Direct Answer
A Hacker News thread titled “Run Claude and Codex in the Browser” is drawing attention to a simple idea: your coding agents already live in a terminal, so why should that terminal be chained to one desk? New tools put that terminal behind an end-to-end encrypted link you can open in any browser — same session, same context, even from your phone. This is the browser-remote trend for AI coding agents.
The agents themselves have not changed. Claude Code and Codex still run locally or in a sandbox. What is new is the surface: instead of SSH-ing in or keeping a laptop open, you hand the agent a task on your desktop and watch it work live from a browser tab or a mobile layout.
Why this is happening now
Three things converged:
- Agents are long-lived. A Claude Code or Codex session can run for minutes or hours on a real task. You do not want to babysit it from one machine.
- Context is expensive to lose. Re-establishing a working directory, the conversation, and the half-finished diff is friction. Keeping one session reachable from anywhere preserves that state.
- The terminal is the universal surface. Because these agents are terminal-first, any tool that can stream a terminal can host them — desktop app, browser, phone.
What a browser-remote setup looks like
Based on the tooling surfaced in the HN discussion, the pattern is consistent: a desktop app for your OS (macOS, Windows, Linux) that hosts your projects, sessions, and agents; a “remote access” toggle that produces an encrypted pairing link; and a browser (or mobile) view that mirrors the same workspace live. SSH remote login is often built in for connecting out to other machines.
The appeal is not just convenience. It is continuity: the same link, the same session, the same context as your desktop — so you can start a refactor at a desk and check on it from a phone without losing the thread.
Things to watch
- Encryption is the whole point. A pairing link that exposes your terminal should be end-to-end encrypted and password-gated. Treat any tool that skips this as a hard pass.
- It is still your machine. The agent runs where your files are. Browser access is a window, not a separate computer — permissions and filesystem access stay local.
- Latency matters for interactive edits. Streaming a terminal is light, but if you are pasting code back and forth, the round-trip shows.
For a deeper look at how the two agents behind these sessions actually differ, read our Claude Code vs Codex comparison, and if you are building your own setup, our best coding-agent setup after six months covers what holds up in daily use.
FAQ
Can I really run Claude Code from a phone? The browser-remote pattern is designed for exactly this: the same encrypted link switches to a mobile layout automatically, so you can open a session, hand it a task, and watch it work. You are not running a new agent — you are viewing the one already running on your desktop or server.
Is running agents in a browser safe? It is as safe as the encryption behind the link. A reputable setup uses end-to-end encryption plus a password, and the agent keeps running on your own machine — the browser is only a view. Avoid anything that uploads your workspace or exposes an unencrypted port.
Does this replace SSH? Not really — it complements it. SSH remote login is usually built in for connecting out to other machines, while the browser link is for quickly checking or driving a session from anywhere without opening a terminal client. Most setups give you both.
Your coding agent already makes you faster. aiFiesta makes your AI tools cheaper — $12/mo for 9+ premium models instead of $20/mo each.