<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Shuaipeng Guo</title>
    <description>The latest articles on DEV Community by Shuaipeng Guo (@shuaipeng_guo_79180b50553).</description>
    <link>https://dev.to/shuaipeng_guo_79180b50553</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4195131%2F53f8d228-36e6-4378-a7cf-ddc0322c8401.png</url>
      <title>DEV Community: Shuaipeng Guo</title>
      <link>https://dev.to/shuaipeng_guo_79180b50553</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/shuaipeng_guo_79180b50553"/>
    <language>en</language>
    <item>
      <title>Light, Small, Fast: AI-SSH-Pro v0.8.0 Fits Multi-Session SSH in ~3.5MB</title>
      <dc:creator>Shuaipeng Guo</dc:creator>
      <pubDate>Thu, 05 Nov 2026 01:12:43 +0000</pubDate>
      <link>https://dev.to/shuaipeng_guo_79180b50553/light-small-fast-ai-ssh-pro-v080-fits-multi-session-ssh-in-35mb-5blg</link>
      <guid>https://dev.to/shuaipeng_guo_79180b50553/light-small-fast-ai-ssh-pro-v080-fits-multi-session-ssh-in-35mb-5blg</guid>
      <description>&lt;p&gt;Most SSH “suites” grow into hundred-megabyte installers. That tax shows up every time you reimage a laptop, push an upgrade over a shaky VPN, or toss a client onto a USB stick for on-site work.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;AI-SSH-Pro&lt;/strong&gt; v0.8.0 is built around three plain goals:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Goal&lt;/th&gt;
&lt;th&gt;What you get&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Light&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;A desktop Windows client meant to stay open with many sessions&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Small&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Installer &lt;strong&gt;under 10MB&lt;/strong&gt;; recommended NSIS setup about &lt;strong&gt;3.5MB&lt;/strong&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Fast&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Quicker downloads and upgrades when bandwidth is the bottleneck&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5eqoiuqt7oz018nbnliv.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F5eqoiuqt7oz018nbnliv.png" alt="Terminal + AI assistant" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Site: &lt;a href="https://ai-ssh-pro.vercel.app" rel="noopener noreferrer"&gt;https://ai-ssh-pro.vercel.app&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;GitHub release: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases/tag/v0.8.0" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro/releases/tag/v0.8.0&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Repo: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Gitee mirror: &lt;a href="https://gitee.com/CodeYG/ai-ssh-pro" rel="noopener noreferrer"&gt;https://gitee.com/CodeYG/ai-ssh-pro&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Why install size is an ops feature
&lt;/h2&gt;

&lt;p&gt;SSH is infrastructure muscle memory. If the client itself is heavy, every rebuild and every remote assist costs minutes you never budget for.&lt;/p&gt;

&lt;p&gt;v0.8.0 keeps the download story boring on purpose:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Package&lt;/th&gt;
&lt;th&gt;Notes&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;AI-SSH-Pro_0.8.0_x64-setup.exe&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Recommended NSIS installer (~3.5MB)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;AI-SSH-Pro_0.8.0_x64_en-US.msi&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;MSI alternative&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The product name is &lt;strong&gt;AI-SSH-Pro&lt;/strong&gt;. Installers and shortcuts are presented as a normal desktop app — not as a billboard for the underlying UI stack.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7s56dj60v2afzts4ksp0.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7s56dj60v2afzts4ksp0.png" alt="Ops home" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Still a full day-to-day SSH desk
&lt;/h2&gt;

&lt;p&gt;Small does not mean “terminal only”:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Multi-session SSH&lt;/strong&gt; — several hosts without a window per box
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;SFTP&lt;/strong&gt; — browse and transfer on the active session
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;ProxyJump&lt;/strong&gt; — bastion / jump-host paths without a second tool
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ops toolbox&lt;/strong&gt; — common helpers beside the terminal
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Confirm-before-execute AI&lt;/strong&gt; — suggestions first; nothing runs until you approve
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The AI path is intentionally gated: observe → plan → run only after you confirm. Useful for drafting checks and safe command lists; not an unsupervised autopilot on production.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzvnyvoq4tasfp5uw474i.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzvnyvoq4tasfp5uw474i.png" alt="AI panel with suggested command" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Flwa56u5cqhho8vw8z1nv.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Flwa56u5cqhho8vw8z1nv.png" alt="Dual-pane SFTP" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Where “fast” shows up
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Thin VPN / jump laptop upgrades&lt;/strong&gt; — a few megabytes is a different wait than a few hundred
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;USB field kits&lt;/strong&gt; — keep the installer next to your other emergency tools
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Handing a teammate a link&lt;/strong&gt; — they install and connect instead of babysitting a download bar
&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Get v0.8.0
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Open &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases/tag/v0.8.0" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro/releases/tag/v0.8.0&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Grab the ~3.5MB setup.exe (or the MSI)
&lt;/li&gt;
&lt;li&gt;Install, launch, connect from your host list
&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Source and issues: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro&lt;/a&gt;&lt;br&gt;&lt;br&gt;
China-friendly mirror: &lt;a href="https://gitee.com/CodeYG/ai-ssh-pro" rel="noopener noreferrer"&gt;https://gitee.com/CodeYG/ai-ssh-pro&lt;/a&gt;&lt;br&gt;&lt;br&gt;
Product site: &lt;a href="https://ai-ssh-pro.vercel.app" rel="noopener noreferrer"&gt;https://ai-ssh-pro.vercel.app&lt;/a&gt;  &lt;/p&gt;

&lt;p&gt;If you want a Windows SSH client that stays light enough to move with you — and AI that waits for a human confirm — try v0.8.0 and tell us what jump-host / SFTP / confirm UX still needs polish.&lt;/p&gt;

</description>
      <category>ssh</category>
      <category>devops</category>
      <category>opensource</category>
      <category>windows</category>
    </item>
    <item>
      <title>Under 10MB: AI-SSH-Pro v0.8.0 — Multi-session SSH with Confirm-Before-Execute AI</title>
      <dc:creator>Shuaipeng Guo</dc:creator>
      <pubDate>Wed, 04 Nov 2026 02:21:46 +0000</pubDate>
      <link>https://dev.to/shuaipeng_guo_79180b50553/under-10mb-ai-ssh-pro-v080-multi-session-ssh-with-confirm-before-execute-ai-1eff</link>
      <guid>https://dev.to/shuaipeng_guo_79180b50553/under-10mb-ai-ssh-pro-v080-multi-session-ssh-with-confirm-before-execute-ai-1eff</guid>
      <description>&lt;p&gt;&lt;strong&gt;AI-SSH-Pro&lt;/strong&gt; is a lightweight Windows desktop SSH client for people who live in terminals all day: many hosts, jump boxes, file transfers, and the occasional “can AI draft that command?” moment — without letting an assistant silently run anything on production.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;v0.8.0&lt;/strong&gt; makes the download story simple: the Windows installer is &lt;strong&gt;under 10MB&lt;/strong&gt;. The recommended NSIS setup is about &lt;strong&gt;3.5MB&lt;/strong&gt; (roughly 3–4MB). That is small enough to keep on a USB stick, ship over a flaky VPN, or upgrade without watching a progress bar for minutes.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Website: &lt;a href="https://ai-ssh-pro.vercel.app" rel="noopener noreferrer"&gt;https://ai-ssh-pro.vercel.app&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;GitHub release: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases/tag/v0.8.0" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro/releases/tag/v0.8.0&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Repo: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Gitee mirror: &lt;a href="https://gitee.com/CodeYG/ai-ssh-pro" rel="noopener noreferrer"&gt;https://gitee.com/CodeYG/ai-ssh-pro&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Why “small” matters for an SSH client
&lt;/h2&gt;

&lt;p&gt;SSH tools often grow into multi-hundred-megabyte installers. For ops and platform work, that friction is real: slower upgrades, harder offline installs, more disk noise on jump laptops.&lt;/p&gt;

&lt;p&gt;v0.8.0 keeps the focus on a &lt;strong&gt;desktop Windows&lt;/strong&gt; package you can actually download quickly:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Package&lt;/th&gt;
&lt;th&gt;Notes&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;AI-SSH-Pro_0.8.0_x64-setup.exe&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Recommended NSIS installer (~3.5MB)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;AI-SSH-Pro_0.8.0_x64_en-US.msi&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;MSI alternative&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Product-facing name is simply &lt;strong&gt;AI-SSH-Pro&lt;/strong&gt; — installer and shortcuts do not carry a framework brand name. You get a client that feels like a normal desktop app, not a branded demo of its runtime.&lt;/p&gt;




&lt;h2&gt;
  
  
  Confirm-before-execute AI (not silent autopilot)
&lt;/h2&gt;

&lt;p&gt;The AI path is intentionally conservative:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;observe → plan → run only after you confirm&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Suggested commands show up first. You decide whether they land in the shell. That is the whole point for production hosts: assistance stays in the loop; the final click stays with you.&lt;/p&gt;

&lt;p&gt;Use it when you want help diagnosing disk pressure, drafting a safe checklist, or turning a natural-language ask into a labeled command — without treating the model as an unsupervised operator. If a suggestion looks wrong, you cancel. If it looks right, you confirm. Nothing “helpful” sneaks onto the wire first.&lt;/p&gt;

&lt;p&gt;That design matches how careful teams already work: review, then run. The difference is that the draft can come from AI while the gate remains human.&lt;/p&gt;




&lt;h2&gt;
  
  
  Everyday SSH capabilities kept in one place
&lt;/h2&gt;

&lt;p&gt;v0.8.0 keeps the day-to-day ops surface area you expect from a multi-session client:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Multi-session SSH&lt;/strong&gt; — work across several hosts without juggling a separate window for every box
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;SFTP&lt;/strong&gt; — browse and transfer files over the active session
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;ProxyJump&lt;/strong&gt; — bastion / jump-host workflows without bolting on a second tool
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ops toolbox&lt;/strong&gt; — common operational helpers beside the terminal, not in a separate app
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The pitch is continuity: connect, transfer, jump, and get confirmable AI help in one lightweight desktop client. Fewer context switches means fewer chances to paste the right command into the wrong session.&lt;/p&gt;




&lt;h2&gt;
  
  
  Who it is for
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Sysadmins and SREs who keep many sessions open
&lt;/li&gt;
&lt;li&gt;Developers who SSH into staging/prod and want AI drafts with a hard confirm gate
&lt;/li&gt;
&lt;li&gt;Anyone tired of oversized installers for a tool that should feel local and fast
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If you need “AI that never asks,” this is not that product. If you want &lt;strong&gt;AI that waits for you&lt;/strong&gt;, it matches the design.&lt;/p&gt;




&lt;h2&gt;
  
  
  Get v0.8.0
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Open the release page: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases/tag/v0.8.0" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro/releases/tag/v0.8.0&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Download &lt;code&gt;AI-SSH-Pro_0.8.0_x64-setup.exe&lt;/code&gt; (recommended, ~3.5MB) or the MSI
&lt;/li&gt;
&lt;li&gt;Install, launch, and connect from your host list
&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Source and issues: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro&lt;/a&gt;&lt;br&gt;&lt;br&gt;
China-friendly mirror: &lt;a href="https://gitee.com/CodeYG/ai-ssh-pro" rel="noopener noreferrer"&gt;https://gitee.com/CodeYG/ai-ssh-pro&lt;/a&gt;&lt;br&gt;&lt;br&gt;
Product site: &lt;a href="https://ai-ssh-pro.vercel.app" rel="noopener noreferrer"&gt;https://ai-ssh-pro.vercel.app&lt;/a&gt;  &lt;/p&gt;

&lt;p&gt;A Chinese intro is already live on CSDN for readers who prefer that channel. This English post is the parallel announcement for international developer communities.&lt;/p&gt;

&lt;p&gt;Feedback on jump-host flows, SFTP, and the confirm-before-execute UX is especially welcome — Issues and stars help more than slogans.&lt;/p&gt;

</description>
      <category>ssh</category>
      <category>devops</category>
      <category>windows</category>
      <category>ai</category>
    </item>
    <item>
      <title>Dual-pane SFTP with a resumable transfer queue: a practical walkthrough of AI-SSH-Pro</title>
      <dc:creator>Shuaipeng Guo</dc:creator>
      <pubDate>Thu, 22 Oct 2026 13:13:03 +0000</pubDate>
      <link>https://dev.to/shuaipeng_guo_79180b50553/dual-pane-sftp-with-a-resumable-transfer-queue-a-practical-walkthrough-of-ai-ssh-pro-11al</link>
      <guid>https://dev.to/shuaipeng_guo_79180b50553/dual-pane-sftp-with-a-resumable-transfer-queue-a-practical-walkthrough-of-ai-ssh-pro-11al</guid>
      <description>&lt;p&gt;Most SSH clients treat file transfer as an afterthought: a separate window, a separate login, and a progress bar that vanishes the moment your Wi-Fi blinks. In &lt;strong&gt;AI-SSH-Pro&lt;/strong&gt;, SFTP runs over the SSH session you already have open, in a dual-pane tab next to your terminal, with a transfer queue that remembers where it stopped.&lt;/p&gt;

&lt;p&gt;This post walks through how that works in day-to-day use.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Website: &lt;a href="https://guoshuaipeng.github.io/ai-ssh-pro/" rel="noopener noreferrer"&gt;https://guoshuaipeng.github.io/ai-ssh-pro/&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Download: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro/releases&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Foeo82cr2c93ax1o4iyw1.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Foeo82cr2c93ax1o4iyw1.png" alt="Dual-pane SFTP: local on the left, remote on the right, transfer queue below" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Screenshots use the English UI and the built-in simulated demo host, so no real server or data appears.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  The layout: two panes, one session
&lt;/h2&gt;

&lt;p&gt;Open SFTP from the toolbar (or from a connected host) and you get a tab named after the host:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Left pane:&lt;/strong&gt; your local machine, with a drive picker on Windows and a clickable breadcrumb path.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Right pane:&lt;/strong&gt; the remote host, starting in your home directory, with a &lt;strong&gt;Mode&lt;/strong&gt; column (&lt;code&gt;rwxr-xr-x&lt;/code&gt;) next to size and modified time.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Middle:&lt;/strong&gt; &lt;strong&gt;Upload&lt;/strong&gt; and &lt;strong&gt;Download&lt;/strong&gt; buttons for whatever is selected.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Bottom:&lt;/strong&gt; the &lt;strong&gt;Transfers&lt;/strong&gt; panel.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Both panes share the same toolbar ideas: up, refresh, new folder, recent folders, search, and a name filter. Columns sort folders first, then by name, time, or size. Clicking the same header flips the direction. Name sorts A to Z first, time newest first, and size largest first, which is usually what you want when you are hunting for a fresh log or a huge file.&lt;/p&gt;

&lt;p&gt;Because it reuses the authenticated SSH connection, there is no second password prompt and no separate "site manager" to keep in sync.&lt;/p&gt;

&lt;h2&gt;
  
  
  Follow the terminal
&lt;/h2&gt;

&lt;p&gt;The small &lt;strong&gt;Follow terminal&lt;/strong&gt; checkbox is the feature I use most. With it on, when you &lt;code&gt;cd /var/log/nginx&lt;/code&gt; in the terminal tab, the remote pane moves to the same directory. When you want to grab the file you were just looking at with &lt;code&gt;tail&lt;/code&gt;, it is already on screen.&lt;/p&gt;

&lt;p&gt;It is honest about failures, too. If the terminal moves somewhere the SFTP account cannot read (for example, you &lt;code&gt;sudo -i&lt;/code&gt; into &lt;code&gt;/root&lt;/code&gt;), the pane stays where it is and tells you why, instead of throwing an error dialog:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;The terminal moved to /root, but the SFTP account cannot read it (sudo may be in use). The remote pane stays at /home/demo.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  A transfer queue that survives disconnects
&lt;/h2&gt;

&lt;p&gt;Every upload and download becomes a task in the &lt;strong&gt;Transfers&lt;/strong&gt; panel. Tasks are grouped into &lt;strong&gt;Active&lt;/strong&gt;, &lt;strong&gt;Paused / Interrupted&lt;/strong&gt;, &lt;strong&gt;Failed&lt;/strong&gt;, and &lt;strong&gt;Finished&lt;/strong&gt;. Each one shows bytes, percent, speed, and ETA on one line.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Faudcnip3wwcu106gqnz1.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Faudcnip3wwcu106gqnz1.png" alt="Transfer queue with an interrupted download, Resume and Retry all" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;A few details make it hold up on flaky networks:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Partial files are explicit.&lt;/strong&gt; Data streams into &lt;code&gt;name.aisshpart&lt;/code&gt; first and is only renamed to the real filename when complete, so you never mistake half a tarball for a finished one.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Resume is the default, but it checks before it continues.&lt;/strong&gt; The source file's modification time is recorded when the task is created. If the source changed since then, the client refuses to append to the old partial file and starts over, so you don't end up with a stitched-together file.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The queue is persisted.&lt;/strong&gt; If the app or the session dies mid-transfer, unfinished tasks come back as &lt;em&gt;Interrupted&lt;/em&gt; rather than disappearing. With &lt;strong&gt;Auto-resume&lt;/strong&gt; on, they continue when that host reconnects; otherwise press &lt;strong&gt;Resume&lt;/strong&gt; or &lt;strong&gt;Retry all&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Failures say what happened.&lt;/strong&gt; You get messages like &lt;em&gt;Transfer stopped: session dropped&lt;/em&gt;, &lt;em&gt;host unreachable&lt;/em&gt;, or &lt;em&gt;authentication error&lt;/em&gt;, not just "failed".&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Name conflicts ask first.&lt;/strong&gt; When the destination already has the file, choose &lt;strong&gt;Overwrite&lt;/strong&gt; or &lt;strong&gt;Keep both&lt;/strong&gt;. Keep both writes &lt;code&gt;report (1).csv&lt;/code&gt;, then &lt;code&gt;report (2).csv&lt;/code&gt;, and so on.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Finished tasks stay visible until you clear them, which is handy when you need to confirm later that a file really did land.&lt;/p&gt;

&lt;h2&gt;
  
  
  Finding files without running &lt;code&gt;find&lt;/code&gt;
&lt;/h2&gt;

&lt;p&gt;Click search on the remote pane to look for file names in subfolders:&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fjo77q0kanou2mdh4fc5w.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fjo77q0kanou2mdh4fc5w.png" alt="Search files in subfolders" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;It walks the folder with plain SFTP directory listings, so &lt;strong&gt;no shell command runs on the server&lt;/strong&gt;. That matters on locked-down accounts and audited hosts. It is deliberately bounded: up to 5 levels deep, at most 400 folders and 200 hits. It skips &lt;code&gt;.git&lt;/code&gt;, &lt;code&gt;node_modules&lt;/code&gt;, &lt;code&gt;.cache&lt;/code&gt;, &lt;code&gt;__pycache__&lt;/code&gt;, &lt;code&gt;/proc&lt;/code&gt;, and &lt;code&gt;/sys&lt;/code&gt;, and it doesn't follow symlinked folders. Press Enter on a hit and the pane jumps to it.&lt;/p&gt;

&lt;p&gt;For the current folder only, the &lt;strong&gt;Filter by name&lt;/strong&gt; box narrows the list as you type.&lt;/p&gt;

&lt;h2&gt;
  
  
  Small things that add up
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Edit in place.&lt;/strong&gt; Double-click a text file under 2 MB to open it in the built-in editor. If the file changed on the server after you opened it, saving warns you instead of silently overwriting someone else's change. Larger files can be opened read-only (first 2 MB) or downloaded, and binary files are offered as downloads.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Permissions without a shell.&lt;/strong&gt; Change mode with octal (&lt;code&gt;755&lt;/code&gt;, &lt;code&gt;0644&lt;/code&gt;) or symbolic input (&lt;code&gt;u+x&lt;/code&gt;, &lt;code&gt;go-w&lt;/code&gt;, &lt;code&gt;a+r&lt;/code&gt;). It is applied with SFTP &lt;code&gt;chmod&lt;/code&gt;, not by typing a command into your terminal.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Recent folders per host.&lt;/strong&gt; The last 10 remote folders for each host, and your recent local folders, are one click away.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Paste a path.&lt;/strong&gt; Paste &lt;code&gt;"/var/log/app"&lt;/code&gt; with quotes or a Windows-style path, and the path box normalizes it before jumping.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;New file never overwrites.&lt;/strong&gt; Creating a file with an existing name is refused.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Density.&lt;/strong&gt; On windows 1366 px wide or narrower, the panes default to a compact layout. You can force Comfort if you prefer more space.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Where the AI assistant fits
&lt;/h2&gt;

&lt;p&gt;AI-SSH-Pro also has an AI ops assistant docked next to the terminal. For files, it can prepare an action card such as "open &lt;code&gt;/var/log&lt;/code&gt; on web-01 in SFTP" or "download &lt;code&gt;config.yaml&lt;/code&gt; to this folder". &lt;strong&gt;Nothing runs until you click confirm&lt;/strong&gt;, and uploads ask a second time because they can overwrite a remote file. Every AI-initiated action lands in the audit log.&lt;/p&gt;

&lt;p&gt;You bring your own OpenAI-compatible model (OpenAI, DeepSeek, Qwen, and others). Keys stay in your OS's secure storage, and the assistant never receives your passwords or private keys.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it in two minutes
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Grab an installer for Windows, macOS, or Linux from &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases" rel="noopener noreferrer"&gt;GitHub Releases&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;Open the built-in &lt;strong&gt;demo host&lt;/strong&gt; and click &lt;strong&gt;SFTP&lt;/strong&gt;. It is simulated and read-only, so you can browse, search, download, and watch the Transfers panel without a server.&lt;/li&gt;
&lt;li&gt;Connect a real host, turn on &lt;strong&gt;Follow terminal&lt;/strong&gt;, and &lt;code&gt;cd&lt;/code&gt; around.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;More on the &lt;a href="https://guoshuaipeng.github.io/ai-ssh-pro/" rel="noopener noreferrer"&gt;project site&lt;/a&gt;. AI-SSH-Pro is open source under MIT. Feedback and &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/issues" rel="noopener noreferrer"&gt;issues&lt;/a&gt; are very welcome.&lt;/p&gt;

&lt;p&gt;What's the one thing your current SFTP client gets wrong on a bad connection?&lt;/p&gt;

</description>
      <category>devops</category>
      <category>ssh</category>
      <category>opensource</category>
      <category>productivity</category>
    </item>
    <item>
      <title>Suggest only vs Confirm before run vs Auto run: picking an AI execution policy in AI-SSH-Pro</title>
      <dc:creator>Shuaipeng Guo</dc:creator>
      <pubDate>Wed, 21 Oct 2026 13:12:31 +0000</pubDate>
      <link>https://dev.to/shuaipeng_guo_79180b50553/suggest-only-vs-confirm-before-run-vs-auto-run-picking-an-ai-execution-policy-in-ai-ssh-pro-22be</link>
      <guid>https://dev.to/shuaipeng_guo_79180b50553/suggest-only-vs-confirm-before-run-vs-auto-run-picking-an-ai-execution-policy-in-ai-ssh-pro-22be</guid>
      <description>&lt;p&gt;An AI that can type into your SSH session is useful. An AI that can type into your &lt;strong&gt;production&lt;/strong&gt; SSH session without asking is a different story.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;AI-SSH-Pro&lt;/strong&gt; is an open-source SSH desktop client with an AI ops assistant docked next to the terminal. The assistant proposes commands with risk labels; you choose how much autonomy it gets. That choice is the &lt;strong&gt;AI execution policy&lt;/strong&gt;: three levels that stay consistent across Settings, each host's Safety card, and the current tab.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Website: &lt;a href="https://guoshuaipeng.github.io/ai-ssh-pro/" rel="noopener noreferrer"&gt;https://guoshuaipeng.github.io/ai-ssh-pro/&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Download: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro/releases&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7dbq29uhlt04jsaghc9x.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7dbq29uhlt04jsaghc9x.png" alt="AI execution policy in Settings → AI" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Screenshots use the English UI and the built-in demo host where applicable. Some sample AI replies in the demo may be in Chinese.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  The three policies at a glance
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Policy&lt;/th&gt;
&lt;th&gt;Read-only / low risk&lt;/th&gt;
&lt;th&gt;Medium risk (writes, restarts)&lt;/th&gt;
&lt;th&gt;High risk&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Suggest only&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;May run&lt;/td&gt;
&lt;td&gt;Shown as suggestion only — you copy or type yourself&lt;/td&gt;
&lt;td&gt;Suggestion only&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Confirm before run&lt;/strong&gt; (default)&lt;/td&gt;
&lt;td&gt;May run&lt;/td&gt;
&lt;td&gt;Needs your confirmation&lt;/td&gt;
&lt;td&gt;Needs confirmation (stricter for dangerous ops)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Auto run&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Runs automatically&lt;/td&gt;
&lt;td&gt;Runs automatically&lt;/td&gt;
&lt;td&gt;Still asks&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Risk labels (&lt;strong&gt;low / medium / high&lt;/strong&gt;) sit on every suggested command block next to &lt;strong&gt;Copy / Run&lt;/strong&gt;. The policy does not invent risk; it decides whether a labelled command is allowed to execute without you.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where you set it
&lt;/h2&gt;

&lt;p&gt;You can set policy in three places. The client computes one &lt;strong&gt;effective&lt;/strong&gt; policy for the tab so you are never guessing which layer won.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Settings → AI&lt;/strong&gt; — global default for new sessions.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Host Safety card&lt;/strong&gt; — per-host default (and protection level: Normal / Production / Read-only).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tab policy menu&lt;/strong&gt; (input row) — temporary override for this connection only. It lives in memory and &lt;strong&gt;resets on disconnect&lt;/strong&gt;.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7p2mg779mxxajrmujg49.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7p2mg779mxxajrmujg49.png" alt="Host Safety card: protection level and AI policy" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;When a host is Read-only, the Safety card can show something like &lt;em&gt;"In effect: Suggest only (limited by read-only guard)"&lt;/em&gt; — the UI matches what the main process will actually allow.&lt;/p&gt;

&lt;h2&gt;
  
  
  Scenario 1: laptop / jump box / learning
&lt;/h2&gt;

&lt;p&gt;On a throwaway VM or the built-in &lt;strong&gt;demo host&lt;/strong&gt;, friction gets in the way of learning. &lt;strong&gt;Auto run&lt;/strong&gt; (or Confirm with a light hand) lets you ask &lt;em&gt;"show me listening ports"&lt;/em&gt; and see &lt;code&gt;ss&lt;/code&gt; output without clicking every time.&lt;/p&gt;

&lt;p&gt;Still useful habits even here:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Keep high-risk behind a prompt (Auto run already does that).&lt;/li&gt;
&lt;li&gt;Prefer attaching terminal output to a question when you already have a log open, instead of retyping context.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The demo host is simulated, so you can click Run freely while you learn the policy UI.&lt;/p&gt;

&lt;h2&gt;
  
  
  Scenario 2: staging
&lt;/h2&gt;

&lt;p&gt;Staging is where most teams want &lt;strong&gt;Confirm before run&lt;/strong&gt;. Diagnostics (&lt;code&gt;systemctl status&lt;/code&gt;, &lt;code&gt;tail&lt;/code&gt; on logs, &lt;code&gt;ss&lt;/code&gt;, &lt;code&gt;journalctl&lt;/code&gt;) stay low-friction. Restarts and config edits wait for a click.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fvjj52courpnpmhgstbqv.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fvjj52courpnpmhgstbqv.png" alt="Confirm dialog before an AI-proposed command runs" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;A typical flow:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;You describe the symptom (or attach terminal output).&lt;/li&gt;
&lt;li&gt;The assistant returns read-only commands labelled &lt;strong&gt;Risk: low&lt;/strong&gt; — they may run under Confirm.&lt;/li&gt;
&lt;li&gt;It proposes &lt;code&gt;sudo systemctl restart …&lt;/code&gt; labelled medium/high — you get a confirm card, then decide.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;That split is the point of Confirm: speed on investigation, a hard stop before mutation.&lt;/p&gt;

&lt;h2&gt;
  
  
  Scenario 3: production
&lt;/h2&gt;

&lt;p&gt;Mark the host &lt;strong&gt;Production&lt;/strong&gt; on the Safety card. You get:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A &lt;strong&gt;red production banner&lt;/strong&gt; on the terminal tab so the context is always visible.&lt;/li&gt;
&lt;li&gt;Stricter confirmation for dangerous commands (including typing the host name where required).&lt;/li&gt;
&lt;li&gt;Prefer &lt;strong&gt;Confirm before run&lt;/strong&gt;, or even &lt;strong&gt;Suggest only&lt;/strong&gt; if you want the AI as a second pair of eyes only.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3yoao6dgpnl039bs4rxg.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3yoao6dgpnl039bs4rxg.png" alt="Production banner on the terminal tab" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fje4bt2eygrfsfh33yp07.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fje4bt2eygrfsfh33yp07.png" alt="Production-style confirm / ask flow" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Auto run on a production host is almost never worth the minutes you save. One mistaken restart at peak traffic costs more than a confirm click.&lt;/p&gt;

&lt;h2&gt;
  
  
  Scenario 4: read-only hosts (and why UI alone is not enough)
&lt;/h2&gt;

&lt;p&gt;Some accounts should never receive writes from AI: auditors, shared jump users, or "view-only" break-glass profiles. Mark the host &lt;strong&gt;Read-only&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;The important part: the block is enforced in Electron's &lt;strong&gt;main process&lt;/strong&gt;, not only in the renderer. Even if someone clicks approve on a write, the AI path cannot inject mutating input into that terminal. Suggest only / Confirm / Auto run cannot override that guard — the effective policy is capped.&lt;/p&gt;

&lt;p&gt;That is the difference between a preference and a control.&lt;/p&gt;

&lt;h2&gt;
  
  
  Per-tab overrides without permanent foot-guns
&lt;/h2&gt;

&lt;p&gt;Need Auto run for five minutes of noisy log grepping on staging? Raise the tab policy from the input-row menu. Disconnect (or close the tab), and it falls back to the host / global default. No "I left production on Auto run last Tuesday" landmine.&lt;/p&gt;

&lt;p&gt;Conversation context stays &lt;strong&gt;per tab / per host&lt;/strong&gt;, so the policy you see matches the machine you are talking about. AI-initiated commands and dangerous ops also land in the &lt;strong&gt;audit log&lt;/strong&gt; for later review or export.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fz0wv9tc744qc4xmmgy38.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fz0wv9tc744qc4xmmgy38.png" alt="Conversation history grouped by host" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Bring your own model
&lt;/h2&gt;

&lt;p&gt;The assistant talks to any OpenAI-compatible provider (OpenAI, DeepSeek, Qwen, and others). API keys stay in the OS secure store on your machine — they are not uploaded to a product cloud. Policy is local to the client either way: changing model does not change how Confirm / Suggest / Auto behave.&lt;/p&gt;

&lt;h2&gt;
  
  
  A practical default matrix
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Host type&lt;/th&gt;
&lt;th&gt;Protection&lt;/th&gt;
&lt;th&gt;Suggested AI policy&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Demo / local VM&lt;/td&gt;
&lt;td&gt;Normal&lt;/td&gt;
&lt;td&gt;Auto run or Confirm&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Staging&lt;/td&gt;
&lt;td&gt;Normal&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;Confirm before run&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Production&lt;/td&gt;
&lt;td&gt;Production&lt;/td&gt;
&lt;td&gt;Confirm (or Suggest only)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Audit / view-only&lt;/td&gt;
&lt;td&gt;Read-only&lt;/td&gt;
&lt;td&gt;Effective Suggest only (main-process guard)&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Start global at Confirm, then tighten hosts that matter. Loosen only where you can afford mistakes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Install from &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases" rel="noopener noreferrer"&gt;GitHub Releases&lt;/a&gt; (Windows, macOS, Linux).&lt;/li&gt;
&lt;li&gt;Open the &lt;strong&gt;demo host&lt;/strong&gt; and flip Settings → AI between the three policies; watch which commands ask and which run.&lt;/li&gt;
&lt;li&gt;Add a real host, set Production or Read-only on the Safety card, and confirm the banner / effective policy text.&lt;/li&gt;
&lt;li&gt;Point &lt;strong&gt;Settings → AI&lt;/strong&gt; at your OpenAI-compatible provider when you are ready for live triage.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;More on the &lt;a href="https://guoshuaipeng.github.io/ai-ssh-pro/" rel="noopener noreferrer"&gt;project site&lt;/a&gt;. AI-SSH-Pro is MIT-licensed; stars, feedback, and &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/issues" rel="noopener noreferrer"&gt;issues&lt;/a&gt; are welcome.&lt;/p&gt;

&lt;p&gt;Which policy do you default to on production boxes — Confirm, Suggest only, or something else entirely?&lt;/p&gt;

</description>
      <category>devops</category>
      <category>ssh</category>
      <category>opensource</category>
      <category>ai</category>
    </item>
    <item>
      <title>One Developer + One AI Bot: Building an SSH Client in 45 Self-Iterating Rounds</title>
      <dc:creator>Shuaipeng Guo</dc:creator>
      <pubDate>Wed, 21 Oct 2026 05:17:19 +0000</pubDate>
      <link>https://dev.to/shuaipeng_guo_79180b50553/one-developer-one-ai-bot-building-an-ssh-client-in-45-self-iterating-rounds-167j</link>
      <guid>https://dev.to/shuaipeng_guo_79180b50553/one-developer-one-ai-bot-building-an-ssh-client-in-45-self-iterating-rounds-167j</guid>
      <description>&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqu8ke7bvwq7xvedn5y0x.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqu8ke7bvwq7xvedn5y0x.png" alt="AI-SSH-Pro: one developer + one AI bot, 45 self-iterating rounds" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;I'm building &lt;strong&gt;AI-SSH-Pro&lt;/strong&gt;, an open-source SSH client with a built-in AI assistant (Electron + React + TypeScript): terminal, SFTP, command history and AI chat in one window.&lt;/p&gt;

&lt;p&gt;The unusual part is the team. It's me plus one AI bot. I treat the bot as a whole product team — product manager, UX/UI designer, developer, QA, and retro lead — and let it iterate round after round on its own. My job is direction, feedback, and deciding when to ship.&lt;/p&gt;

&lt;p&gt;This post covers how that works, the actual role prompts I use, and real retrospective notes from the repo.&lt;/p&gt;

&lt;h2&gt;
  
  
  The numbers
&lt;/h2&gt;

&lt;p&gt;All of these come straight from &lt;code&gt;git log&lt;/code&gt; and the docs folder, as of 2026-10-21:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;405 commits&lt;/strong&gt; in total (first commit on 2026-04-02), &lt;strong&gt;319&lt;/strong&gt; of them in just two days (Oct 20–21).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;45 rounds&lt;/strong&gt; of continuous self-iteration completed since Oct 20; round 46 is in progress.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;124 design docs&lt;/strong&gt; under &lt;code&gt;docs/iterations/&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;407 regression tests&lt;/strong&gt;, all green on the latest full &lt;code&gt;npm test&lt;/code&gt; run.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;11 retro entries&lt;/strong&gt; in &lt;code&gt;docs/retro.md&lt;/code&gt;, distilled into &lt;strong&gt;17 rules&lt;/strong&gt; in &lt;code&gt;docs/dev-rules.md&lt;/code&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The loop: one bot, five hats
&lt;/h2&gt;

&lt;p&gt;Every round follows the same pipeline:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Product → UX/UI design → Development → QA → Retro&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Product&lt;/strong&gt; proposes just 3 small items per round, focused on what people use every day: connecting, the terminal, SFTP, history, AI chat.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Design&lt;/strong&gt; screenshots the real UI first, then writes a design doc.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Development&lt;/strong&gt; commits in small steps — one item, one commit.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;QA&lt;/strong&gt; adds regression tests; nothing is done until the full suite passes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Retro&lt;/strong&gt; turns every mistake into an actionable rule that the next round reads first.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;A round takes roughly 30 minutes. Any bug I report jumps the queue.&lt;/p&gt;

&lt;h2&gt;
  
  
  The role prompts
&lt;/h2&gt;

&lt;p&gt;These are the prompts I actually use, lightly trimmed. Feel free to copy them.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Shared context (always first)&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Project: AI-SSH-Pro, an AI SSH client (Electron + React + TypeScript).
Rules: commit locally only; push only for releases. npm test must be fully green.
Use theme components and spacing tokens, never native &amp;lt;select&amp;gt;.
Production hosts show a banner but are not blocked; read-only hosts block writes.
AI-generated commands are only prefilled — the user presses send.
Redact anything before it is sent to the AI.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Product&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;You are the product manager. Only propose high-frequency, everyday features;
niche ideas go to the backlog. Before proposing, grep docs/iterations and src
to dedupe — never re-propose something already built.
Each round: 3 small items, each with the user scenario, what feels awkward today,
and the change the user will actually notice.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;UX / UI design&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;You are the UX/UI designer. Screenshot the real UI first, then write a design doc.
Cover empty, disabled (explain why) and error states, shortcuts, and both
light and dark themes. Principles: dense but not cramped, symmetric layouts,
icon buttons over text buttons, and never drop an existing feature.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Development&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;You are the developer. Implement the design doc in small steps, one commit per item,
no pushing. Toggle-style state must read the latest value from a ref.
Render popovers into document.body so they never get clipped.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;QA&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;You are QA. Every item gets a regression test covering rapid repeated clicks,
acting after switching tabs, and edge cases. Run the full npm test on the final
HEAD — it must be all green. Capture before/after screenshots.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Retro&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;You are the retro lead. After every round and every bug fix, append an entry to
docs/retro.md: time (planned vs actual), where it got stuck, what went wrong,
root cause, and how to avoid it next time (as an actionable rule).
Promote general rules to docs/dev-rules.md. Read them before each round starts.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Self-iteration (one line to start it)&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Iterate continuously: product → design → dev → QA → retro, round after round,
without asking me. ~30 minutes per round; bugs I report jump the queue.
End each round with a 3–5 line report.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Quick request / bug (the one I use most)&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;[screenshot] This doesn't respond / this looks messy — designers, please fix it.
State the root cause or design rationale first, then fix, test, commit, and write a retro entry.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Before and after
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Split view: from a dedicated toolbar to pane headers
&lt;/h3&gt;

&lt;p&gt;The old split toolbar ate a whole row, used a native dropdown, and had a cryptic "left/right" button. Now the controls live in each pane's header, primary and secondary panes are clearly labeled, and broadcast input is a real toggle.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F9yvwadp9ighkxwq39bxz.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F9yvwadp9ighkxwq39bxz.png" alt="Split view before and after" width="800" height="272"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The split button in the top-right became a menu: off, side-by-side, or stacked, with the current mode checked.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ff7agr2jya1m8cq3xruzu.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ff7agr2jya1m8cq3xruzu.png" alt="Split mode menu" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Snippets: one line each, built-ins grew from 3 to 26
&lt;/h3&gt;

&lt;p&gt;Each snippet used to be a tall card. Now it's a single row, action buttons appear on hover, and double-click or Enter only prefills the command — it never runs it. Risky commands like delete or reboot are deliberately left out of the built-in list.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmdauq1otibp0oeudo661.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmdauq1otibp0oeudo661.png" alt="Snippets panel before and after" width="799" height="363"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  SFTP header: it started with one screenshot
&lt;/h3&gt;

&lt;p&gt;This is the screenshot I sent the bot: the breadcrumb duplicated the path input, the two sides didn't match, and too many text buttons wrapped onto a second line. The bot came back with a design and shipped it: one symmetric row per side, click the address bar (or press Ctrl+L) to type a path, and rarely used actions tucked into a "More" menu.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Faskw81vpd9064h24b1ta.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Faskw81vpd9064h24b1ta.png" alt="My feedback screenshot of the old SFTP header" width="800" height="418"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  AI chat panel
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fry1p9wkell8xmmdnk3di.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fry1p9wkell8xmmdnk3di.png" alt="AI chat panel" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Real retros
&lt;/h2&gt;

&lt;p&gt;Each of these is taken from &lt;code&gt;docs/retro.md&lt;/code&gt; and matches a real incident.&lt;/p&gt;

&lt;h3&gt;
  
  
  1. "The split button doesn't respond"
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;What I saw:&lt;/strong&gt; the split toggle worked only sometimes; clicking it again after enabling often did nothing.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Root cause:&lt;/strong&gt; a missing &lt;code&gt;useCallback&lt;/code&gt; dependency meant the closure read stale state. Tests only covered a single click.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Rule:&lt;/strong&gt; toggle callbacks that fire repeatedly always read the latest value from a ref. Toggle tests must cover two clicks in a row, three rapid clicks, and acting after switching tabs.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  2. The split redesign took an hour
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;What I saw:&lt;/strong&gt; a 30-minute task took about 60.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Root cause:&lt;/strong&gt; the requirements changed mid-way with no new estimate; the full test suite ran after every small edit; the screenshot script kept turning split view off.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Rule:&lt;/strong&gt; run only related tests while iterating and the full suite once on the final HEAD; re-estimate before acting on a requirement change; screenshot scenarios must be idempotent.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  3. A wall of script after connecting to a server
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;What I saw:&lt;/strong&gt; connecting to a test server echoed the shell-integration bootstrap script into the terminal.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Root cause:&lt;/strong&gt; it was only verified on the demo host, which has no real TTY, so the echo was invisible.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Rule:&lt;/strong&gt; anything written into a remote shell must be checked once in a real interactive bash session for echo and history. Anything meant to be hidden needs an explicit end marker and a render-layer gate.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  4. "All tabs" flashes and disappears
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;What I saw:&lt;/strong&gt; clicking "All tabs" showed the list for a split second.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Root cause:&lt;/strong&gt; the auto-close condition was "close when tabs don't overflow", but the button had since been made always visible — the condition was never updated.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Rule:&lt;/strong&gt; popovers may auto-close only on a state &lt;em&gt;transition&lt;/em&gt;, never on the current state; when changing an entry point's visibility, grep every effect on that state; every popover needs a "stays open after opening" test.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  5. Product kept re-proposing finished features
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;What I saw:&lt;/strong&gt; the product role repeatedly suggested things that already existed.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Root cause:&lt;/strong&gt; different wording from the existing implementation, and dedupe relied on memory.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Rule:&lt;/strong&gt; every proposal must grep docs/iterations, the backlog and src with at least two synonyms, and list the search terms and hits in the proposal.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  6. A context menu covering a modal
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;What I saw:&lt;/strong&gt; the tab context menu stayed on top of the host-key dialog.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Root cause:&lt;/strong&gt; the fix assumed all menus shared one close path, but the tab menu had its own state in App.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Rule:&lt;/strong&gt; before touching popover behavior, grep every popover implementation and list its owner; closing goes through a single broadcast event that every menu must listen to; layering fixes must be verified with an "after" screenshot.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What the human actually does
&lt;/h2&gt;

&lt;p&gt;In this setup I do three things:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Direction:&lt;/strong&gt; decide where the product goes — for example, "only build high-usage features".&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Feedback:&lt;/strong&gt; use the app, take a screenshot, and say "this is messy" or "this doesn't respond".&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Release decisions:&lt;/strong&gt; when to cut a test build and when to push is my call.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Breaking down requirements, design, code, tests and retros are all handled by the bot. The retro loop means each mistake is (ideally) made only once — the rulebook keeps growing, and iterations keep getting steadier.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Website: &lt;a href="https://guoshuaipeng.github.io/ai-ssh-pro" rel="noopener noreferrer"&gt;https://guoshuaipeng.github.io/ai-ssh-pro&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Download: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro/releases&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Feedback is very welcome — and if you're also letting an AI drive your development, I'd love to hear how you run it.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>productivity</category>
      <category>opensource</category>
      <category>electron</category>
    </item>
    <item>
      <title>Debugging an nginx 502 with an AI assistant next to your terminal</title>
      <dc:creator>Shuaipeng Guo</dc:creator>
      <pubDate>Tue, 20 Oct 2026 13:07:27 +0000</pubDate>
      <link>https://dev.to/shuaipeng_guo_79180b50553/debugging-an-nginx-502-with-an-ai-assistant-next-to-your-terminal-2o63</link>
      <guid>https://dev.to/shuaipeng_guo_79180b50553/debugging-an-nginx-502-with-an-ai-assistant-next-to-your-terminal-2o63</guid>
      <description>&lt;p&gt;It's 3 AM, the pager goes off, and the site is answering every request with &lt;strong&gt;502 Bad Gateway&lt;/strong&gt;. You know the drill: SSH in, read the nginx error log, check the upstream, check the port, check memory. None of it is hard. Doing it half-asleep, without skipping a step, is.&lt;/p&gt;

&lt;p&gt;This post walks through that triage the way I now do it in &lt;strong&gt;AI-SSH-Pro&lt;/strong&gt;, an open-source SSH desktop client with an AI assistant docked next to the terminal. The idea is simple: the assistant reads output, proposes the next command and explains it. You decide what actually runs.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Website: &lt;a href="https://guoshuaipeng.github.io/ai-ssh-pro/" rel="noopener noreferrer"&gt;https://guoshuaipeng.github.io/ai-ssh-pro/&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Download: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro/releases&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F4862vhw2fkdwutbjlbed.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F4862vhw2fkdwutbjlbed.png" alt="Terminal on the left, AI assistant on the right, bound to the same tab" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Screenshots use the English UI and the built-in demo host, so all data is simulated. Some sample AI replies in the demo are in Chinese.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  A quick refresher: what a 502 actually means
&lt;/h2&gt;

&lt;p&gt;nginx returns 502 when it is acting as a reverse proxy and gets no valid response from the upstream. In practice it is almost always one of these:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;What you see in &lt;code&gt;error.log&lt;/code&gt;
&lt;/th&gt;
&lt;th&gt;Usual cause&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;connect() failed (111: Connection refused)&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;The upstream process (php-fpm, a Node or Java app) is down, or listening on a different port&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;connect() to unix:/run/php/php-fpm.sock failed (2: No such file or directory)&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Socket path in nginx config doesn't match php-fpm's &lt;code&gt;listen&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;upstream prematurely closed connection&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;The app crashed mid-request, often OOM-killed&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;upstream sent too big header&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Response headers exceed &lt;code&gt;proxy_buffer_size&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;So the job is: find which line you have, then find why.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: describe the symptom, get read-only commands back
&lt;/h2&gt;

&lt;p&gt;Connect to the web host, open the AI panel on the right and just type what you see, e.g. &lt;em&gt;"nginx on web-01 returns 502, help me check"&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;The panel is bound to the terminal tab you're in, so the assistant knows which host it is talking about. A good first answer is a short explanation plus a couple of safe, read-only commands:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;systemctl status nginx &lt;span class="nt"&gt;--no-pager&lt;/span&gt;
&lt;span class="nb"&gt;sudo tail&lt;/span&gt; &lt;span class="nt"&gt;-n&lt;/span&gt; 50 /var/log/nginx/error.log
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Each suggested command shows up in a compact block with &lt;strong&gt;Copy / Run&lt;/strong&gt; and a &lt;strong&gt;risk label&lt;/strong&gt; (low / medium / high). Both of these are read-only, so they're labelled low.&lt;/p&gt;

&lt;p&gt;If the log shows &lt;code&gt;Connection refused&lt;/code&gt;, you already know the upstream isn't answering. That one line saves a lot of guessing.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: follow the thread
&lt;/h2&gt;

&lt;p&gt;Next questions write themselves: is the upstream running, is it listening where nginx expects, did the kernel kill it?&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nb"&gt;sudo &lt;/span&gt;ss &lt;span class="nt"&gt;-ltnp&lt;/span&gt; | &lt;span class="nb"&gt;grep&lt;/span&gt; &lt;span class="nt"&gt;-E&lt;/span&gt; &lt;span class="s1"&gt;':(80|443|8080|9000)\b'&lt;/span&gt;
systemctl status php-fpm &lt;span class="nt"&gt;--no-pager&lt;/span&gt;      &lt;span class="c"&gt;# or your app's unit&lt;/span&gt;
&lt;span class="nb"&gt;sudo &lt;/span&gt;journalctl &lt;span class="nt"&gt;-k&lt;/span&gt; &lt;span class="nt"&gt;--since&lt;/span&gt; &lt;span class="s2"&gt;"1 hour ago"&lt;/span&gt; | &lt;span class="nb"&gt;grep&lt;/span&gt; &lt;span class="nt"&gt;-i&lt;/span&gt; &lt;span class="nt"&gt;-E&lt;/span&gt; &lt;span class="s2"&gt;"oom|killed process"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Instead of reading raw &lt;code&gt;ps&lt;/code&gt; output yourself, you can ask &lt;em&gt;"list the top processes by memory"&lt;/em&gt;. In the screenshot above the assistant proposes:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;ps aux &lt;span class="nt"&gt;--sort&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;-%mem | &lt;span class="nb"&gt;head&lt;/span&gt; &lt;span class="nt"&gt;-n&lt;/span&gt; 11
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It's labelled &lt;strong&gt;Risk: low&lt;/strong&gt;, and because it's read-only it ran right away under the tab's policy ("Ran automatically under the AI execution policy"). The reply then summarizes the output: which process is using the most memory, how much, and whether it's something you should touch.&lt;/p&gt;

&lt;p&gt;If you'd rather paste context than describe it, the input row has a button to &lt;strong&gt;attach terminal output&lt;/strong&gt; to your question, so the assistant reads exactly what you're looking at.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: the fix is a change, so you approve it
&lt;/h2&gt;

&lt;p&gt;Once the cause is clear, the fix usually mutates the host: restart the upstream, correct the socket path, bump a buffer, reload nginx.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nb"&gt;sudo &lt;/span&gt;nginx &lt;span class="nt"&gt;-t&lt;/span&gt; &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="nb"&gt;sudo &lt;/span&gt;systemctl reload nginx
&lt;span class="nb"&gt;sudo &lt;/span&gt;systemctl restart php-fpm
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;This is where the execution policy matters. AI-SSH-Pro has three levels:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Policy&lt;/th&gt;
&lt;th&gt;What happens&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Suggest only&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Read-only commands run; changes are shown as suggestions only&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;strong&gt;Confirm before run&lt;/strong&gt; (default)&lt;/td&gt;
&lt;td&gt;Read-only commands run; changes need your confirmation; high-risk needs a second confirmation&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Auto run&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Low- and medium-risk commands run automatically; high-risk still asks&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;You set it globally in &lt;strong&gt;Settings → AI&lt;/strong&gt;, per host on the host's Safety card, or for the current tab only from the policy menu in the input row. The tab override is in memory only and resets on disconnect.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7dbq29uhlt04jsaghc9x.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7dbq29uhlt04jsaghc9x.png" alt="AI execution policy settings" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;For an incident on a real web node, &lt;strong&gt;Confirm before run&lt;/strong&gt; is the sweet spot: diagnostics flow without friction, and nothing restarts until you click.&lt;/p&gt;

&lt;h2&gt;
  
  
  Guardrails that matter at 3 AM
&lt;/h2&gt;

&lt;p&gt;Tired people make mistakes, so a few safety nets are built in rather than left to discipline:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Production banner.&lt;/strong&gt; Hosts marked Production show a red banner on the tab, and dangerous commands require typing the host name.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Read-only hosts are enforced in the main process.&lt;/strong&gt; If a host is marked Read-only, the AI cannot write to its terminal, even if you click approve. The check happens in Electron's main process, not just in the UI.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The effective policy is visible.&lt;/strong&gt; The host Safety card shows what is actually in effect, e.g. "In effect: Suggest only (limited by read-only guard)".&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3yoao6dgpnl039bs4rxg.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3yoao6dgpnl039bs4rxg.png" alt="Production banner on the terminal tab" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7p2mg779mxxajrmujg49.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7p2mg779mxxajrmujg49.png" alt="Host Safety card with protection level and AI policy" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: keep the trail for the postmortem
&lt;/h2&gt;

&lt;p&gt;Every terminal tab has its own conversation, and history is grouped &lt;strong&gt;per host&lt;/strong&gt;. Next time web-01 acts up, open the history and the previous investigation is right there: which commands ran, what they returned, what fixed it. &lt;strong&gt;+ New chat&lt;/strong&gt; archives the current conversation instead of deleting it.&lt;/p&gt;

&lt;p&gt;AI-initiated commands and dangerous operations also land in the &lt;strong&gt;audit log&lt;/strong&gt;, which you can filter by time and export when writing up the incident.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fz0wv9tc744qc4xmmgy38.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fz0wv9tc744qc4xmmgy38.png" alt="Conversation history grouped by host" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Bring your own model
&lt;/h2&gt;

&lt;p&gt;The assistant talks to any OpenAI-compatible provider (OpenAI, DeepSeek, Qwen, and others). API keys are stored with the OS secure store on your machine and are not uploaded to any product cloud.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it without touching a server
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Grab the installer for Windows, macOS or Linux from &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases" rel="noopener noreferrer"&gt;GitHub Releases&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;Open the built-in &lt;strong&gt;demo host&lt;/strong&gt;. Everything is simulated, so you can click Run freely.&lt;/li&gt;
&lt;li&gt;Add your model provider in &lt;strong&gt;Settings → AI&lt;/strong&gt;, then try it on a real box with &lt;strong&gt;Confirm before run&lt;/strong&gt;.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;More details are on the &lt;a href="https://guoshuaipeng.github.io/ai-ssh-pro/" rel="noopener noreferrer"&gt;project site&lt;/a&gt;. AI-SSH-Pro is MIT-licensed; stars, feedback and &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/issues" rel="noopener noreferrer"&gt;issues&lt;/a&gt; are very welcome.&lt;/p&gt;

&lt;p&gt;How do you triage 502s today? I'd love to hear which checks you run first.&lt;/p&gt;

</description>
      <category>devops</category>
      <category>nginx</category>
      <category>ssh</category>
      <category>ai</category>
    </item>
    <item>
      <title>AI-SSH-Pro: a multi-session SSH client with an AI ops assistant you stay in control of</title>
      <dc:creator>Shuaipeng Guo</dc:creator>
      <pubDate>Tue, 20 Oct 2026 01:43:48 +0000</pubDate>
      <link>https://dev.to/shuaipeng_guo_79180b50553/ai-ssh-pro-a-multi-session-ssh-client-with-an-ai-ops-assistant-you-stay-in-control-of-50ke</link>
      <guid>https://dev.to/shuaipeng_guo_79180b50553/ai-ssh-pro-a-multi-session-ssh-client-with-an-ai-ops-assistant-you-stay-in-control-of-50ke</guid>
      <description>&lt;p&gt;&lt;strong&gt;AI-SSH-Pro&lt;/strong&gt; is a local, multi-session SSH desktop client with a built-in, Cursor-style AI ops assistant. The AI proposes commands; you decide how they run: &lt;strong&gt;Suggest only&lt;/strong&gt;, &lt;strong&gt;Confirm before run&lt;/strong&gt;, or &lt;strong&gt;Auto run&lt;/strong&gt;.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Website: &lt;a href="https://guoshuaipeng.github.io/ai-ssh-pro/" rel="noopener noreferrer"&gt;https://guoshuaipeng.github.io/ai-ssh-pro/&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Download: &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro/releases&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Source (MIT): &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro" rel="noopener noreferrer"&gt;https://github.com/guoshuaipeng/ai-ssh-pro&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;blockquote&gt;
&lt;p&gt;The UI ships in both English and Chinese — switch the language in &lt;strong&gt;Settings&lt;/strong&gt;. Screenshots below use the English UI with the built-in demo host (simulated data; some sample AI replies are in Chinese).&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Who it's for
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Ops / SRE engineers&lt;/strong&gt; watching several machines at once who want one place to see everything.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Backend developers&lt;/strong&gt; who regularly SSH in to read logs or restart services, and want help structuring the troubleshooting.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Small-team leads&lt;/strong&gt; who want AI in the loop, with per-host control over what it may execute.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Xshell / FinalShell users&lt;/strong&gt; looking for familiar tabs and a session tree.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Ops home: every host at a glance
&lt;/h2&gt;

&lt;p&gt;The app opens on an ops dashboard:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Online hosts and open alerts at the top.&lt;/li&gt;
&lt;li&gt;One card per host with live CPU, memory and disk.&lt;/li&gt;
&lt;li&gt;One click into Terminal, SFTP or Monitor.&lt;/li&gt;
&lt;li&gt;Quick actions for common playbooks, such as a rolling nginx restart or disk cleanup.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7s56dj60v2afzts4ksp0.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7s56dj60v2afzts4ksp0.png" alt="Ops home with host metrics" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Multi-session SSH, done properly
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Tabs, split panes and session groups, with search and &lt;code&gt;#tag&lt;/code&gt; filtering.&lt;/li&gt;
&lt;li&gt;Jump hosts (multi-level ProxyJump), port forwarding and host-key verification.&lt;/li&gt;
&lt;li&gt;Import from &lt;code&gt;~/.ssh/config&lt;/code&gt;; reconnect to recent sessions in one click.&lt;/li&gt;
&lt;li&gt;Production tabs carry a prominent banner, so you always know where you are.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3yoao6dgpnl039bs4rxg.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3yoao6dgpnl039bs4rxg.png" alt="Terminal with production banner" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The AI panel
&lt;/h2&gt;

&lt;p&gt;The assistant lives in a panel to the right of the terminal, so you never switch windows mid-incident.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Suggested commands are grouped into one compact block with &lt;strong&gt;Copy / Run&lt;/strong&gt; and a risk label (low / medium / high).&lt;/li&gt;
&lt;li&gt;Each reply shows how long it took and can be copied in one click.&lt;/li&gt;
&lt;li&gt;The input row holds everything: model, execution policy, attach terminal output, send.&lt;/li&gt;
&lt;li&gt;Connect any OpenAI-compatible provider (OpenAI, DeepSeek, Qwen, …). Your API keys stay on your machine.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Three execution policies
&lt;/h2&gt;

&lt;p&gt;Pick one in &lt;strong&gt;Settings → AI&lt;/strong&gt;:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Policy&lt;/th&gt;
&lt;th&gt;Behavior&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Suggest only&lt;/td&gt;
&lt;td&gt;Runs read-only commands; changes are shown as suggestions only&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Confirm before run (default)&lt;/td&gt;
&lt;td&gt;Read-only commands run; changes need confirmation; high-risk needs a second confirmation&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Auto run&lt;/td&gt;
&lt;td&gt;Low- and medium-risk commands run automatically; high-risk still asks&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;ul&gt;
&lt;li&gt;Under Auto run you can additionally allow high-risk commands to run automatically. It's off by default.&lt;/li&gt;
&lt;li&gt;The policy menu in the input row can override the policy for the current tab only.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7dbq29uhlt04jsaghc9x.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7dbq29uhlt04jsaghc9x.png" alt="AI execution policy settings" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Per-host safety
&lt;/h2&gt;

&lt;p&gt;Every host profile has a &lt;strong&gt;Safety&lt;/strong&gt; card:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Protection level: Normal / Production / Read-only.&lt;/li&gt;
&lt;li&gt;AI policy for this host: follow the global setting, or pick one.&lt;/li&gt;
&lt;li&gt;The card shows the policy &lt;strong&gt;actually in effect&lt;/strong&gt;, e.g. "In effect: Suggest only (limited by read-only guard)".&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Production hosts are not capped: your chosen policy applies as-is, and the banner reminds you of the environment. Only &lt;strong&gt;read-only&lt;/strong&gt; hosts block AI writes.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7p2mg779mxxajrmujg49.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7p2mg779mxxajrmujg49.png" alt="Host safety card" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Conversations bound to tabs
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Each terminal tab is its own conversation; two tabs on the same host give you #1 and #2.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;+ New chat&lt;/strong&gt; archives the current conversation instead of deleting it.&lt;/li&gt;
&lt;li&gt;History is grouped by host — click to restore.&lt;/li&gt;
&lt;li&gt;With no terminal bound, the assistant is plain Q&amp;amp;A and never runs anything.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fz0wv9tc744qc4xmmgy38.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fz0wv9tc744qc4xmmgy38.png" alt="Conversation history grouped by host" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  SFTP and the rest of the toolbox
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Dual-pane SFTP&lt;/strong&gt;: local and remote side by side, drag-and-drop, transfer queue.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Monitoring&lt;/strong&gt;: resource charts, multi-host dashboard, alert rules and an alert workbench.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Toolbox&lt;/strong&gt;: processes, ports, services, logs, Docker containers and Compose.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Productivity&lt;/strong&gt;: snippets, playbooks, batch execution, broadcast input, session recording and replay.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Audit&lt;/strong&gt;: sensitive operations are logged, filterable by time and exportable.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F75pghhlja7fbhi84ysmt.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F75pghhlja7fbhi84ysmt.png" alt="Dual-pane SFTP" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxrrm7o2tyztph0fjgq1l.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxrrm7o2tyztph0fjgq1l.png" alt="Tabs and workspace" width="799" height="491"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Security by design
&lt;/h2&gt;

&lt;p&gt;Safety is layered, not a single switch:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Main-process enforcement&lt;/strong&gt;: before the AI writes to a terminal, the main process re-checks host protection. On a read-only host, the AI cannot write even if you click approve.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;One rule everywhere&lt;/strong&gt;: settings, host card, AI input and background execution all compute the effective policy the same way — no "UI says confirm, backend just runs".&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Bounded overrides&lt;/strong&gt;: per-tab overrides live in memory only, reset on close or disconnect, and still respect read-only.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Credentials stay local&lt;/strong&gt;: passwords, key passphrases and API keys are encrypted with the OS secure store and never uploaded to any product cloud.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Traceable&lt;/strong&gt;: AI-initiated commands and dangerous operations are recorded in the audit log.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Get started
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Download the latest installer from &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/releases" rel="noopener noreferrer"&gt;GitHub Releases&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;Launch it and try the built-in &lt;strong&gt;demo host&lt;/strong&gt; — all data is simulated, nothing touches a real server.&lt;/li&gt;
&lt;li&gt;Add your own model provider in &lt;strong&gt;Settings → AI&lt;/strong&gt;.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;If you want a comfortable multi-session SSH client where the AI proposes and &lt;strong&gt;you&lt;/strong&gt; decide how it executes, give AI-SSH-Pro a try. Stars and &lt;a href="https://github.com/guoshuaipeng/ai-ssh-pro/issues" rel="noopener noreferrer"&gt;issues&lt;/a&gt; are very welcome.&lt;/p&gt;

</description>
      <category>devops</category>
      <category>ssh</category>
      <category>opensource</category>
      <category>ai</category>
    </item>
  </channel>
</rss>
