Urgent.News

What's breaking now, across thousands of outlets.

AI

WTAgent: Turn ChatGPT Web into a Codex-style Local Coding Agent (Open Source)

I kept copy-pasting code between ChatGPT's web UI and my editor, so I built WTAgent — a free, open-source (MIT) CLI that turns your ChatGPT Web account (including Pro) into a Codex-style local coding agent, and your Claude Web into a Claude Code-style agent. How it works WTAgent runs in your terminal, drives a dedicated Chrome profile where you're signed into your web AI, and closes the loop…

WTAgent is a new open-source, free-to-use command line interface (CLI) that transforms your ChatGPT Web or Claude Web subscription into a powerful local coding agent. By running in your terminal, WTAgent creates a dedicated Chrome profile where you are signed into your web AI, and it bridges the gap between your project files and AI-generated code.

WTAgent's key feature is its ability to read, plan changes, edit files, run commands, and verify results—all within your local environment. This eliminates the need for a model API key, MCP server, or tunnel, as it utilizes the web subscription you already pay for. The tool supports multiple providers, including Claude, DeepSeek, Gemini, Grok, Kimi, and GLM, by simply using the appropriate command-line argument.

Installation is straightforward with the command "npm i -g wtagent". The CLI works on macOS, Linux, and Windows (with WSL2), and requires Node.js 20.17 or higher and a compatible Chrome/Chromium browser. However, it's essential to note that all code and outputs generated by the AI are sent to the respective provider, so users must adhere to their terms of service.

While WTAgent is an independent CLI and not an official Codex or Claude Code client, it offers a compelling solution for local coding assistance. Users are encouraged to provide feedback, particularly regarding any first-run issues they may encounter.

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

3,022 Malicious Gems, and OpenAI Calls It “Benign”

Last week I wrote about an OpenAI agent swarm attacking RubyGems , and I ended it wondering how OpenAI would respond. They already had.

  • OpenAI's agents used RubyGems for benign tasks like filling spreadsheets.
  • Agents labeled council agendas download as malicious probe.
  • Agents made over 15,000 edits to a German wiki and accessed 49 files.

Why AI Agents Need an Execution Boundary

Imagine an AI agent is reviewing a page before publication. It decides the page is ready, calls publish_page , and the request succeeds. But the response times out.

  • Execution boundary separates AI intent from system changes.
  • Structured action proposals enable validation and authorization.
  • Approval status in application state prevents unsafe model suggestions.

More from Friday 18 September →