Show HN: Yado – Claude Code and Codex on Your iPhone

What happened

A solo developer has shipped Yado, an iOS app that puts Claude Code and OpenAI Codex on your iPhone without a laptop anywhere in the picture. The app went live on Hacker News under a Show HN post and is priced at $19 per month with a seven-day free trial.

This piece breaks down how Yado is architected, what it costs, who stands to benefit most from it, and what open questions remain for developers considering it as a primary coding environment. It is aimed at engineers who already pay for Claude or ChatGPT subscriptions and want to understand whether a mobile-first AI coding setup is actually viable day to day.

What happened

How it works

A developer released Yado, a standalone iOS application that connects Claude Code and OpenAI Codex to a dedicated Linux machine hosted on Fly.io. Each user gets their own Fly.io instance, so the compute is isolated rather than shared. The app is available now at $19 per month, and new users can try it free for seven days before the billing cycle begins. The developer announced it through a Show HN post, describing the project as a way to run AI coding agents from a phone with no secondary device required. The announcement did not specify the underlying Linux distribution, the amount of RAM or storage provisioned per machine, or how long idle machines remain active before being paused or recycled.

How it works

Why it matters

Yado communicates with Claude Code using its native stream-json protocol, which is carried over a WebSocket relay between the iPhone and the Fly.io machine. Because the relay sits between the phone and the Linux box, the agent session continues running even when the phone screen locks or the app moves to the background. When the agent finishes a turn, Yado sends a push notification to the device so the user can check results without keeping the screen on. The app also supports connecting Model Context Protocol servers and CLI tools, which means developers can extend the agent's reach to external APIs or local scripts running on the Fly.io machine. Live previews are included, though the announcement did not detail the mechanism — whether that is a browser preview over a tunnel, a port-forwarding arrangement, or a rendered output within the app itself.

Why it matters

The project is a direct challenge to the assumption that serious AI-assisted coding requires a laptop or desktop. Claude Code and Codex are sophisticated agents that can write, edit, test, and iterate on code autonomously over multiple steps, and Yado tries to make those full loops available on a device most developers carry everywhere. The WebSocket relay design is particularly notable because it decouples agent execution from device availability, which is a meaningful architectural choice rather than a simple port of a desktop tool. The pricing is also structured to avoid a second AI subscription: Yado uses the customer's own Claude or ChatGPT subscription for model access and charges separately only for the infrastructure and app layer.

Who is affected

Who is affected

The most immediate audience is developers who already hold active Claude or ChatGPT subscriptions and want to extend those tools beyond their desk. Freelancers working across locations, engineers who review and iterate on code between meetings, and hobbyist builders who do not maintain a dedicated development machine are reasonable early adopters. The $19 per month fee is additive to whatever Anthropic or OpenAI already charges, so the total cost depends on which subscription tier a user holds. Teams running shared Codex or Claude Code workflows would need to evaluate whether per-user Fly.io isolation fits their collaboration model, since the announcement describes individual machines rather than any shared or organization-level account structure.

What to watch next

What to watch next

Several details left unaddressed in the launch post will matter for builders evaluating Yado seriously. The provisioned specs of the Fly.io machine — CPU count, RAM, and disk size — determine whether it can handle large repositories, dependency installations, or long build cycles without throttling. The persistence model for the machine also needs clarification: whether the Linux environment survives between sessions or is recreated on each connection has a direct effect on workflow continuity. MCP and CLI integration is advertised as easy, but the specifics of how those connections are authenticated and whether they survive the WebSocket relay during background execution are worth testing. Developers should also verify how Yado handles Claude Code or Codex API rate limits and errors on the relay side, since a phone notification arriving for a failed turn rather than a completed one would degrade the core use case.