<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[Smith Neo]]></title><description><![CDATA[Smith Neo]]></description><link>https://sneo.hashnode.dev</link><image><url>https://cdn.hashnode.com/uploads/logos/6a09c4743104e2aff09d5ce4/70edb833-8da5-4005-916a-8ae2c5cd5763.jpg</url><title>Smith Neo</title><link>https://sneo.hashnode.dev</link></image><generator>RSS for Node</generator><lastBuildDate>Mon, 05 Oct 2026 14:54:20 GMT</lastBuildDate><atom:link href="https://sneo.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[I Ran Out of Codex Quota, So I Built an Open-Source Tool That Lets ChatGPT Web Work With My Local Project]]></title><description><![CDATA[If you use ChatGPT on the web for coding, you know the pain: copy code from a browser tab, paste it into your editor, run it, copy the error back, repeat. ChatGPT can't see your file tree, can't use y]]></description><link>https://sneo.hashnode.dev/i-built-an-open-source-tool-for-chatgpt-web</link><guid isPermaLink="true">https://sneo.hashnode.dev/i-built-an-open-source-tool-for-chatgpt-web</guid><category><![CDATA[chatgpt]]></category><category><![CDATA[mcp server]]></category><category><![CDATA[AI]]></category><category><![CDATA[Rust programming]]></category><category><![CDATA[Open Source]]></category><dc:creator><![CDATA[yang sun]]></dc:creator><pubDate>Wed, 30 Sep 2026 07:01:51 GMT</pubDate><content:encoded><![CDATA[<p>If you use ChatGPT on the web for coding, you know the pain: copy code from a browser tab, paste it into your editor, run it, copy the error back, repeat. ChatGPT can't see your file tree, can't use your terminal, and has no idea whether its changes actually work on your machine.</p>
<p>And if you've ever hit your <strong>Codex quota</strong> halfway through a task, it's worse: you're stuck until the limit resets.</p>
<p>So I built <strong>Nexus</strong>, and today I'm releasing <strong>v1.20.0</strong>.</p>
<blockquote>
<p>⭐ If you find it interesting, a star on GitHub is the easiest way to support it: <a href="https://github.com/ysunyang979-sys/Nexus-Web-ChatGPT">ysunyang979-sys/Nexus-Web-ChatGPT</a></p>
</blockquote>
<h2>What is Nexus?</h2>
<p>Nexus is an open-source <strong>Local AI Control Plane</strong> for <strong>ChatGPT Web</strong>. It's a desktop app built with <strong>Tauri 2 + Rust + React</strong> that runs on your machine and connects <code>chatgpt.com</code> to your local workstation through a secure MCP tunnel, with strict, auditable security boundaries.</p>
<p>It's released under <strong>AGPL-3.0</strong>, with installers for Windows, macOS (Apple Silicon) and Linux.</p>
<h2>What can ChatGPT do through Nexus?</h2>
<p>Nexus ships with <strong>332 tools across 24 capability domains</strong>:</p>
<ul>
<li><p><strong>Filesystem</strong> access and <strong>AST-aware</strong> code edits</p>
</li>
<li><p><strong>Git</strong> support, including <strong>managed worktrees</strong> so ChatGPT can experiment in isolated branches</p>
</li>
<li><p>Persistent <code>node-pty</code> <strong>terminals</strong> and background jobs</p>
</li>
<li><p>Built-in <strong>LSP code intelligence</strong> (TypeScript / Python)</p>
</li>
<li><p>Headless and interactive <strong>browser automation</strong></p>
</li>
<li><p><strong>Native Windows Computer Use</strong> (WinRT OCR + Accessibility UI Tree)</p>
</li>
<li><p>A local <strong>SQLite-backed store</strong> for memory, rules, knowledge and prompts</p>
</li>
</ul>
<h2>Security was the first design goal</h2>
<p>Giving an AI access to your computer only makes sense if you stay in control:</p>
<ul>
<li><p><strong>Per-project sandboxing</strong>: authorize specific folders and switch live, without restarting, between <code>read-only</code> / <code>read-write</code> access and <code>disabled</code> / <code>safe-only</code> / <code>project-code</code> execution.</p>
</li>
<li><p><strong>Path protection</strong>: paths are canonicalized (<code>fs.realpathSync.native</code>) to block <code>../</code> traversal and NTFS junction / symlink escapes.</p>
</li>
<li><p><strong>Command risk classification</strong>: AST-based rating as <code>SAFE</code>, <code>CAUTION</code> or <code>DANGEROUS</code>. Destructive commands such as <code>rm -rf /</code>, <code>format</code> and <code>reg delete</code> are blocked by default.</p>
</li>
<li><p><strong>Secret masking</strong>: <code>.env*</code>, <code>*.pem</code>, <code>*.key</code>, <code>id_rsa</code> and cloud credentials are masked automatically.</p>
</li>
<li><p><strong>ACID action ledger (WAL)</strong>: every side-effecting tool call is written to a write-ahead log <em>before</em> it runs, giving you a live audit trail, crash recovery and idempotent replay after network drops.</p>
</li>
</ul>
<h2>Set it up in a few minutes</h2>
<p>Full step-by-step tutorial: <strong><a href="https://nexus.ycu.de5.net/">https://nexus.ycu.de5.net/</a></strong></p>
<p><strong>1. Install Nexus.</strong> Download the installer from <a href="https://github.com/ysunyang979-sys/Nexus-Web-ChatGPT/releases/latest">GitHub Releases</a>, or build from source:</p>
<pre><code class="language-bash">pnpm install &amp;&amp; pnpm run build
</code></pre>
<p>Wait for the status bar to show <strong>Engine Connected</strong>.</p>
<p><strong>2. Create a tunnel and an API key</strong> on the OpenAI Platform. Both are free to create:</p>
<ul>
<li><p>Tunnel: <a href="https://platform.openai.com/settings/organization/tunnels">https://platform.openai.com/settings/organization/tunnels</a> (copy the Tunnel ID)</p>
</li>
<li><p>API key: <a href="https://platform.openai.com/settings/organization/api-keys">https://platform.openai.com/settings/organization/api-keys</a> (this is your Runtime API Key)</p>
</li>
</ul>
<blockquote>
<p>⚠️ An API key is a secret. Never share it or commit it to a repo. Create a <strong>dedicated key just for Nexus</strong> so you can revoke it independently.</p>
</blockquote>
<p><strong>3. Generate a local MCP token.</strong> In Nexus, go to <strong>Settings → MCP &amp; Tokens</strong>, pick a scope preset and click <strong>Auto-Generate Token</strong>. The local endpoint is <code>http://127.0.0.1:18080/mcp</code>. On Windows the token is stored encrypted with DPAPI, which is a Windows-specific feature.</p>
<p><strong>4. Configure the tunnel.</strong> Open <strong>Tunnel Settings</strong>, enter the Tunnel ID, Runtime API Key and your local MCP token, then click <strong>Save &amp; Connect</strong> and <strong>Test Readiness</strong> until it shows <strong>ChatGPT Ready</strong>.</p>
<p><strong>5. Create the MCP app on ChatGPT Web.</strong> On <code>chatgpt.com</code>, open <strong>Plugins</strong>, click <strong>+</strong>, choose <strong>Create MCP app</strong>, set the connection to <strong>Tunnel</strong>, select your tunnel, set authentication to <strong>No authentication</strong>, tick the confirmation box and click <strong>Create</strong>.</p>
<p><strong>6. Start working.</strong> Open a new chat, enable the plugin, and ask ChatGPT to read files, edit code, run commands or manage Git worktrees in your authorized project.</p>
<p><strong>Tip:</strong> start with <code>read-only</code> access and <code>safe-only</code> execution, then loosen permissions per project once you're comfortable.</p>
<h2>What's next</h2>
<p>Right now Nexus is dedicated to ChatGPT Web. Support for more AI clients and platforms is planned for future releases.</p>
<h2>If you like it, please give it a star ⭐</h2>
<p>Nexus is free and open source, and a GitHub star is the simplest way to support it. It helps other developers find the project and keeps me motivated to keep improving it.</p>
<ul>
<li><p>⭐ <strong>Star on GitHub</strong>: <a href="https://github.com/ysunyang979-sys/Nexus-Web-ChatGPT">https://github.com/ysunyang979-sys/Nexus-Web-ChatGPT</a></p>
</li>
<li><p>🌐 <strong>Website &amp; tutorial</strong>: <a href="https://nexus.ycu.de5.net/">https://nexus.ycu.de5.net/</a></p>
</li>
</ul>
<p>Bug reports, feature requests and pull requests are all welcome. Thanks for reading!</p>
]]></content:encoded></item></channel></rss>