Urgent.News

What's breaking now, across thousands of outlets.

AI

I Built an Open-Source Remote Console to Control Codex, Claude Code, Antigravity & Grok From My Phone

If you run Codex , Claude Code , Antigravity , and Grok across multiple projects (or even multiple Codex accounts), two pains hit you fast: At your desk : Juggling 4+ IDE windows and terminals eats gigabytes of RAM just to check who is stuck waiting for permission. Away from your desk : You walk away for coffee or a commute and have no clean way to inspect diffs, approve tool calls, interrupt a…

If you operate Codex, Claude Code, Antigravity, and Grok across various projects or accounts, two challenges arise: at your workstation, managing multiple IDE windows and terminal sessions can consume substantial RAM, particularly when waiting for permission. When away from your workstation, you lack a straightforward method to review changes, approve tool calls, halt an infinite loop, or queue the next command from your mobile device.

Existing Codex bridge scripts typically support only a single engine or necessitate setting up Tailscale or chatbot configurations manually. To address these issues, I developed Agent Control—an open-source (AGPL-3.0), self-hosted desktop and mobile remote console with a built-in five-language interface (English, simplified Chinese, traditional Chinese, Japanese, and Korean). Website: https://hub.guuguo.top/agent-control GitHub: https://github.com/guuguo/agent-control

Security and privacy are paramount in remote-controlling local coding agents. Agent Control is built around stringent local boundaries: it only reads your local Codex, Claude, and Gemini history files in read-only mode, ensuring upstream databases remain unaltered. Your code and conversations never visit a third-party cloud database, with remote connections routing directly through your selected tunnel or self-hosted relay, safeguarded by one-time pairing codes and HTTP-only session cookies.

Agent Control resolves several key issues:

1. Multi-engine and multi-account quota management—automatically identifies multiple local Codex accounts, displaying per-account quotas alongside Claude Code, Antigravity, and Grok sessions, grouped by project workspace.

2. Mobile approvals and diff reviews—users can scan a QR code from their phone's browser to view live output, compare code changes, approve or reject tool calls, or queue subsequent actions.

3. Zero-configuration remote access—starts with a free Cloudflare Quick Tunnel, which can be swapped at any time for Tailscale Funnel, Cloudflare Named Tunnel, or a self-hosted Docker/Sealos Relay.

4. Hands-free voice assistance—includes a wake-locked voice mode that announces task completion or pending approvals and allows approving or dispatching tasks verbally using low-latency models like Claude Sonnet, GPT, Antigravity, or Grok.

For further details, consult the comprehensive documentation available in English, simplified Chinese, traditional Chinese, Japanese, and Korean, explore the security architecture, and install the tool with a single command line instruction. The project page, https://hub.guuguo.top/agent-control, and the GitHub repository, https://github.com/guuguo/agent-control, provide additional resources. Feedback and support are welcomed!

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

Build-Time AI: How We Validate 5,000 Prices Across 8 Languages Without a Server

The problem We run a transport booking site in Morocco — navetteforyou.com . Eight languages, 641 pages, thousands of prices. One wrong number and a customer books a ride expecting 300 MAD when it costs 600. Manual QA across 5,000+ price points in eight languages is not viable. So we built an AI-assisted build-time validation pipeline.

  • Navetteforyou.com validates 5,000 prices in 8 languages without server
  • AI-assisted build-time pipeline checks prices, minimums, consistency
  • Zero pricing errors in production since deployment, 4-second build time

More from Tuesday 17 November →