ghcr.io/wiva-hq/coderail container (amd64 and arm64), a web UI on port 8080 behind HTTPS, with all state kept in ./data.On GitHub, someone published wiva-hq/CodeRail — an open-source project for remotely managing coding agents. The idea is simple: one Docker container on your own server, a 'bridge' on each work machine, and from there you open Claude Code and Codex chats, tool approvals, terminals, files, git, and dev-server previews from a browser or your phone. The repo has four commits as of publication — the project is brand new.
The mechanics in detail. You clone the repo, copy .env.example to .env, set SETUP_TOKEN, run docker compose up -d, then create the first admin at https://your-site/setup. You add a machine via 'Machines → Download for Mac' and a confirmation code — or with a one-liner on Linux: curl -fsSL https://your-site/install.sh | sh. The bridge keeps an outbound connection to the server (no open ports or VPN needed), launches local claude and codex via ACP adapters, and exposes terminals, files, git, and a browser to the outside. The server checks every request against a role — Owner, Operator, or Viewer, down to individual folders and agents — and writes an audit log. Under the hood: Bun + SQLite on the server, a single Bun binary in the bridge, Svelte 5 on the web, and a menubar app for macOS.
This looks like a hit on a real pain point rather than a trendy topic. An agent like Claude Code or Codex runs for minutes at a time, keeps asking for approvals, and the moment you step away from the laptop, the process stalls. CodeRail promises exactly what's missing: the session lives on with the browser closed, approvals arrive as a push notification on your phone, and history can be read offline. This isn't 'yet another chat with a model' — it's about the agent no longer demanding your physical presence at the desk.
What it changes — our assessment: the market for remotes like this is still empty. From experience, until now people have hand-assembled such setups from tmux, SSH, and tunnels, and it only worked for those on friendly terms with the terminal. Roles down to the folder level and an audit of every change — that's the language of teams, not lone wolves. That's exactly why I'd view CodeRail not as a hobbyist toy but as a bid for a place in a small studio's workflow.
Who wins: those who already run their own server and pay for Claude Code and Codex subscriptions — they get control without a middleman and without someone else's keys. Who loses: DIY SSH setups and closed 'cloud' wrappers around agents, where your code and logs go to a third party. The main risk here isn't technical but human: a bridge on every machine plus a server that stores the entire history of chats and commands.
The weak spot of this story is its youth: four commits, a README instead of an operational track record. Not a word about who is behind the wiva-hq handle, no data on terminal latency through the bridge, no independent audit report — 'signed and notarized builds for Mac' is, for now, just the authors' claim. The installation instructions only mention macOS and Linux; Windows support isn't mentioned in the text, although a mac/ folder does exist. And most importantly: the authors honestly state that there is no end-to-end encryption — the server sees everything. You should verify the security model yourself, using the SECURITY.md file.
Prediction: within a quarter, CodeRail will get Windows instructions and at least one third-party review of its security model — if the project doesn't die at the README stage. Easy to check via the commit history and the Releases section.
Source: GitHub — wiva-hq/CodeRail
Source: github.com · post in Telegram