hours for the Starter plan), with prepaid overage available at a lower rate if you exceed the included quota. You can start for free with a trial that provides 10 box-hours.

Submit to 240+ Directories with 1-Click

Maximize your product's SEO and drive massive traffic by automatically submitting it to over 240 curated startup directories using DirSubmit.

Subscribe to Our Newsletter

Get weekly curated tool recommendations and stay updated with the latest product news