Boxes.dev: A Cloud-Native Agentic Development Environment for Claude Code and Codex

Boxes.dev is a cloud-only agentic development environment (ADE) designed to move AI-driven coding from local machines to dedicated cloud compute. By providing every agent thread with its own isolated cloud computer, filesystem, and compute resources, it removes the constraints of localhost development, such as the need for git worktrees and the requirement to keep laptops powered on for background agent tasks.

Solving the Constraints of Localhost Agentic Coding

Developing with AI agents on localhost often creates bottlenecks in resource management and workflow parallelization. Boxes.dev addresses these specific pain points:

  • Elimination of Git Worktrees: Instead of manually managing clunky worktrees to parallelize tasks, Boxes.dev allows each Claude Code or Codex thread to start from a snapshot of the full environment, providing an independent filesystem for each task.
  • Persistent Remote Execution: Agents can run continuously in the cloud, removing the need for developers to keep laptops open or maintain home servers (e.g., Mac minis) to prevent agent processes from stopping.
  • End-to-End Testing in Isolation: Because agents have their own dedicated cloud compute, they can run the full application and perform end-to-end testing in isolation without interfering with other active development threads.
  • Mobile-First Accessibility: The platform includes a fully-featured mobile app, treating coding as a communication task ("texting") rather than a tethered desktop activity.

Core Features and Integration

Boxes.dev is designed to mirror the user experience of Claude Code and Codex while adding cloud orchestration layers:

  • Automated Porting: The platform uses coding agents to scan a user's local development setup and automatically port it to the cloud.
  • Subscription Integration: Users continue to use their existing subscriptions for Codex and Claude Code.
  • Extended Tooling: Beyond the core environment, the platform offers scheduled automations and a Slack integration to streamline agent workflows.

Technical Considerations and Community Feedback

Following its introduction, the developer community raised several technical and strategic questions regarding the shift to cloud-native ADEs:

Security and Credentials

Users have questioned the security model, specifically how the platform handles Model Context Protocol (MCP) credentials. Concerns include whether secrets for stdio servers (reading from local config) or HTTP servers (using OAuth tokens) are stored within the Boxes.dev cloud infrastructure.

Defensibility and Model Agnosticism

There is ongoing discussion regarding the platform's defensibility as AI labs like Anthropic and OpenAI move "down the stack" toward remote execution. Some users suggested that a model-agnostic platform supporting OpenCode or self-hosted models could provide a competitive advantage over being tied to frontier labs.

Compute and Orchestration

Critics of the cloud-only approach argue that local compute remains faster and more cost-effective for some developers. Common points of contention include:

  • Latency: The preference for the "instant-response" of a local laptop over a cloud VM.
  • Control: A desire for the ability to "bring your own compute" or run the platform on a private VPS rather than a proprietary cloud.
  • Cost: Questions regarding the "box-hours" pricing model and how costs are handled when agents are idle or waiting for user input.

Comparison to Alternative Workflows

Some developers currently use a hybrid approach, such as running Claude Code on a private VPS and accessing it via Termux or SSH on mobile, suggesting that the primary value of Boxes.dev lies in its integrated orchestration and UX rather than the mere availability of remote compute.

Sources