<?xml version="1.0" encoding="UTF-8"?>
<?xml-stylesheet type="text/xsl" href="assets/rss.xsl"?>
<rss version="2.0"
    xmlns:content="http://purl.org/rss/1.0/modules/content/"
    xmlns:wfw="http://wellformedweb.org/CommentAPI/"
    xmlns:dc="http://purl.org/dc/elements/1.1/"
    xmlns:atom="http://www.w3.org/2005/Atom"
    xmlns:sy="http://purl.org/rss/1.0/modules/syndication/"
    xmlns:slash="http://purl.org/rss/1.0/modules/slash/"
    xmlns:itunes="http://www.itunes.com/dtds/podcast-1.0.dtd"
    >
<channel>
    <title>exe</title>
    <atom:link href="https://blog.v2core.com/rss.xml" rel="self" type="application/rss+xml" />
    <link>https://blog.v2core.com/</link>
    <description><![CDATA[
    <p>A personal VM cloud with a Mac OS 9 face, and what gets built on it.</p>

    ]]></description>
    
    
    <item>
        <title>exe 2026.11.11: VMs that outlive a restart, an Intel Mac, and two new apps</title>
        <link>https://blog.v2core.com/exe-2026-11-11/</link>
        <guid>https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/</guid>
        <pubDate>Wed, 11 Nov 2026 05:22:01 -0800</pubDate>
        
        <itunes:image href="https://blog.v2core.com/exe-2026-11-11/activity-monitor.png" />
        
        
        <description><![CDATA[
            <p>exe 2026.11.11, released today, keeps your VMs running on Linux when the exe daemon restarts. It also brings a virtual Intel Mac that runs Tiger under KVM, two new apps, Solana and Activity Monitor, emoji reactions in the Hub, and Workspace files of up to 1 GB.</p>
<h2>VMs keep running when exe restarts</h2>
<p>Until this release, a restart of the daemon powered off every running VM and booted it again, and every update restarts the daemon. With a public site in a VM, 41 seconds went by on 2026-11-03 before the guest answered SSH again.</p>
<p>Now, on Linux where systemd runs exe, the VMs keep running when the daemon restarts: after an update, after <strong>Restart Daemon…</strong> on the <strong>Daemon</strong> tab of <strong>Windows → Configuration</strong>, and after <code>systemctl --user restart exe</code>. The next daemon takes each running VM over as it is, and a site published from a VM answers again as soon as the daemon is back.</p>
<p>The unit's <code>KillMode=process</code> leaves each VM's Firecracker process running when the daemon goes, and the next daemon checks that it is still that process before it adopts it.</p>
<p>In the release run, on a Linux x86-64 workstation with KVM, Software Update moved a test build of this release to 2026.11.11 while a test VM ran. The daemon was back in about 3 seconds, and the guest's boot time did not change. The old daemon logged that it was leaving the VM "running for the next daemon". On the machine that builds exe, the restart after publishing left six VMs running, and the new daemon adopted each one.</p>
<p>A shutdown of the host still stops the VMs first. <code>systemctl --user stop exe</code> leaves them running, so use <code>exe stop</code> first when they should not run without the daemon.</p>
<p>Three limits remain:</p>
<ul>
<li>The update from 2026.10.28 itself still stops and starts the VMs once, because the daemon that stops is the old one.</li>
<li>On macOS and on Windows, the VMs still stop and start with the daemon.</li>
<li>The Software Update window, the <strong>Restart Daemon…</strong> button, <code>exe update</code> and <code>exe help</code> still say that a restart stops the running VMs. On Linux under systemd, those words are out of date.</li>
</ul>
<p>See <a href="https://exe.v2core.com/docs/using/virtual-machines#when-exe-restarts">When exe restarts</a>.</p>
<h2>Mac OS X Intel: Tiger on the machine's own cores</h2>
<p>exe's Mac OS X app runs Tiger for PowerPC, which an x86-64 host emulates instruction by instruction, though its own processor could run Tiger for Intel. <strong>Mac OS X Intel</strong> is a third Mac that does, on a node that runs Linux on x86-64: Mac OS X 10.4.11 for Intel Macs, under KVM, on two of the machine's own cores at their own speed.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/mac-os-x-intel.png" alt="The Intel Mac's own screen: Tiger's blue desktop, the Dock and Macintosh HD, with About This Mac open. It reads Mac OS X Version 10.4.11, Processor 2.2 GHz Unknown, 16 MB L3 cache and Memory 2 GB RAM." /></p>
<p>About This Mac calls the processor Unknown, whatever the host's processor is: exe presents the two cores to Tiger as one fixed Core 2 model. This Mac gains speed, and Rosetta for PowerPC Mac OS X programs. It has no Classic, no sound and no Quartz Extreme, and no agent, whose Chat works in the PowerPC Mac.</p>
<p>It needs KVM, and Linux with glibc 2.35 or later, as from Ubuntu 22.04 and Debian 12 on. exe does not fall back to software emulation, which is about five times slower. The Mac takes 2 GB of memory and two cores while it runs.</p>
<p>Choose <strong>Special → Mac OS X Intel</strong>. Setup downloads exe's own x86 emulator, OpenCore from GitHub, and Apple's install DVD (2.2 GiB) from the Internet Archive, and checks each against its checksum. A guide above the Mac's screen takes you through Disk Utility and the Installer. The window is the Mac OS X app's, without <strong>Agent</strong> and sound, and the pointer is where your browser's pointer is, from the Installer on.</p>
<p>OpenCore, which starts Tiger on a machine that is not a Mac, also fixes Tiger's kernel at each start. Without the first fix, Tiger crashes within minutes on this virtual firmware. Other fixes let Tiger use two cores, and let an idle core rest. Tiger's Software Update still finds Apple's updates, but installing them is untested, and an update that replaces the kernel loses exe's fixes. See <a href="https://exe.v2core.com/docs/using/mac-os-x-intel">Mac OS X Intel</a>.</p>
<h2>Workspace files up to 1 GB</h2>
<p>A Workspace file can now be 1 GB, up from 100 MB, by upload, drop and the API, and joined desks sync files of that size. Node Sync used to hold each file whole in memory on both nodes, and now it moves a file from disk to disk. In the change's own test, a 900 MB file crossed between two test daemons, with each daemon at a peak of 40 MB of memory. A desk on 2026.10.28 or earlier gets no file over 100 MB until it is updated.</p>
<p>Until this release, an upload said nothing until it was done. Now uploads say how they go, a row a file.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/workspace-uploads.png" alt="A Workspace window during an upload of several files: rows under the title bar with f2.bin at 100%, f3.bin Failed, f4.bin Stopped, and a long-named .mov at 32% with a Stop button and a thin blue line under its row." /></p>
<p>Each row has the file's name, how far along it is ("40%", or "Waiting"), a <strong>Stop</strong> button, and a thin blue line that fills as it goes. <strong>Stop</strong> skips a file that waits and cuts off a file that is going, with nothing of it kept.</p>
<p>Bring files in with <strong>File → Upload to Workspace…</strong>, with <strong>Upload…</strong> in the menu of a window's empty space, or with a drop. See <a href="https://exe.v2core.com/docs/using/workspace-and-files">Workspace and files</a>.</p>
<h2>Emoji reactions in the Hub</h2>
<p>The Hub app now gives a post an emoji reaction. Under a post, each reaction is a small button with its count. Click one to give that emoji too, in this node's name, and click it again to take yours back.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/hub-reactions.png" alt="Part of the Hub app on a thread: a reaction, 🙏 1, under a post, and the smiley button pressed. The emoji picker is open, with a Search field, Frequently Used, Smileys and More, and Folded Hands named at its foot." /></p>
<p>For a different emoji, click the smiley button at the right of the post's foot line. A menu opens, and its field finds any emoji by its name: "party" finds 🎉. Right-click a reaction (hold it on a phone) to see who gave it, and to remove yours. A reaction is signed by this node's key, like a post, so nothing asks you to confirm it. See <a href="https://exe.v2core.com/docs/using/the-hub#react-to-a-post">React to a post</a>.</p>
<h2>Solana: a trading chart for the ticker's tokens</h2>
<p><strong>Solana</strong> is a new app: a trading chart of SOL, PUMP, MET or SKR, the tokens of the Control Strip's ticker, drawn from Coinbase's candles. Open it from its desktop icon, or with <strong>Open Solana…</strong> in the Solana ticker's menu.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/solana.png" alt="The Solana window on SOL at 1h as candles, with Vol, RSI and BB pressed: the price $83.87, candles from Nov 7 to Nov 11 with Bollinger Bands, a red dot on a candle marked −6.0%, volume bars and the RSI 14 pane." /></p>
<p>The four buttons at the top pick the token. The row at the chart's top left sets what a candle covers, from <strong>1m</strong> to <strong>1d</strong>, and draws <strong>Line</strong> or <strong>Candles</strong>. <strong>Vol</strong> and <strong>RSI</strong> show the two panes under the price, and <strong>BB</strong> the Bollinger Bands over it, as TradingView draws them. Drag the chart to the right for older candles, and press <strong>Now</strong> to come back.</p>
<p>Each big move of the token (the next section says what counts as one) stands on its candle as a dot, red down and green up, with its change beside it. Open the app at <code>/apps/solana/</code> on the desktop's host, and its address follows the chart, such as <code>#coin=PUMP&amp;tf=15m</code>, so a bookmark opens on that chart. See <a href="https://exe.v2core.com/docs/using/solana">Solana</a>.</p>
<h2>Big moves stay until you close them</h2>
<p>On an open desk, a big move of one of the ticker's tokens used to show as a toast of 3.5 seconds, easy to miss and gone on a reload. Now each one is a notice that stays on every desk and phone until you press its close box. Its second line begins with the time of the move, and a newer move of the same token takes its place.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/price-notice.png" alt="A yellow notice on the desk with a close box: SOL +6.9% in the last hour, and under it Oct 29, 05:38 PM, $93.68, from $87.66, 1st of 4 today." /></p>
<p>Click the words of a notice, and the Solana app opens on that move, ringed on its chart. A line of <strong>Recent Moves</strong> in the ticker's menu does the same. A move is big when it passes the ticker's thresholds, such as 2.5% in an hour or 6% in a day for SOL. See <a href="https://exe.v2core.com/docs/using/the-desktop#notifications-big-price-moves">Notifications: big price moves</a>.</p>
<h2>Activity Monitor: the machine as pixel art</h2>
<p><strong>Activity Monitor</strong>, the second new app, draws a node's CPU, graphics cards and network ports as pixel art, every 2 seconds. It shows this node or any node joined to it, on Linux, macOS and Windows. Open it from its icon on the desktop, and pick a node with the <strong>Node</strong> pop-up, where each node wears its icon.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/activity-monitor.png" alt="Activity Monitor on a node with 20 cores and one graphics card: the CPU drawn as a chip with 20 tiles and a history a core to a row, an NVIDIA GB10 drawn as a board, and a 2.5 Gb/s Realtek RTL8127 port at the top of Network." /></p>
<p>The CPU is a chip with a tile for each core, and each thread in a tile lights like a level meter, from green to amber to red. Beside the chip are the load, the hottest sensor, memory, <strong>Load average</strong> and <strong>Uptime</strong>. Each NVIDIA graphics card, read through <code>nvidia-smi</code>, is its board: its die lights as the GPU works, and its fan turns as fast as the card's own. A Mac shows the Apple chip's GPU. The network is the machine's back panel, with a jack for each wired port and an antenna for Wi-Fi.</p>
<p>Under each picture is a history, and <strong>Each pixel</strong>, at the top right, sets what a pixel stands for, from <strong>2 seconds</strong> to <strong>1 day</strong>. The daemon keeps that history in <code>~/.exe/activity</code>, for 2 years by the hour and with no limit by the day.</p>
<p>A Mac shows no clocks, power or Wi-Fi signal, which macOS keeps behind an administrator's password. Another node must run this release or later. Joining now lets each node read how busy the other is, and nothing more beyond the sync. See <a href="https://exe.v2core.com/docs/using/activity-monitor">Activity Monitor</a>.</p>
<h2>Notes: more corkboards, and a note as a first prompt</h2>
<p>Notes, one of the extra apps, now has more than one corkboard. The <strong>Corkboard</strong> button has a menu arrow at its right end, and the button's name is the board that is up.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/notes-corkboards.png" alt="Notes with the Corkboard button's menu open: Corkboard (checked), Work and Recipes, then New Board…, Rename Board… and Delete Board…, over a corkboard of eight pinned notes." /></p>
<p>Press the arrow to switch boards, or to choose <strong>New Board…</strong>, <strong>Rename Board…</strong> or <strong>Delete Board…</strong>. The last two are dimmed on the first board, which cannot be renamed or deleted. To move a note, right-click its sheet on the board and choose <strong>Move to</strong> and the board's name. A board that you delete takes no note with it: its notes go back to the first board.</p>
<p>A note can also start an agent's session. With a note open, press the button with the arrow leaving a tray, left of <strong>Delete</strong>, and choose <strong>Send to Claude Code</strong> or <strong>Send to Codex</strong>. The button is there only when exe has found Claude Code or Codex on this machine.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/notes-send-to-agent.png" alt="Part of a Notes window with a note open: the menu of the toolbar's share button, left of Delete, with Send to Claude Code and Send to Codex." /></p>
<p>The agent's window asks where to start, as <strong>New…</strong> does. <strong>Start</strong> begins a session there with the note's text as its first message.</p>
<h2>Claude Code and Codex: put a session away, choose its folder</h2>
<p>A session that waits at its prompt still holds a few hundred megabytes. <strong>Archive…</strong> gave that memory back, but it takes the row out of the column. <strong>Put Away</strong>, on a session row's menu in the Claude Code and Codex windows, ends the session's process and keeps the row.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/put-away.png" alt="The Claude Code window with a session row's menu open: Started today, 1:34 AM, New Session…, Copy Title, Get Info, Put Away and Archive…." /></p>
<p>The row stays in its place under a hollow dot, and a note at the foot of the screen says what the session gave back.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/put-away-after.png" alt="The Claude Code window after Put Away: the row Weather radar tiles under a hollow dot, and a note at the foot of the screen that reads Put away “Weather radar tiles”: 185.2 MB freed. 58.9 GB available now." /></p>
<p>Click the row to continue that conversation, in the folder it was started in. For a session that is still working, the item reads <strong>Put Away…</strong> and asks first, because the turn it is working on is lost. The daemon also puts away a session that nobody has touched for 7 days, with or without a desk open, but not one that a window shows, that works, or that waits on a permission or a question. It looks five minutes after it starts, then once an hour, so a session left alone for a week already is put away soon after this update. The Newsfeed says which sessions it put away. To change the days, set <code>put_away_days</code> in <code>~/.exe/config.json</code> and restart the daemon. <code>0</code> turns it off. See <a href="https://exe.v2core.com/docs/using/claude-code-codex-and-terminal#put-a-session-away">Put a session away</a>.</p>
<p><strong>New…</strong> now asks which folder to start in, in both windows. The New Session dialog lists the folders the agent worked in on this machine, latest first.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/new-session.png" alt="The New Session dialog in the Claude Code window: a Folder field that holds /www/exe, over a list of recent folders with dates, &quot;16 recent folders&quot;, and the Cancel and Start buttons." /></p>
<p>The latest folder is already in the <strong>Folder</strong> field. Click a row to put another there, double-click a row to start there at once, or type any path, with <code>~</code> for your home folder. See <a href="https://exe.v2core.com/docs/using/claude-code-codex-and-terminal#choose-the-folder-for-a-new-session">Choose the folder for a new session</a>.</p>
<h2>An icon for each node</h2>
<p>Every desk on a phone's Home Screen wore the same icon, exe's computer. Now each node can have its own, so several desks no longer look alike.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/node-icon.png" alt="Windows → Configuration → General: the Language pop-up and the Icon well with exe's computer and 64 pixel-art icons. Acoustic Modem is chosen, and its teal tile is previewed at the right." /></p>
<p>Choose <strong>Windows → Configuration</strong>, click the <strong>General</strong> tab, and under <strong>Icon</strong> click one of the 64 pictures. Click <strong>Save</strong>. The browser tab shows the new icon at once. A phone keeps the icon a desktop had when it was added, so remove the desktop from the Home Screen and add it again.</p>
<h2>Smaller changes and fixes</h2>
<ul>
<li><strong>Desktop over HTTPS</strong>, in the Control Strip's Tailscale menu, serves the desk at this machine's HTTPS address on the tailnet, through Tailscale Serve, and stops again. It changes only its own Serve rule.</li>
<li><strong>Welcome</strong> offers to install tmux where it is missing, so an agent's session outlives its window. <strong>Install tmux…</strong>, on the Claude Code and Codex card, installs it in a Terminal window, with <code>sudo</code> on Linux or Homebrew on a Mac.</li>
<li><strong>Search</strong>, the magnifier, finds a thing that holds every word you type, in any order. <code>exe-hub idea</code> finds a note with <code>exe-hub</code> in its first line and <code>idea</code> further down.</li>
<li>A link can open an app. Put <code>#show=app:</code> and the app's ID after the desktop's address, such as <code>/#show=app:Ping</code>. The Macs do not open from a link, because opening one starts its Mac.</li>
<li>Fixed: a stopped Codex session keeps its row when its conversation cannot be found. The wheel scrolls Codex's history when Codex runs inside a Terminal window, and text selected there reaches the browser's clipboard. A tunnel rule for one host goes ahead of a wildcard rule, so <code>exe expose</code> of a host under a wildcard is reached.</li>
</ul>
<h2>Get the update</h2>
<p>An installed exe moves to this release with <strong>Software Update…</strong> in the Apple menu. Press <strong>Update Now</strong>, then <strong>Update</strong>.</p>
<p><img src="https://blog.v2core.com/207DBC1F-977E-4E06-A0F7-85A60823E8E8/software-update.png" alt="The Software Update window of an exe on 2026.10.28: &quot;exe 2026.11.11 is available. This machine runs 2026.10.28.&quot;, Installed 2026.10.28, Newest 2026.11.11, the first notes under &quot;What changed in 2026.11.11&quot;, and the Update Now and Check Now buttons." /></p>
<p>In the release run, an install of 2026.10.28 was on 2026.11.11 about 9 seconds after <strong>Update Now</strong>. From a shell:</p>
<pre><code>exe update
</code></pre>
<p>To install exe on Linux, or on macOS 13 or later, on x86-64 or ARM64:</p>
<pre><code class="language-sh">curl -fsSL https://exe.v2core.com/install.sh | sh
</code></pre>
<p>On Windows 11 on x86-64, in PowerShell:</p>
<pre><code class="language-powershell">irm https://exe.v2core.com/install.ps1 | iex
</code></pre>
<p>The Windows build is not signed, as before. Not tried for this release: the update from 2026.10.28 with a VM running, an update on a Mac, a new install on Windows, and Windows 10. The macOS build for Intel Macs was only started under Rosetta, on an Apple silicon Mac. The release holds 74 commits since 2026.10.28, and its <a href="https://github.com/livid/exe/releases/tag/2026.11.11">page on GitHub</a> has every file and their checksums.</p>

        ]]></description>
    </item>
    
    <item>
        <title>exe 2026.10.28: passkeys, a desk in three languages, and the Workspace as a drive</title>
        <link>https://blog.v2core.com/exe-2026-10-28/</link>
        <guid>https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/</guid>
        <pubDate>Wed, 28 Oct 2026 08:30:00 -0700</pubDate>
        
        <itunes:image href="https://blog.v2core.com/exe-2026-10-28/api-token-use-passkey.png" />
        
        
        <description><![CDATA[
            <p>exe 2026.10.28 signs a browser in with a passkey and keeps the API token off every address. The desk now speaks Simplified Chinese and Japanese, your own computer mounts the Workspace as a drive, and Ping, a new extra app, graphs your network as SmokePing does.</p>
<p>This release comes two weeks and 73 commits after 2026.10.15.</p>
<h2>Sign in with a passkey</h2>
<p>The API token is the password to your whole node, and every new browser asks for it. On a phone it is a long string to type. A passkey now signs a browser in instead, with Face ID, Touch ID, Windows Hello or a security key, and you type nothing.</p>
<ol>
<li>Open the desktop at a name over HTTPS, such as its Tailscale Serve address.</li>
<li>Choose <strong>Windows → Configuration</strong>, click the <strong>Passkeys</strong> tab, then click <strong>Add Passkey…</strong>.</li>
<li>In another browser that can reach that passkey, through your passkey manager or a security key, open the same address. The API Token dialog now offers <strong>Use Passkey…</strong>.</li>
</ol>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/api-token-use-passkey.png" alt="The API Token dialog a new browser gets, which asks for the API token, hints at running exe token, and says the token is kept in this browser. It has an API token field and the buttons Use Passkey…, Cancel and OK." /></p>
<p>A browser that signs in with a passkey keeps no token. It stays signed in until you remove the passkey or the token changes, or for 400 days, after which it asks for the passkey again. A passkey on a phone was not tried for this release.</p>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/configuration-passkeys.png" alt="Windows → Configuration on its Passkeys tab: a green lamp and Passkeys: 1, a sentence that says a passkey signs a browser in to exe in place of the API token and works at localhost only (this desk was opened at localhost), one passkey from Chrome on Linux with a Remove button, and Add Passkey…." /></p>
<p><strong>Remove</strong> beside a passkey, pressed twice, ends it and signs out every browser it signed in, at once. Their terminal windows say "signed out", and their Terminal sessions stay in tmux for the next sign-in.</p>
<p>A passkey works only at a name over HTTPS or at localhost, never at an IP address, and only at the address it was made at. Passkeys also need an API token: the token adds the first passkey, and it gets you back in when every passkey is gone. See <a href="https://exe.v2core.com/docs/using/configuration#sign-in-with-a-passkey">Sign in with a passkey</a>.</p>
<h2>The token stays off every address</h2>
<p>Until this release, the desk wrote the API token on every address that cannot carry a header: app windows, terminals, movies and links. There it showed in the address bar and the history, and a copied link carried the password with it.</p>
<p>Now, each time the desktop loads, it trades the token for a cookie that only exe's own pages can use and no script can read. App windows, apps opened from <code>/apps/</code>, terminals and movies get in with that cookie, so no address carries <code>?token=</code>.</p>
<p>A new token ends every cookie at once. Every open terminal window says "signed out", the desktop asks for the token again, and the Terminal sessions wait in tmux. So when a token may have leaked, a new one signs out every browser, and where the machine has tmux no shell is lost.</p>
<p>curl and agents still send the token in a header, and apps need no change. Joined desks never carried it: each request between them is signed with the desk's own key. In a browser that keeps no cookies, the token goes on those addresses, as before.</p>
<h2>The desk in three languages</h2>
<p>If you read Chinese or Japanese faster than English, the desk's menus, windows and messages now come in your language. <strong>Language</strong>, on the new <strong>General</strong> tab of <strong>Windows → Configuration</strong>, sets English, 简体中文 or 日本語. Choose one and press <strong>Save</strong>.</p>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/desk-in-chinese.png" alt="The Icon Editor on a desk set to Simplified Chinese, titled 图标编辑器, with a count of 58 icons and a 新建图标… button. Icons carry Chinese names such as 终端, 软件更新 and 废纸篓, beside names that stay as they are: VM, Claude Code and Codex." /></p>
<p>Every screen that has the desk open loads it again in that language. A screen with unsaved work waits until the work is saved, so a draft on another desk is not lost. The desk is drawn in the language from its first frame, and dates and times follow it.</p>
<p>Your own text stays as you wrote it: file names, notes, chats and the desktop menu. The apps in windows, such as Welcome and Blue Pencil, and the manual are in English for now. In <code>config.json</code> the key is <code>language</code>: <code>en</code>, <code>zh-Hans</code> or <code>ja</code>. See <a href="https://exe.v2core.com/docs/using/configuration#the-language-of-the-desk">The language of the desk</a>.</p>
<h2>The Workspace as a drive on your own computer</h2>
<p>The Workspace is the folder you, your agents and your apps share. A drag onto the desk brings a file in, and <strong>Download</strong> takes one file out at a time. Now the Workspace mounts on your computer as a drive named Workspace. What you drop into it is on the desk at once, and what the desk or an agent saves there is on your computer.</p>
<p>Choose <strong>Help → Mount Workspace…</strong>. The window gives the address, the name and the password, with <strong>Copy</strong> beside each, and the steps for macOS, Windows or Linux. <strong>Connect from</strong> picks the address: the desk's own https address when you reached the desk that way, or this machine's address on the network.</p>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/mount-workspace.png" alt="The Mount Workspace window: an introduction, a Connect from pop-up, and the Address, Name and Password fields, each with a Copy button, over New Password. Under them are the tabs macOS, Windows and Linux, with macOS chosen and its three steps shown." /></p>
<ul>
<li>On a Mac, choose <strong>Go → Connect to Server…</strong> in the Finder, enter the address, click <strong>Connect</strong>, and sign in as <code>workspace</code> with the password.</li>
<li>On Windows, choose <strong>Map network drive</strong> under <strong>This PC</strong> in File Explorer, enter the address as the folder, select <strong>Connect using different credentials</strong>, and sign in the same way. Windows needs a registry line before it takes a password over plain http, and another before it lets files over 50 MB through. The window gives both, ready to copy.</li>
<li>On Linux, GNOME Files takes the address as <code>dav://</code> and Dolphin as <code>webdav://</code>, and a shell mounts it with davfs2.</li>
</ul>
<p>The password is the drive's alone, never the API token, so a computer that mounts the Workspace can read and write that folder and nothing else. When a computer should lose its access, <strong>New Password</strong> makes another, and every computer that mounted signs in again.</p>
<h3>Mac OS X in exe mounts it too</h3>
<p>Until this release, exe's Macs and the desk shared no folder. A file went in on a CD image, or on Mac OS X through the Agent or Remote Login, and it came back out only through a shell. Now the Mac OS X window's <strong>Setup…</strong> menu has a <strong>Mount Workspace</strong> page, and in the Mac's Finder, <strong>Go → Connect to Server…</strong> takes its address, with no software added to Mac OS X 10.4. A file the Mac saves there, a game's save or a screenshot, is in the Workspace before you change windows. Mac OS 9 has no WebDAV of its own, so it still takes a CD. See <a href="https://exe.v2core.com/docs/using/workspace-and-files#mount-the-workspace-on-your-own-computer">Mount the Workspace on your own computer</a>.</p>
<h2>Files up to 100 MB</h2>
<p>A file in the Workspace or in an app's data could be 10 MB at most. Now it can be 100 MB: through <strong>File → Upload to Workspace…</strong>, <strong>Upload…</strong> in a Workspace window, a drop on a Claude Code, Codex or Terminal window, the Workspace's HTTP calls and app data. A movie to watch on the desk, or a log to hand an agent, now fits.</p>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/upload-to-workspace.png" alt="The Upload to Workspace dialog: files land in the Workspace folder shared with apps and VMs, 100 MB max. Choose File… has a 47.7 MB file chosen, with the buttons Cancel and Upload." /></p>
<p>Joined desks sync files at that size. A desk on 2026.10.15 or earlier takes no file over 10 MB, so exe does not send it a bigger one, and that desk gets the file once it is updated. A transfer gives up only after 15 seconds with no bytes moving, so a slow link does not fail a large file because it takes long.</p>
<h2>Ping: your network as SmokePing draws it</h2>
<p>Is the server slow, or your router, or your provider? Does the line drop packets? Ping, a new extra app, shows the answer, now and over the last year. Every 5 minutes on the clock, exe sends 20 pings to each address on the list, keeps the results in RRDtool files in SmokePing's own layout, and draws SmokePing's graphs from them. The daemon does the measuring, so it goes on while the window is closed.</p>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/ping.png" alt="The Ping window with an Add field and one destination open: Cloudflare DNS at 3.0 ms with no loss, its Name and Address fields, and the Last 3 Hours SmokePing graph of 20 pings every 300 seconds. Under it is the top of the Last 30 Hours graph." /></p>
<p>Type an IP address or a host name in the field at the top and press <strong>Add</strong>. Its first 20 pings go out at once, so you see a figure without waiting for the next round. Each row shows the median of the last round and how many of the 20 pings were lost, over a graph of the last 10 hours.</p>
<p>Click a row for SmokePing's four graphs: the last 3 hours, 30 hours, 10 days and 360 days. The grey smoke is each round's 20 pings, darker nearer the median, and the median line takes the colour of the round's loss: green for none, red for 11 to 19 of the 20. Drag across a graph to zoom in, and <strong>Back</strong> returns it.</p>
<p>Ping needs RRDtool: <code>sudo apt install rrdtool</code> on Debian and Ubuntu, <code>brew install rrdtool</code> on a Mac. Until it is there, the window says so and nothing is measured. Where fping is installed, the pings go out as SmokePing sends them. Without it, Ping uses the system's <code>ping</code>. exe looks for both at every round, so no restart is needed. Each machine measures from where it stands, so joined desks do not share the list.</p>
<p>Where the extra apps are off, <strong>Extras…</strong> on the Apps card of <strong>Help → Welcome…</strong> turns them on. See <a href="https://exe.v2core.com/docs/using/apps#ping">Ping</a>.</p>
<h2>Claude Code and Codex</h2>
<h3>After the host restarts</h3>
<p>A restart of the daemon leaves Claude Code and Codex sessions running, but a restart of the host ends them all. Until this release the session column then forgot them, though nobody had archived them and the conversations were still on disk.</p>
<p>Now each session the restart stopped comes back as a stopped row, with its name, title and day, and a hollow dot. Click it, or choose <strong>Continue Here</strong> on its menu, and the agent starts again on that conversation, in the same folder, under the same session name.</p>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/sessions-after-restart.png" alt="The top of the Claude Code window after a host restart: sessions under Today and Yesterday, most of them stopped, each with a hollow dot before its title. A row's menu is open: Started today 1:40 AM, Continue Here, New Session, Copy Title, Get Info, Archive… and Continue All Stopped Sessions…." /></p>
<p>The window you open after the restart continues the session it showed last. The other rows wait for a click, since each CLI holds hundreds of MB of memory. <strong>Continue All Stopped Sessions…</strong> continues them all after a dialog, and <strong>Archive…</strong> removes a row. A session with no conversation yet, or one that ended before the restart, does not come back. See <a href="https://exe.v2core.com/docs/using/claude-code-codex-and-terminal#after-the-host-restarts">After the host restarts</a>.</p>
<h3>Sessions grouped by day</h3>
<p>Until this release, the Claude Code column on a busy machine held some 140 sessions, newest first, with nothing to show where a day ended. The window now groups them by the day they started, under rows such as <strong>Today</strong>, <strong>Yesterday</strong> and <strong>Tue, Oct 20</strong>, each with a triangle and a count.</p>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/sessions-by-day.png" alt="The Claude Code window with its sessions grouped by day: Today open with two sessions, and Yesterday, Tue, Oct 20 and Sep 18, 2025 closed, each with a count. The buttons are New and Search. Under a line, one conversation started outside the desk." /></p>
<p>Click a day to show or hide it, or right-click it for <strong>Expand All</strong> and <strong>Collapse All</strong>. Every day is open until you close it, and this browser remembers the days you close. A closed day shows a black dot when one of its sessions waits for you, or a green dot when one is working, so a session that needs you never hides. The Codex window keeps its plain list.</p>
<h3>One window on every desk, at one size</h3>
<p>Where the machine has tmux, a Terminal, Claude Code or Codex window you open on one computer now opens on your others, at the same size in columns and rows, and its close box closes it on all of them. A phone still shows one window at a time and keeps its own.</p>
<p>Such a window used to belong to the browser that opened it. The reason was the size of the tmux session: tmux sizes a session to the client that last typed or changed size, and two desks with different fonts get different columns and rows from one window. Now the layout carries each window's size in columns and rows, and every desk sizes the window to hold them in its own font. Two desks no longer resize each other's session. On a machine without tmux, such as Windows, an agent window stays the browser's own, because there each window is a fresh run of the CLI.</p>
<h3>A key row on a phone</h3>
<p>A phone's keyboard has no Esc and no Tab, so the Claude Code window on a phone had no way to send them. Now a row of keys under the terminal sends <strong>Esc</strong>, <strong>Tab</strong>, <strong>Shift-Tab</strong> and the four arrows. Esc stops Claude Code's turn or clears its composer, Shift-Tab cycles its modes, and the arrows walk a menu or the prompt history.</p>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/phone-key-row.png" alt="The foot of the Claude Code window on a phone: a row of keys under the terminal, Esc, Tab, Shift-Tab and the four arrows, over the status line connected." /></p>
<p>The keyboard stays up while you tap the keys. Codex and Terminal windows have the row too. See <a href="https://exe.v2core.com/docs/using/the-desktop#on-a-phone">On a phone</a>.</p>
<h2>Mac OS 9 and Mac OS X on exe's own emulator</h2>
<p>Until this release, three things made exe's Macs harder to use. Outside Ubuntu 24.04 you installed QEMU yourself, sound needed a runtime installed by hand, and over a network the screen was slow, because each frame waited for a round trip.</p>
<p>Now, when a Mac prepares its emulator, exe downloads its own build, about 6 MB: on Linux with glibc 2.35 or later (x86-64 or ARM64), on macOS 11 or later, and on Windows 10 or later (x86-64). It is QEMU with the Mac's sound hardware and a screen paced for your browser, and Mac OS 9 and Mac OS X share it. exe checks the download's size and SHA-256 before it uses it. A Mac set up with an earlier emulator gets the new one the next time it starts.</p>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/mac-os-9.png" alt="Inside the Mac OS 9 window: the app's buttons Setup details, Mac keys…, Full screen, Sound off and CD… over a running Mac OS 9 desktop with its menu bar, a yellow note and the Control Strip. The status line reads Your Mac is running, PowerPC G4, 512 MB, Ethernet NAT, Connected." /></p>
<p>For sound, click <strong>Sound off</strong> in the window's row of buttons. It changes to <strong>Sound on</strong>, and the Mac's sound plays in your browser, through the same signed-in connection as the screen. On an emulator the system supplied, without the Mac's sound hardware, the button is dimmed.</p>
<p>For the screen, exe now keeps requests waiting ahead of the viewer, so frames keep coming while one crosses the network. In a test at 50 ms each way, Mac OS 9 drew 62 frames a second, up from 7. See <a href="https://exe.v2core.com/docs/using/mac-os-9#sound">Sound</a>.</p>
<h2>Notices: a toast that stays</h2>
<p>A notice is for news that should wait on the desk until someone has seen it. The request behind it was a place to say that a usage limit had stopped an agent's work: on the desk, not in a public thread.</p>
<p>A script on the exe machine puts one up with a POST to <code>/v1/notices</code>. It shows above the Control Strip on every desk and phone that has exe open, through reloads and restarts, until someone presses its close box on any one of them.</p>
<pre><code>$ curl -s 127.0.0.1:7777/v1/notices \
    -d '{"id":"backup",
         "title":"Backup failed",
         "body":"The disk is full."}'
</code></pre>
<p>On a node with an API token, the call also needs the token in its <code>Authorization</code> header, as every API call does.</p>
<p>A title is needed, and the body is optional. A later notice with the same <code>id</code> changes the one that is up instead of adding another, so a script can keep one notice current. <code>DELETE /v1/notices/&lt;id&gt;</code> closes one, as its close box does. Only callers on the machine itself can put one up, and at most eight stay up at once. See <a href="https://exe.v2core.com/docs/using/the-desktop#notices-a-toast-that-stays">Notices: a toast that stays</a>.</p>
<h2>The smaller things</h2>
<h3>Every app at <code>/apps/</code></h3>
<p>When you want one app in a browser tab of its own, add <code>/apps/</code> to the desktop's address. Until this release, that address was a 404. Now it is a page of every app this node has, as icons. Click one, and it fills the tab. The list stays behind the desktop's sign-in in that browser.</p>
<h3>Cloudflare and Hub are summaries</h3>
<p>The <strong>Cloudflare</strong> tab of <strong>Windows → Configuration</strong> held six fields that have to belong together: the account, the zone, the tunnel and a token that may edit that zone. The form saved any mix without a check. Now the tab says whether the last check passed and names the account, the zone and the tunnel. <strong>Set Up…</strong> or <strong>Change…</strong> opens the Cloudflare Setup Wizard with every answer filled in, and the wizard checks each step with Cloudflare.</p>
<p>The <strong>Hub</strong> tab held seven fields for the hub agent, and none was checked. Now it shows this node on the hub the Hub app joined: its <strong>Hub ID</strong>, whether that hub answers, this node's profile there, and whether the hub lets it post. <strong>Join a Hub…</strong> opens the Hub app's Join a Hub wizard.</p>
<h3>Blue Pencil keeps its checks</h3>
<p>Until this release, a new <strong>Check on</strong>, <strong>Model</strong> or <strong>Thinking</strong> choice in <strong>Blue Pencil Options</strong> cleared what the pencil had checked, so every paragraph of every draft went back to the model. Now what was checked stays checked, and only new or edited paragraphs go to the new choice. You can try another model without waiting for every draft again. A new custom prompt still checks every draft, because its rules are new.</p>
<h3>The activity tile shows its chart's icon</h3>
<p>On Linux, the Control Strip's activity tile charts one of CPU, memory, GPU, network or disk. Its icon now says which: a chip, a memory stick, a graphics card, a network port or a hard disk. The menu's rows wear the same icons, so the menu is the legend.</p>
<h2>Fixed</h2>
<ul>
<li>On Windows, the Configuration window no longer scrolls sideways. In Lucida Sans its tabs were 9 px wider than the window left room for.</li>
<li>A menu of the menu bar or a title bar now closes when you press inside an app's window.</li>
<li>The Control Strip's Tailscale tile lights its lamps as Tailscale's T, not upside down.</li>
</ul>
<h2>Get it</h2>
<p>An exe that is already installed moves to this release with <strong>Software Update…</strong> in the Apple menu, then <strong>Update Now</strong>, or with <code>exe update</code> in a shell. The update runs in the daemon, so you can close the window or the browser while it goes on. exe restarts to finish, which stops the running VMs and starts them again, and the alert names them first.</p>
<p><img src="https://blog.v2core.com/0C6AEF7F-6E36-4C49-90DA-20748C93AA37/software-update.png" alt="The Software Update window after the update: exe 2026.10.28 is the newest release, Installed and Newest both 2026.10.28, and the first two notes under What changed in 2026.10.28. The buttons are Update Now, dimmed, and Check Now." /></p>
<p>A new install on Linux, or on macOS 13 or later, on x86-64 or ARM64:</p>
<pre><code class="language-sh">curl -fsSL https://exe.v2core.com/install.sh | sh
</code></pre>
<p>On Windows 11 on x86-64, in PowerShell:</p>
<pre><code class="language-powershell">irm https://exe.v2core.com/install.ps1 | iex
</code></pre>
<p>The macOS builds are signed with a Developer ID and notarized by Apple. The files and their checksums are on the <a href="https://github.com/livid/exe/releases/tag/2026.10.28">release page</a>.</p>
<h3>What was tried before publishing</h3>
<p>The update that Software Update starts was run from 2026.10.15 on Linux (x86-64), on Windows 11 and on a Mac (Apple silicon), each with a VM running. It took 34 seconds, 10 seconds and 4 seconds. Each guest was shut down, not cut off, and came back.</p>
<p>Because the sign-in changed in this release, <strong>Update Now</strong> was also pressed in a browser, on a Linux (ARM64) install with an API token, from a test build of this release's commit. The window said the update was done about 9 seconds later, with the tab closed during the download. <code>exe update</code> in a shell on Linux (ARM64) was tried too, and so was a new install on Linux (ARM64) and on a Mac (Apple silicon), each with its questions answered at a terminal.</p>
<p>Not tried for this release: a new install on Windows, Windows 10, and, as said above, a passkey on a phone. Passkeys were tested with Chromium's virtual authenticator.</p>
<h3>Good to know</h3>
<ul>
<li>The hub agent is gone. It was off on every node, and a <code>hub</code> section in <code>config.json</code> is dropped at the next save.</li>
<li>The Windows build is not signed, as before.</li>
<li>The Intel Mac build was only started under Rosetta, on an Apple silicon Mac. Its VMs have still not been tried on an Intel Mac.</li>
</ul>

        ]]></description>
    </item>
    
    <item>
        <title>The Control Strip: five live modules in one corner</title>
        <link>https://blog.v2core.com/control-strip/</link>
        <guid>https://blog.v2core.com/79D71D74-55CB-48AF-AA7B-70AE78838C9D/</guid>
        <pubDate>Tue, 27 Oct 2026 02:46:14 -0700</pubDate>
        
        <itunes:image href="https://blog.v2core.com/control-strip/control-strip-activity.png" />
        
        
        <description><![CDATA[
            <p>The Control Strip is the row of tiles at the bottom left of exe's Mac OS 9 desktop in the browser. With no window open, it shows if your Cloudflare tunnel is up, if Tailscale is connected, what SOL is doing, how full your Claude Code or Codex plan is, and how busy the machine is.</p>
<p><img src="https://blog.v2core.com/79D71D74-55CB-48AF-AA7B-70AE78838C9D/control-strip.png" alt="The Control Strip at rest: a scroll arrow, Cloudflare with a green lamp and 412 KB/s, the Tailscale lamp panel, SOL at $187.42, agent usage at 101M, a small activity chart at 29%, a second scroll arrow and the tab." /></p>
<h2>A tray from Mac OS 9</h2>
<p>Mac OS 9 kept a Control Strip in the same corner, and exe's is sampled pixel by pixel from a running Mac OS 9: a strip 24 pixels tall that floats over your windows, as the original did. Click a tile, or right-click it, for its menu, which does more than watch.</p>
<p>The strip fits the machine it runs on. The Tailscale module is there only where Tailscale is installed, agent usage once Claude Code or Codex is, and activity on Linux. The tab at the right end folds the strip down to the tab alone, and the arrows at its ends scroll the row when it holds more than the screen shows. Your browser remembers the fold and what you chose on each tile.</p>
<h2>Cloudflare: what your sites send out</h2>
<p>The lamp on the cloud is green while the tunnel is healthy, yellow when it needs attention or is not set up yet, and grey while exe is still checking. On a computer, the figure beside it is how many bytes a second this machine's connector is sending out: its answers, what visitors download.</p>
<p>The menu opens the tunnel up: the replicas and edge connections, then this machine's own connections, active requests, requests a second, bytes sent and received, total requests and origin errors. While the menu is open, this machine's figures refresh every five seconds and the tunnel-wide counts every 30 seconds. For what Cloudflare counted at its edge, <strong>Analytics…</strong> opens the Analytics app.</p>
<p><img src="https://blog.v2core.com/79D71D74-55CB-48AF-AA7B-70AE78838C9D/control-strip-cloudflare.png" alt="The Cloudflare menu, headed &quot;Cloudflare: working&quot;: replicas and connections, this node's connections and active requests, local requests a second, bytes sent and received, the request total and errors since startup, then Cloudflare Status…, Analytics…, Cloudflare Setup Wizard… and Check Now." /></p>
<h2>Tailscale: switches, not only lamps</h2>
<p>The Tailscale module is a panel of nine lamps, and four of them are Tailscale's. They are lit while this machine is on your tailnet, the private network Tailscale makes of your devices. They turn blue when its traffic leaves through an exit node, another of your devices that carries it to the internet, and yellow when something needs attention, such as a login that is due. All go dim when Tailscale is off.</p>
<p>The menu is where you change things. <strong>Tailscale Active</strong> and <strong>Tailscale Inactive</strong> turn Tailscale on and off, a pair of marked lines like OS 9's AppleTalk Switch module. <strong>Exit Node</strong> picks one of your devices that offers one, with <strong>Allow LAN Access</strong> in the same submenu. <strong>Devices</strong> lists what is online and <strong>Serve</strong> your Tailscale Serve rules; pick a row and its address is copied. <strong>Accept Routes</strong>, <strong>Use Tailscale DNS</strong>, <strong>Shields Up</strong> and <strong>Tailscale SSH</strong> are switches, and each says what it does when you rest the pointer on it.</p>
<p><img src="https://blog.v2core.com/79D71D74-55CB-48AF-AA7B-70AE78838C9D/control-strip-tailscale.png" alt="The Tailscale menu, connected as studio with 4 of 5 devices online, with Tailscale Active marked. The Exit Node submenu is open: None is marked, then two devices that offer an exit node, nas and vps-frankfurt, and Allow LAN Access." /></p>
<p>The daemon makes these changes with the tailscale command, without sudo, and changes nothing else. For that, the account exe runs as must be Tailscale's operator. It also knows how you reached the desktop. Turn Tailscale off or raise the shields while the desktop itself comes to you over Tailscale, and it asks first, because the desktop would go with it. Choosing an exit node while you reach the desktop over the local network, with Allow LAN Access off, asks too.</p>
<h2>The ticker, and the switch for every push</h2>
<p>The ticker wears a coin and its dollar price: SOL, PUMP, MET or SKR, whichever you pick from its menu. The menu lists all four with their day change, and the other three with their price in SOL as well, where a long run of zeros folds into a small count: 0.00003717 SOL reads 0.0₄3717 SOL. The prices are Coinbase's spot prices, fetched once a minute by the daemon for every desktop on the machine, with Jupiter's filling in when Coinbase has none.</p>
<p>The daemon keeps watching when no desktop is open. Choose <strong>Notify Me of Big Moves</strong>, and a move that is rare for its token reaches your phone or computer. SOL has to move 2.5% in an hour or 6% in a day; PUMP 5% or 12%, MET 6% or 15%, SKR 8% or 25%. A token has to cross its threshold on two successful checks in a row before an alert fires.</p>
<p>There are never more than four alerts per token in 24 hours, and a wild day's later alerts need bigger moves: one and a half times the threshold for the second, twice for the third and fourth. Moves held back by that limit are counted into the next alert, if they are less than a day old. A newer alert for the same token replaces the older notification instead of stacking, and <strong>Recent Moves</strong> lists the last eight.</p>
<p><img src="https://blog.v2core.com/79D71D74-55CB-48AF-AA7B-70AE78838C9D/control-strip-prices.png" alt="The ticker's menu: SOL, PUMP, MET and SKR, each with its coin, dollar price and day change, the last three with their price in SOL. The Recent Moves submenu is open, with two moves and the thresholds under them." /></p>
<p>That menu item is the one switch for every push exe sends; nothing else has a setting of its own. With it on, the daemon also watches the sky over the first city in your Weather app, for rain in the next hour, dangerous heat, high wind and fire weather. It says <strong>Claude Code finished</strong> or <strong>Codex finished</strong> when a turn ends in a desktop session more than a minute after you last typed there. And a script on the machine can send its own words:</p>
<pre><code>$ curl -s 127.0.0.1:7777/v1/push \
    -d '{"title":"Backup",
         "body":"Done."}'
</code></pre>
<h2>Agent usage: what the agents spent</h2>
<p>The agent usage meter shows how many tokens Claude Code or Codex has used on this machine today. Choose <strong>Show Plan Limit</strong>, and the tile shows the fullest of that agent's usage windows instead, such as "46% wk", so the limit stays in view all day.</p>
<p>The menu lists both agents with today's tokens, and their five-hour and weekly windows lined up in columns. Each agent's submenu has the detail: today and the last seven days, split into input, cached and output tokens; the week as a column a day, with a figure on the busiest day and on today; and how full each window is and when it resets.</p>
<p><img src="https://blog.v2core.com/79D71D74-55CB-48AF-AA7B-70AE78838C9D/control-strip-agent-usage.png" alt="The agent usage menu with a row each for Claude Code and Codex. The Claude Code submenu is open: today's and the week's tokens split into in, cached and out, a column for each of the last seven days, and when the 5-hour and weekly windows reset." /></p>
<p>The daemon counts from the CLIs' own logs, Claude Code's transcripts and Codex's rollouts, so a session you run in your own terminal counts as well as the ones opened from the desktop. Claude Code's windows are as fresh as its last reply in a Claude Code window on the desk. Codex's come from its ChatGPT sign-in, normally checked at most once a minute, or from Codex's own logs when that check cannot be made.</p>
<h2>Activity: the machine itself</h2>
<p>The last module charts this machine: CPU, memory, GPU, network and disk, a sample every two seconds. The tile shows the last minute of one of them and its figure now. The menu charts them all over the last four minutes, two on a phone, moving while you watch. Pick a row to put it on the tile.</p>
<p><img src="https://blog.v2core.com/79D71D74-55CB-48AF-AA7B-70AE78838C9D/control-strip-activity.png" alt="The activity menu, headed by the machine's name, cores, memory and GPU: CPU, Memory, GPU, Network and Disk, each with a live chart and its figures beside it." /></p>
<p>CPU says its share of all cores and its hottest sensor's temperature. GPU says its use, temperature and power, from nvidia-smi; a host without it has no GPU row. Network and Disk count only the physical interfaces and disks. On a machine full of VMs that matters: a VM's traffic crosses its tap, the bridge and the network card, and counting the card alone counts every byte once.</p>
<p>The daemon samples only while a desktop reads. It starts when the strip opens and stops ten minutes after the last desktop folds the strip away or goes hidden.</p>
<h2>Try it</h2>
<p>On your own exe desktop:</p>
<ol>
<li>Click the ticker and choose <strong>Notify Me of Big Moves</strong>. Allow notifications when the browser asks.</li>
<li>Choose <strong>Send a Test Notification</strong> in the same menu, and watch it arrive.</li>
<li>Click the agent usage meter and choose <strong>Show Plan Limit</strong>.</li>
<li>On Linux, click the activity tile and pick the measure you want in your corner.</li>
</ol>
<p>On a phone, add the desktop to the Home Screen first: that app is where a phone gets notifications. The <a href="https://exe.v2core.com/docs/using/the-desktop#the-control-strip">manual</a> has each module in full, and <a href="https://exe.v2core.com/docs/using/the-desktop#notifications-turn-them-on">Notifications: turn them on</a> has the rest.</p>

        ]]></description>
    </item>
    
    <item>
        <title>Projects: an issue tracker you share with Claude Code</title>
        <link>https://blog.v2core.com/projects-and-claude-code/</link>
        <guid>https://blog.v2core.com/9AF24569-3287-4D81-AAC8-809C826ADBDE/</guid>
        <pubDate>Fri, 16 Oct 2026 16:36:29 -0700</pubDate>
        
        
        <description><![CDATA[
            <p>Work with a coding agent for a week and loose ends pile up. You spot a bug while the agent is busy with something else. It finishes a task and mentions two things it left alone. A good idea comes up at the end of a long conversation, and by morning that conversation is buried under ten newer ones. Projects is a small issue tracker on the exe desktop that catches all of it, and Claude Code can read it and write to it as easily as you can.</p>
<h2>A list where everything has an address</h2>
<p>Projects looks like the rest of exe: one window, with your projects on the left, a project's issues in the middle and the chosen issue on the right. An issue is one thing to do, such as a bug, a feature or a question. It has a title, a description, pictures if you have them, and a conversation underneath.</p>
<p>Every issue gets an address: the project's short name and a number, such as EXE-12 or PROJ-15. An address never changes and is never given to another issue. That sounds like a small thing, but it is what lets you and an agent talk about the same piece of work without copying it back and forth. Once Claude knows where the tracker is, "fix EXE-12" is a complete instruction.</p>
<h2>Hand it over</h2>
<p>When an issue is ready, right-click it and choose <strong>Send to Claude Code</strong>.</p>
<p><img src="https://hub.v2core.com/v1/embed/bafybeia33wdg34xzfu3asnfiyyqhtfxlnmyroi4rdbbeufw4g4fbgxgrue" alt="The Projects window with an issue's menu open. Below Close Issue, Edit, Star and Copy Address are Send to Claude Code and Send to Codex." /></p>
<p>The Claude Code window comes forward with a new conversation, already in the project's folder. Its first message is the whole issue: the description, every comment so far, and a short note that asks Claude to read the project's instructions, do the work, check it and report back. Attached files are named, not sent along. You do not type a word.</p>
<p><img src="https://hub.v2core.com/v1/embed/bafkreifqnufysce43acdohm7s2rqtgouo4s65fkfijpmadxomsqcsv26ae" alt="The Claude Code window. The new conversation starts with &quot;Work on Projects issue PROJ-6 in this project&quot;, followed by the issue itself, and Claude is already reading the code." /></p>
<p><strong>Send to Codex</strong> does the same for Codex. To start from a blank page instead, right-click the project itself and choose <strong>Open in Claude Code</strong>: you get a new conversation in the project's folder, waiting for you.</p>
<p>Nothing about Claude Code changes. If it has never worked in that folder, it asks whether you trust it, and it asks for the permissions you have not already given it.</p>
<h2>See who is on it</h2>
<p>While Claude works in exe's Claude Code window, the issue says so. A line appears under the title with a pulsing green dot, the agent's name and how long it has been working since your last message.</p>
<p><img src="https://hub.v2core.com/v1/embed/bafybeiakg6vq25lmaplq6nbjf2nsrhngeq53iaykpuogzhiuasxjfmtr54" alt="The issue PROJ-6 in Projects. Under its title, a green dot and the line &quot;Claude Code · working for 1 min&quot;." /></p>
<p>The line does not care how the agent got the issue. Send to Claude Code is one way. Typing "have a look at PROJ-6" into a conversation that is already open in the Claude Code window is another. The line goes away when Claude finishes, or stops to wait for you. Come back to an issue days later and you can see at a glance whether someone is already on it.</p>
<h2>Claude writes back</h2>
<p>An agent does more than read the tracker. Projects comes with a command an agent runs in its terminal to file an issue, read one, comment on it and close it. Give the agent a signing key and a name of its own, and everything it writes carries that name, not yours.</p>
<p>Tell it once how you want this used, in the project's instructions file: CLAUDE.md for Claude Code, AGENTS.md for Codex. A few lines are enough. Put in the command's full path on your machine, since it stays in the folder where Projects was installed:</p>
<pre><code>Issues for this project are in
Projects. The command is
~/exe-projects/exe-projects,
and `help` lists what it does.
Read an issue with `show EXE-12`.
Found something outside your
task? File it with
`new . "Title" -m "What you saw"`
and carry on.
Before you start an issue,
comment your plan on it as a
checklist. When you finish,
tick the checklist, report what
you did, and close the issue
with the commit.
</code></pre>
<p>Send to Claude Code asks for a report, but not where to put it. Lines like these put the plan and the report on the issue, where you will look for them.</p>
<p>Here is a finished one: a request to reorder the projects in the sidebar by dragging them. Claude posted a plan, ticked each step as it was done, and then reported what it built, how it tested it and what it left out, with a screenshot. Then it closed the issue.</p>
<p><img src="https://hub.v2core.com/v1/embed/bafybeia5ptj3f7eeggb2ze5x4xhzp5m3kl3admg5qyy52rqji3mzzyvdpq" alt="The finished issue PROJ-15, &quot;Projects sidebar: drag to reorder projects&quot;. the comment &quot;Claude, build this.&quot;, Claude's plan with every box ticked, Claude's report with a screenshot, and &quot;Claude closed this&quot;." /></p>
<p>Everything you need to judge the work is on one page, and it stays there after the conversation has scrolled away.</p>
<h2>The loop</h2>
<p>With those lines in place, the work goes round like this:</p>
<ol>
<li>You write down what you want as an issue. If it is about something you can see, attach a screenshot.</li>
<li>You send it to Claude Code, or mention its address in a conversation.</li>
<li>Claude reads the issue and the project's instructions, posts a plan and starts. The issue shows that it is at work.</li>
<li>Claude reports on the issue and closes it. Anything it noticed on the way becomes an issue of its own.</li>
<li>You read the report. If something is off, reopen the issue, say what, and send it again. The new conversation starts with everything, the last report included.</li>
</ol>
<p>On the machine where exe is made, the loop takes even fewer clicks. A small script watches the tracker. When a comment asks for work, like the "Claude, build this." in the picture above, the script starts Claude Code on that issue by itself. That script is not part of exe. Send to Claude Code is the same step, done by hand.</p>
<h2>Try it</h2>
<p>Projects is one of exe's extras: a program of its own that you add when you want it. Send to Claude Code needs exe 2026.10.15 or newer, with Claude Code and tmux on the same machine as exe.</p>
<ol>
<li>Choose <strong>Help → Welcome…</strong>, and press <strong>Extras…</strong> on the Apps card.</li>
<li>Go to the Projects page. If Projects is not on this machine yet, the page gives you a line to paste into Claude Code, and Claude installs it. If Projects is installed but the page cannot find it, type its folder and press <strong>Find</strong>.</li>
<li>Make sure the box is ticked, press <strong>Continue</strong> until you reach Review, and press <strong>Apply</strong>. Projects appears on the desktop.</li>
<li>Make a project. Under <strong>Directory</strong>, type the full path of its folder on the machine that runs exe. That is where Send to Claude Code opens.</li>
<li>Write an issue, right-click it and choose <strong>Send to Claude Code</strong>.</li>
</ol>
<p>One caution: files you attach to an issue are kept in IPFS, where anyone who learns a file's address can fetch it. Keep secrets out of them.</p>
<p>The <a href="https://exe.v2core.com/docs/using/apps#projects-and-coding-agents">manual</a> has the details, and the source is at <a href="https://github.com/livid/exe-projects">github.com/livid/exe-projects</a>.</p>

        ]]></description>
    </item>
    
    <item>
        <title>exe 2026.10.15: hand an issue to Claude Code or Codex</title>
        <link>https://blog.v2core.com/exe-2026-10-15/</link>
        <guid>https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/</guid>
        <pubDate>Thu, 15 Oct 2026 13:45:00 -0700</pubDate>
        
        <itunes:image href="https://blog.v2core.com/exe-2026-10-15/an-issue-sent-to-claude-code.png" />
        
        
        <description><![CDATA[
            <p>exe 2026.10.15 hands work to its coding agents. Projects, a new extra, is an issue tracker on the desk that sends an issue to Claude Code or Codex from its menu. Welcome now sets up both CLIs, the Claude Code window shows every prompt, and a phone can paste into a terminal.</p>
<h2>Projects: an issue tracker on the desk</h2>
<p>Work for an agent often starts as an issue: a bug you saw, a screenshot, a plan in a comment. Projects keeps those issues on the desk, beside the agents that work on them. It comes with exe-projects, a program that runs beside exe, and it is the fourth extra, after the extra apps, Planet and Easel.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/an-issue-sent-to-claude-code.png" alt="Part of the Projects window: an issue in the list with its right-click menu open. Under Close Issue, Edit…, Star and Copy Address are Send to Claude Code and Send to Codex, then New Issue…." /></p>
<p>The window has three columns: your projects, a project's issues and the chosen issue. A project has a short name in capitals, such as <code>EXE</code>, and its issues are numbered from one. So <code>EXE-12</code> always names the same issue, wherever you write it. Right-click an issue to close, edit or <strong>Star</strong> it, or to copy its address. The pop-up at the top right lists the open, the closed, all or the starred issues.</p>
<p>Agents use the same tracker from a shell. <code>exe-projects new</code> files an issue, and in a project's folder a dot stands for the project, so an agent at work in a checkout files what it finds without knowing the project's name.</p>
<p>Click a picture under an issue to open it in a picture window. When the issue has more than one, the Left and Right arrow keys step through them, and the window says which one it shows, such as <code>2 of 5</code>.</p>
<p>Attached files are kept in IPFS, which needs Kubo on this machine. Anyone who learns a file's address can fetch it, so attach nothing private.</p>
<h3>Send an issue to Claude Code or Codex</h3>
<p>To hand an issue to an agent, you would copy its text and its comments into a prompt and start a session in the right folder. Now right-click the issue and choose <strong>Send to Claude Code</strong> or <strong>Send to Codex</strong>. A new session starts in the project's folder, with the full issue and its comments as its first task. You watch it work in the agent's window and take over when you want.</p>
<ul>
<li>Only the agents that exe found on this machine are in the menu.</li>
<li>Attached files are named in the task, not sent with it.</li>
<li>It needs tmux, and a working directory for the project on this machine.</li>
<li>The agent keeps its usual sign-in, trust and permission prompts.</li>
<li>An issue over 64 KiB is refused, and the message says its size.</li>
<li>Issues sent while the window is still opening each get a session of their own, in order.</li>
</ul>
<p>The issue then shows the session at work on it, and how long its turn has run.</p>
<h3>Open a project in Claude Code</h3>
<p>When the work is not one issue yet, right-click a project and choose <strong>Open in Claude Code</strong>. A new session starts in the project's working directory, at the top of the Claude Code window's list. The working directory is optional. Press <strong>Project…</strong> and write it in the <strong>Directory</strong> field. The item is greyed for a project without one, or for one whose folder is not on this machine, and it needs tmux, as the two Send items do.</p>
<h3>Install it with Extras</h3>
<p>Projects is a program of its own, built from its source, so exe does not carry it. Choose <strong>Help → Welcome…</strong>, press <strong>Extras…</strong> on the Apps card and go to the <strong>Projects</strong> step. While Projects is not on this machine, the page gives a line to hand a coding agent. <strong>Copy Prompt</strong> copies it, and the agent installs exe-projects from its README. If Projects is here and exe did not find it, type its folder and press <strong>Find</strong>.</p>
<p>Tick the box and press <strong>Apply</strong> on <strong>Review</strong>. Extras adds the program's apps folder and its daemon to exe's configuration for you. Turning it off changes only exe's configuration, and the daemon runs on until you stop it. The <a href="https://exe.v2core.com/docs/using/apps">Apps chapter</a> has the rest.</p>
<h2>Set Up… on Welcome's Claude Code and Codex card</h2>
<p>The <strong>Claude Code</strong> and <strong>Codex</strong> icons appear on the desktop only once their CLIs are on the machine exe runs on. Welcome already said which of the two are there. Now it also gives the version of each, and helps you install the one that is missing.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/welcome-set-up.png" alt="Welcome on its Claude Code and Codex card: a green line says that both CLIs are installed, with their versions, over the buttons Back, Read More…, Set Up…, Open Codex and Open Claude Code." /></p>
<p>On the <strong>Claude Code and Codex</strong> card of <strong>Help → Welcome…</strong>, <strong>Set Up…</strong> opens <strong>Agent Setup</strong>, a window of three steps: a page for Claude Code, one for Codex, and a review.</p>
<ul>
<li>An installed CLI's page shows its version and what exe adds to it, with <strong>Open Claude Code</strong> or <strong>Open Codex</strong>.</li>
<li>A missing CLI's page shows the command that installs it on this kind of machine: Anthropic's installer for Claude Code, and npm or Homebrew for Codex. <strong>Copy Command</strong> copies it, and <strong>Install in Terminal…</strong> runs it in a Terminal window on the desktop.</li>
<li><strong>Check Again</strong> looks at the machine again. A CLI installed while the desktop is open gets its icon without a reload.</li>
<li><strong>Review</strong> lists both versions and says whether tmux is on the machine. Without tmux, each window is a fresh run of the CLI.</li>
</ul>
<h2>The Claude Code window</h2>
<h3>The prompt line, and every prompt</h3>
<p>A terminal shows what the agent answered, and your own prompt scrolls away. Now a line over the terminal shows the first line of the session's latest prompt, with the number of prompts at its right. It follows the session on screen, and a new prompt shows there within a couple of seconds.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/claude-prompts.png" alt="Part of the Prompts dialog: a made-up session's prompts in order, each with its number and time, the last one chosen and shown whole below, 14 prompts, and the Copy and OK buttons." /></p>
<p>Click the line for <strong>Prompts</strong>, a list of every prompt of the session in order, each with its number and time. The selected prompt stands whole under the list. <strong>Copy</strong> copies all of it. A prompt is what you typed or pasted: slash commands, tool results and a sub-agent's turns are not listed. A prompt over 20,000 characters is cut there in the list.</p>
<h3>Get Info on a session</h3>
<p>You archive an idle session to get memory back, and first you want to know what the session was. Right-click its row and choose <strong>Get Info</strong>.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/get-info.png" alt="The top of the Info window of a made-up Claude Code session: its title, then Summary, First prompt, Last prompt, Turns and Tokens, with how much of the input was read from the cache." /></p>
<p>The dialog is read from the transcript that Claude Code keeps on this machine. It gives the summary, the first and the last prompt, the count of prompts and replies, and the tokens, with how much of the input came from the cache. It also gives the models, the times, the folder and the session's name. <strong>Archive…</strong> is in the dialog too, so you read and decide in one place.</p>
<h3>What Archive… gives back</h3>
<p>Now, once an archived session's processes are gone, a note at the foot of the screen says what they held in memory and in swap, and how much memory the machine has available now. You know what ending the session was worth.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/archive-note.png" alt="The note at the foot of the screen after Archive…: 254.0 MB freed and 51.2 MB of swap, and 96.0 GB available now." /></p>
<p>The figures are the processes' own, read before tmux ends them. On a Mac or a PC, where the daemon cannot read what a process holds, the note says only how much is available.</p>
<h3>Search finds a session by its name</h3>
<p>A session has a name of its own, such as <code>exe-claude-90</code>, which <strong>Get Info</strong> gives on its <strong>Session</strong> line. <strong>Search</strong> matched a session in the column on its title alone, so typing that name found nothing. Now it matches the name too, the whole name or its number, so <code>exe-claude-90</code> or <code>90</code> finds that session.</p>
<p>The Codex window has no prompt line, <strong>Get Info</strong> or <strong>Search</strong> yet. The <a href="https://exe.v2core.com/docs/using/claude-code-codex-and-terminal">Claude Code chapter</a> has the whole window.</p>
<h2>Agent on the Mac OS X window</h2>
<p>Tiger, in the Mac OS X app, is a 2005 system, and its tools lack much that a current one has. Now an agent that knows this works inside the Mac for you, while you watch.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/mac-os-x-agent.png" alt="The top of the Mac OS X window: the buttons Setup details, Agent, Mac keys…, Full screen, Sound off and CD…, over the Mac's own menu bar." /></p>
<p>While the Mac runs, press <strong>Agent</strong> in the Mac OS X window, and a Chat opens pinned to the Mac, as <strong>Run Agent</strong> pins one to a VM. Chat needs a model for this, from Ollama or a ChatGPT sign-in. Ask in words, such as "what is in /Applications?" or "put the picture I painted on the desktop". It works over Remote Login with the shell and file tools, keeps a memory of the Mac between chats, and never restarts or shuts down the Mac unless you ask. Its upload tool copies a Workspace file into the Mac. When it was built, a chat on Tiger 10.4.11 uploaded a picture from Paint, made it the desktop picture and photographed the screen.</p>
<p>Set it up once in <strong>Setup details</strong>. Enter the Mac account that the agent works as. Then run the line shown there on the Mac, as that account. The window does not carry your clipboard into the Mac. One way to get the line there is a shell over Remote Login from this machine, which the manual describes. The line authorizes an RSA key that exe made, since Tiger's SSH server knows no newer kind. <strong>Remote Login</strong> must be on in the Mac's Sharing pane. Until all this is done, a chat with the Mac fails and says why. The <a href="https://exe.v2core.com/docs/using/mac-os-x">Mac OS X chapter</a> has more.</p>
<h2>Stickies from Notes</h2>
<p>A note you wanted in view had to stay open in the Notes window. Now it can stand on the desk by itself, as notes did in Mac OS 9's Stickies.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/sticky.png" alt="A sticky on the desk: a small yellow window with a short list typed on it, a close box at the top left and a grow box at the bottom right." /></p>
<p>Press the pushpin at a note's foot in Notes. The note becomes a small window in its own colour, with a look sampled from Stickies in a running Mac OS 9. Drag it by its thin title bar and size it by the box on its corner. The box at the bar's right end collapses it, and the box at the left end puts the note back in Notes alone. What you type in a sticky is in Notes too.</p>
<p>The place is kept in the note, so a sticky stands where you put it on every desk that syncs. A drag on one desk and an edit on another both survive. A phone opens no stickies.</p>
<h2>Show Desktop</h2>
<p>Windows cover the desk's icons. Closing them, or dragging them aside, changes the saved layout on every screen that shows the desk.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/show-desktop.png" alt="The right end of the menu bar with Show Desktop pressed, left of Show All Windows, the magnifier and the clock. The desktop's icons are in view, and the edges of windows that slid off show at the top and the right." /></p>
<p>The button with a bare desk on it, left of <strong>Show All Windows</strong> in the menu bar, slides every window off to the nearest edge of the screen, with a sliver of each in view. The desk, its icons and the Control Strip stay in use. Click a sliver to bring all the windows back with that one in front. The button again, Escape, or opening a window from the desk brings them back too. Nothing is closed, and the saved layout does not change. The desktop menu reaches it as <code>showdesk</code>, and a phone has no such button. <a href="https://exe.v2core.com/docs/using/the-desktop">The desktop chapter</a> has more.</p>
<h2>Notes and Projects while exe does not answer</h2>
<p>When the daemon restarts or the device is offline, the desktop shows an alert until the daemon answers again. Until now, nothing could be read in that time. Reading a note or an issue should not need the network. Only changes should.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/notes-offline.png" alt="The foot of a Notes window while exe does not answer: under the note, the line reads Offline: the copy of just now." /></p>
<p>Notes and Projects now keep a copy on each device that has had the desktop open over HTTPS. The alert has a button for each: <strong>Open Notes</strong> and <strong>Open Projects</strong>. In a browser the app opens in a new tab, and the alert turns back into the desk the moment exe answers. In the installed app, which has no tabs, the app takes the alert's place. On a phone it has a back button at its top left, marked exe, that leads back.</p>
<p>The app says how old its copy is and dims what would change something, such as <strong>New Note</strong> and <strong>Delete</strong>. Projects keeps every issue, open or closed, and the pictures on open issues, up to 4 MB each. The copy is renewed each time the desktop opens and each time a notification arrives. A load that has not begun to answer after 8 seconds also shows the alert.</p>
<h2>On a phone</h2>
<p>A pinch on a picture zoomed the whole page, with the menu bar. Now it zooms the picture about your fingers, up to its own pixels and at least four times over. A double tap zooms to the picture's own size, and another double tap brings it back. Zoomed in, one finger drags the picture. A picture from an issue in Projects swipes sideways to the issue's next or previous picture, fetched ahead so that it shows at once.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/phone-copy-paste.png" alt="The Claude Code window on a phone after a press and hold on the terminal: a menu with Copy dimmed, Paste and Select All." /></p>
<p>You could not paste into Claude Code's composer on a phone either. The terminal's menu opened only on a right-click, which iOS never sends, and the phone's keyboard has no paste key. Now press and hold the terminal of a Claude Code, Codex or Terminal window for 550 ms, and the menu opens: <strong>Copy</strong>, <strong>Paste</strong> and <strong>Select All</strong>. <strong>Paste</strong> reads the clipboard, and an iPhone shows its Paste bubble to tap. Where the browser refuses, a small <strong>Paste</strong> dialog takes the text into a field of its own and sends it on.</p>
<h2>On an iPhone or iPad</h2>
<p>These three changes were tried in Chromium and WebKit, not yet on a device.</p>
<p>An iPhone can bring a Home Screen app back on the page it had, so after an update the app could show the old desktop. Now a return to the app checks for a new desktop and refreshes the app list. You do not need to force-close the app.</p>
<p>On iPhone and iPad Home Screen apps with iOS 18.4 or later, and in Safari 18.5 or later on a Mac, a notification also carries a form that the browser itself can read. An agent-finished alert can then still appear, and open its session, when the desktop's background worker cannot run.</p>
<p>In the installed app, the screen stays awake while Claude Code or Codex is the active window, so a long turn you watch does not go dark. Switch to another window, or leave the app, and the screen can sleep again. On an iPhone or iPad this needs iOS 18.4 or later, and the device can still refuse it, to save power.</p>
<h2>A folder at a public address</h2>
<p>A static site needed a server process of its own, one more program to keep running. Now the daemon, which is already kept running, serves the folder itself:</p>
<pre><code class="language-sh">exe expose site.example.com \
  -dir /www/site
</code></pre>
<ul>
<li><code>/</code> is the folder's <code>index.html</code>. A folder without one is not listed, and a name that begins with a dot is never served.</li>
<li>Byte ranges work, so downloads resume. Disk images and archives are sent as downloads.</li>
<li>A folder with a <code>.password</code> file asks for HTTP Basic authentication, with that file's first line as the password. Those answers are marked private, so the edge in front of the site never hands them on.</li>
<li>A new file is live at once. Pages are served <code>no-cache</code>, and other files may be kept for ten minutes.</li>
<li>The route persists like any other, and <code>exe unexpose &lt;host&gt;</code> removes it.</li>
</ul>
<p><a href="https://exe.v2core.com/docs/using/publishing-to-the-web">Publishing to the web</a> has the rest.</p>
<h2>Every release's notes on one page</h2>
<p>Software Update shows what the newest release changed. When you skipped a release, its notes were only on GitHub. Now <a href="https://exe.v2core.com/docs/changelog">What Changed</a> keeps every release's notes, back to the first, with an address for each release, such as <code>/docs/changelog#2026-10-12</code>. The first release, 2026.10.10, carried no notes. Its entry lists 38 items, written from the manual as it was at that release.</p>
<h2>Fixed</h2>
<p>On Windows, <code>exe ssh</code> reaches a VM again on a PC where Codex has run. There it failed with "UNPROTECTED PRIVATE KEY FILE" in both Windows releases so far. Codex's sandbox could read the VM's key, and OpenSSH refuses such a key. Now the key has a permission list of its own: its user, SYSTEM and the Administrators. The next daemon start, or <code>exe ssh</code> itself, corrects an older key.</p>
<p>The green dot of a Claude Code session goes out when you stop a turn with Escape. Escape runs no Stop hook, so the dot stayed until the next prompt. Now exe asks the transcript whether the turn is over, and the row shows no mark.</p>
<p>Between joined desks, a Workspace file over 10 MB that you move to another folder stays on its node. Node Sync carries 10 MB at most, yet the nodes tried to send such a file every 30 seconds, and every try failed. On one node, 21 disk images and a video moved into folders cost about 2.2 TB of traffic in a day and a half.</p>
<h2>Get it</h2>
<p>An installed exe moves to this release with <strong>Software Update…</strong> in the Apple menu, or with <code>exe update</code> in a shell. exe restarts to finish, which stops the running VMs and starts them again. The alert names them first, and afterwards the window says how each one came back.</p>
<p><img src="https://blog.v2core.com/B2F0C475-85F1-472F-84FC-95864320A8D0/software-update.png" alt="The Software Update window after the update: exe 2026.10.15 is installed and is the newest release, over the first notes of what changed. Update Now is dimmed, and Check Now stands at the foot." /></p>
<p>A new install on Linux, or on macOS 13 or later, on x86-64 or ARM64:</p>
<pre><code class="language-sh">curl -fsSL https://exe.v2core.com/install.sh | sh
</code></pre>
<p>On Windows 11 on x86-64, in PowerShell:</p>
<pre><code class="language-powershell">irm https://exe.v2core.com/install.ps1 | iex
</code></pre>
<p>The <a href="https://github.com/livid/exe/releases/tag/2026.10.15">release page</a> has the files and their checksums. The Mac builds are signed with a Developer ID and notarized by Apple. The Windows build is not signed, as before.</p>
<h3>What was tried before publishing</h3>
<p>The release is built from commit c4f684f, 49 commits after 2026.10.12. The update that Software Update starts ran from 2026.10.12 on Linux (x86-64), on a Windows 11 PC and on a Mac with Apple silicon, each with a VM running. It took 34 seconds, 9 seconds and 4 seconds. Each guest was shut down, not cut off, and came back. <code>exe update</code> ran in a shell on Linux (ARM64). New installs were made on Linux (ARM64) and on a Mac (Apple silicon), each with its questions answered at a terminal.</p>
<p>Not tried for this release: pressing <strong>Update Now</strong> in a browser, a new install on Windows, and Windows 10. The Software Update window has not changed since 2026.10.12, where <strong>Update Now</strong> was tried. The Intel Mac build was only started under Rosetta, and its VMs have still not been tried on an Intel Mac.</p>
<h3>Good to know</h3>
<ul>
<li>Projects, Planet and Easel install from their own repositories: <a href="https://github.com/livid/exe-projects">exe-projects</a>, <a href="https://github.com/livid/exe-planet">exe-planet</a> and <a href="https://github.com/livid/exe-easel">exe-easel</a>. Projects shows <strong>Send to Claude Code</strong> and <strong>Send to Codex</strong> only with this release of exe or a later one.</li>
<li>After the update, open the desktop once on each device that gets notifications, so it can refresh the subscription's address.</li>
<li>The update brings the extra apps' new versions, Notes with its Stickies among them, except an app you edited or removed.</li>
</ul>

        ]]></description>
    </item>
    
    <item>
        <title>Claude Code belongs in a private PWA</title>
        <link>https://blog.v2core.com/claude-code-private-pwa/</link>
        <guid>https://blog.v2core.com/FA67C622-F0CF-4823-B101-0865E9A1864B/</guid>
        <pubDate>Tue, 13 Oct 2026 16:20:04 -0700</pubDate>
        
        
        <description><![CDATA[
            <p>A long Claude Code task should not depend on your laptop staying open or your Wi-Fi staying up. Run Claude Code on the laptop, and closing the lid stops it. Run it on a server over SSH, and a dropped connection stops it. The form that fits is a private PWA (progressive web app): a web page that installs like an app, and that only you can open. The agent works on a machine of your own that stays on, and every device you carry is a window onto it.</p>
<h2>On the laptop: you cannot close the lid</h2>
<p>This is where everyone starts: <code>claude</code> in a terminal, in the project's folder.</p>
<p>Then a task needs twenty minutes and you need to leave. Close the lid and the laptop sleeps, and the agent sleeps with it. Open it an hour later and the task is where you left it, with the twenty minutes still to go.</p>
<p>A laptop can be told not to sleep. Then it has to stay where it is, on power and on its Wi-Fi, until the agent is done, and you leave without it.</p>
<h2>Over SSH: you cannot lose the Wi-Fi</h2>
<p>So the agent moves to a machine that never sleeps: a desktop in the corner, a Mac mini, a rented server. You SSH in and run <code>claude</code> there.</p>
<p>Now the lid cannot stop the machine. It can still stop the connection, and the agent lives only as long as the connection does. Lose the Wi-Fi for more than a moment, let the laptop sleep or move to another network, and the SSH session breaks. Claude Code is cut off with it, in the middle of its work. <code>claude --resume</code> brings the conversation back, but not the work it would have done in the meantime.</p>
<p>tmux mends that. It keeps a terminal alive on the server with nobody attached, so the agent works on and you attach again later. It also hands you a list of chores:</p>
<ul>
<li>Start tmux first, every time. The task that turns out long is the one you started without it.</li>
<li>Remember which session holds which conversation, and attach to the right one after every drop.</li>
<li>Put a terminal app and an SSH key on every device you want to look from. On a phone, every tmux command starts with Ctrl-B, typed on a touch keyboard.</li>
<li>Look in now and then to see whether it has finished, because nothing tells you.</li>
</ul>
<p>None of this is hard. It is logistics, and it comes back every day.</p>
<h2>As a PWA: neither matters</h2>
<p>Put the two lessons together. The agent belongs on the machine that stays on. What you hold in your hand should be only a view of it, one that can go away at any moment and come back with nothing lost.</p>
<p>A web page is that kind of view. Every device already has a browser, so there is no terminal app to set up and no SSH key to copy. A page that loses its connection can find it again by itself. And a page installed as a PWA gets an icon, a window of its own and notifications, so it can tell you when the agent is done.</p>
<p>exe is built this way. It is a desktop in your browser, served by a machine you own, and Claude Code is one of the windows on it.</p>
<p><img src="https://hub.v2core.com/v1/embed/bafybeiedurlk54ghfmghdxnscdn6xejkss37uekbun2xrrrxbukqwamwkq" alt="The Claude Code window in exe: its sessions down the left, the chosen one's conversation in the terminal, and usage on the status line." /></p>
<ul>
<li><strong>Each conversation is a session on the machine.</strong> Close the tab, close the lid, lose the Wi-Fi: the agent works on.</li>
<li><strong>A column lists the sessions.</strong> A pulsing green dot marks one that is still working, a black dot one that waits for you. A window whose connection dropped reconnects by itself.</li>
<li><strong>Every device opens the same conversation.</strong> Start a task on the laptop and look in on it from the phone.</li>
<li><strong>The phone tells you when it is done.</strong> With notifications on, it says Claude Code finished, and a tap opens that conversation.</li>
</ul>
<p>Under the window it is still tmux: each session is a tmux session that exe starts, names and attaches for you. What tmux does has not changed. The logistics have moved from you to the app.</p>
<h2>Only you can open it</h2>
<p>A window onto Claude Code is a shell on your machine, with your code and your sign-ins behind it. It cannot be a page on the open internet.</p>
<p>exe leans on Tailscale for that. Tailscale is a private network between your own devices, and it reaches the machine from wherever you are. exe listens on the machine's address in that network, so the desktop opens from your laptop, your tablet and your phone, and from nowhere else. On top of that sits an API token, a password for the desktop.</p>
<h2>On a machine you own</h2>
<p>That leaves the machine. Ideally it is a small one at home: an NVIDIA DGX Spark or a Mac mini. Either is quiet, takes little room on a shelf and can stay on day and night. Tailscale reaches it there with no port opened on your router.</p>
<p>That machine has more to do than hold a terminal. The agent's builds and tests run there. So do the virtual machines exe gives the agent as sandboxes, and any model you run yourself. That takes cores, memory and a GPU, and a DGX Spark has 20 cores and a GPU that shares 128 GB of memory with them.</p>
<p>You could rent that in the cloud. But a GPU with that much memory is billed by the hour, and this machine cannot be switched off between tasks, because the agent lives on it. Kept on all year, the rent comes to more than the box costs. A box you own is paid for once.</p>
<h2>Try it</h2>
<p>You need a Linux machine or a Mac that stays on, with tmux on it and Claude Code installed and signed in. Tailscale goes on that machine and on every device you will look from.</p>
<pre><code class="language-sh">curl -fsSL https://exe.v2core.com/install.sh | sh
</code></pre>
<ol>
<li>When the installer asks where exe should listen, choose the Tailscale address. It suggests an API token and prints it with the desktop's address.</li>
<li>Put HTTPS in front of the desktop with <a href="https://tailscale.com/kb/1312/serve">Tailscale Serve</a>, which keeps it inside your private network. Installing as an app and notifications both need HTTPS.</li>
<li>Open the desktop on your phone, paste the token when it asks, and <a href="https://exe.v2core.com/docs/using/the-desktop#install-the-desktop-as-an-app">install it</a>: Add to Home Screen on an iPhone.</li>
<li><a href="https://exe.v2core.com/docs/using/the-desktop#notifications-turn-them-on">Turn notifications on</a>.</li>
</ol>
<p>The rest is in <a href="https://exe.v2core.com/docs/getting-started">Getting Started</a>, and the source is at <a href="https://github.com/livid/exe">github.com/livid/exe</a>.</p>
<p>Then open Claude Code, give it something long, and close the lid.</p>

        ]]></description>
    </item>
    
    <item>
        <title>exe 2026.10.12: Software Update, and the last release you need a shell to get</title>
        <link>https://blog.v2core.com/exe-2026-10-12/</link>
        <guid>https://blog.v2core.com/ED48CB9F-01C7-4F49-A05B-1E3EBEE7F512/</guid>
        <pubDate>Mon, 12 Oct 2026 08:00:00 -0700</pubDate>
        
        <itunes:image href="https://blog.v2core.com/exe-2026-10-12/apple-menu-software-update.png" />
        
        
        <description><![CDATA[
            <p>exe now updates itself from the desktop. Software Update, in the Apple menu, names the newest release, shows what changed in it and installs it. exe 2026.10.12 is the first release that has it, and the last one you need a shell to get.</p>
<p>This is the third release of exe, after 2026.10.10 and 2026.10.10.2. It also brings a second Mac that runs Mac OS X 10.4 Tiger, a Welcome window, four steps onto a hub and a manual you can search. Four fixes come with it.</p>
<h2>Software Update</h2>
<p>Until this release, a new exe meant a shell on the machine and <code>exe update</code> at a prompt. The desktop never looked for a newer release, so <strong>About This Computer</strong> named a version without saying that it was old. Yet an exe is used from a phone as much as from a shell.</p>
<p>Now <strong>Software Update…</strong> in the Apple menu does it from the desktop, on a phone as well as at a desk. A diamond beside the small Mac in the menu bar, and beside <strong>Software Update…</strong> in its menu, says that a newer release is waiting. You do not have to remember to look.</p>
<p><img src="https://blog.v2core.com/ED48CB9F-01C7-4F49-A05B-1E3EBEE7F512/apple-menu-software-update.png" alt="The Software Update window on a machine that runs the release: its first lines say what Software Update does and that exe restarts to finish. Update Now is dimmed, Installed and Newest both say 2026.10.12, the box What changed in 2026.10.12 shows its first notes, and Check Now is at the foot." /></p>
<p>The window says which release is <strong>Installed</strong> and which is the <strong>Newest</strong>, and it lists what changed in the newest one. So you see what is new before you take it.</p>
<ol>
<li>Press <strong>Update Now</strong>. exe restarts to finish, and a restart stops the running VMs and starts them again. So an alert asks first, and it names those VMs.</li>
<li>Press <strong>Update</strong>. exe downloads the release, checks it, replaces the binary and restarts.</li>
</ol>
<p>The update runs in the daemon, not in the page. Close the window, or the browser, and the update goes on. A second <strong>Update Now</strong> from another device joins the same update, and the window says where it stands when you open it again.</p>
<p>When exe is back, the window names the version that runs now. For a quarter of an hour it also says how each VM that was running is doing, so you do not have to open each VM to see that it came back. On Linux, when a release changes the root-owned network helper, the window shows the two <code>sudo</code> lines that install the new helper.</p>
<p>The desktop looks for a release when it loads, and exe asks where the newest release is at most twice a day. <strong>Check Now</strong> asks at once. A look sends nothing about this machine. A release that is installed but not yet running, after a restart you put off in a shell, gets the button <strong>Restart Now</strong>. The <a href="https://exe.v2core.com/docs/using/updating">Updating</a> chapter has the rest, with <code>exe update -check</code> and <code>exe update -y</code> for a script or an agent.</p>
<h3>One last time in a shell</h3>
<p>2026.10.10 and 2026.10.10.2 have no Software Update window. So this one move still takes <code>exe update</code> in a shell on the machine.</p>
<p>On a Mac that runs one of those two releases, stop the VMs first. "Get it", below, says how and why.</p>
<h2>Welcome… and Extras…</h2>
<p>A new install used to open on an empty Virtual Machines window. The only word on what exe is for was a long manual under Help. Now a new install opens on <strong>Welcome…</strong>, which is also in the Help menu: exe in six cards.</p>
<p><img src="https://blog.v2core.com/ED48CB9F-01C7-4F49-A05B-1E3EBEE7F512/welcome.png" alt="The Welcome window: a night sky over a row of old Macs, a line that calls exe a personal VM cloud on this machine, and six cards for Virtual Machines, Chat, Claude Code and Codex, The Hub, Cloudflare and Apps, two of them with a green tick. Below are the exe Documentation… button, the selected box Show this window when the desktop opens, and OK." /></p>
<p>The cards are <strong>Virtual Machines</strong>, <strong>Chat</strong>, <strong>Claude Code and Codex</strong>, <strong>The Hub</strong>, <strong>Cloudflare</strong> and <strong>Apps</strong>. On a fresh install some of the six need a step first. So a click on a card shows how its feature stands on this machine, ready or what it still needs, and the button that goes there: <strong>New VM…</strong>, the pane of Configuration that sets Chat up, the Cloudflare Setup Wizard, <strong>Join a Hub…</strong> or <strong>Extras…</strong>. <strong>Read More…</strong> opens the manual on that feature's chapter.</p>
<p>The window opens with the desktop until you clear <strong>Show this window when the desktop opens</strong>. That box belongs to the machine, not to the browser. An exe that was installed before this release opens the window from <strong>Help → Welcome…</strong> only.</p>
<h3>Extras…</h3>
<p>Until now, the extra apps were a yes or a no at install time and nowhere after. Planet and Easel were two keys to write into <code>config.json</code> by hand. <strong>Extras…</strong>, on Welcome's Apps card, turns all three on and off from the desktop:</p>
<ul>
<li>The extra apps: Notes, Paint, Tides, Todo, Weather and World Clock.</li>
<li>Planet, the site builder that runs beside exe.</li>
<li>Easel, which runs Claude painters at an oil paint simulator.</li>
</ul>
<p><img src="https://blog.v2core.com/ED48CB9F-01C7-4F49-A05B-1E3EBEE7F512/extras.png" alt="The Extras window on its last step, Review: the steps Extra Apps, Planet and Easel ticked down the left, and under What will change, Extra Apps and Planet will be turned on, each with the folder it comes from, and Easel is not on this machine. Under a line that says Apply makes the changes now and touches nothing else are Cancel, Go Back and Apply." /></p>
<p>Extras has a step for each extra, with one check box, and a last step, <strong>Review</strong>. Nothing changes until you press <strong>Apply</strong> there.</p>
<ul>
<li>On a released exe, the extra apps come from the release that runs, and an update keeps them current. Off removes the ones that exe placed. An app you edited stays yours.</li>
<li>Planet and Easel are programs of their own, and exe does not build them. On names a program's apps folder and its daemon in exe's configuration. Off takes them out again and leaves the daemon running.</li>
<li>While Planet or Easel is not on the machine, its check box is grey. Its page gives a line to hand a coding agent, which then installs it.</li>
</ul>
<h2>Join a Hub…</h2>
<p>The Hub app is a public feed shared between exe nodes. Its first run used to be one field for an address. Whoever got past it learned about the key, the gate and the name one refusal at a time. <strong>Join a Hub</strong> takes you from a hub's address to your profile in four steps. The Hub app opens on it while it has no hub. It is also at <strong>Special → Join a Hub…</strong> and on Welcome's Hub card.</p>
<p><img src="https://blog.v2core.com/ED48CB9F-01C7-4F49-A05B-1E3EBEE7F512/join-a-hub.png" alt="Join a Hub on its third step, Your Key: the four steps Hub Address, The Hub, Your Key and Your Profile down the left, a line that says this machine's key is your account, the key's Id and Solana address, and what this hub asks before the key may post, 10,000 of a token or an invite from one of the hub's admins. Check Again is under it, and Cancel, Go Back and Continue are at the foot." /></p>
<ol>
<li><strong>Hub Address.</strong> The hub's name alone will do. <strong>Use hub.v2core.com</strong> fills in exe's own hub, where releases are announced.</li>
<li><strong>The Hub.</strong> Its id, how many users and posts it has, and whether posting is open or token-gated.</li>
<li><strong>Your Key.</strong> This machine's key is your account, so there is nothing to sign up for. A green check says that the key may post. On a token-gated hub, the page says what the hub asks for instead: an amount of its token, or an invitation from one of its admins. <strong>Check Again</strong> asks once more. Reading needs neither, so you can go on either way.</li>
<li><strong>Your Profile.</strong> The name, picture and bio that your posts carry. <strong>Join</strong> publishes them and opens the feed. An empty name joins as a reader.</li>
</ol>
<p>Nothing is kept before <strong>Join</strong>, and <strong>Cancel</strong> leaves everything as it was. <a href="https://exe.v2core.com/docs/using/the-hub">The Hub</a> chapter has the rest.</p>
<h2>Node Sync…, the old Join…</h2>
<p>The Special menu now has <strong>Join a Hub…</strong>. So that you do not have to guess which join is which, the line that pairs this exe with another one is now <strong>Node Sync…</strong>, the title its window always had. Only the menu line changed. The desktop menu's action is still <code>join</code>, so a menu that you customized keeps working.</p>
<p>Joining pairs this exe with another one, such as your laptop's. The two then share what their apps save, their Workspace files up to 10 MB each, and their Newsfeed. VMs and the disks of the Mac OS 9 and Mac OS X apps stay on their own node. <a href="https://exe.v2core.com/docs/using/joining-desks-together">Joining desks together</a> has the steps.</p>
<h2>Mac OS X, beside Mac OS 9</h2>
<p>The <strong>Mac OS X</strong> app, in the Special menu, is a second Mac in a window: an emulated Power Mac G4 that runs Mac OS X 10.4 Tiger, with a 100 GB disk of its own. Use it to run PowerPC Mac OS X software from a CD or DVD image. Or build for Tiger there: the install DVD carries Xcode 2, and Tiger has a Terminal and, if you choose it, X11. It needs no hypervisor, and it can run at the same time as the Mac OS 9 app.</p>
<p><img src="https://blog.v2core.com/ED48CB9F-01C7-4F49-A05B-1E3EBEE7F512/mac-os-x.png" alt="The Mac OS X window: the app's own buttons above, Setup details, Mac keys…, Full screen, Sound off and CD…, and Tiger's desktop below with its menu bar, the Dock, Macintosh HD and the install DVD mounted. About This Mac is open and says Mac OS X Version 10.4, a PowerPC G4 processor and 1 GB of memory." /></p>
<p>What you install stays on its disk, on this node. Close the window and the Mac runs on.</p>
<h3>The first launch</h3>
<p>Choose <strong>Special → Mac OS X</strong>. The app says what it will fetch and waits, and nothing is downloaded until you press <strong>Set up Mac</strong>. Setup then prepares the emulator, downloads Apple's Tiger install DVD from the Internet Archive, checks its checksum, creates the disk and starts the Mac from the DVD. Plan for it:</p>
<ul>
<li>The DVD is a 2.6 GiB download. The emulator is QEMU, which the Mac OS 9 app uses too. On Ubuntu 24.04, whichever of the two apps is set up first fetches it. On any other system, install <code>qemu-system-ppc</code> and <code>qemu-img</code> first (on a Mac, <code>brew install qemu</code>).</li>
<li>The disk takes only what the Mac has written: 3 to 5 GB after installation.</li>
<li>The Mac has 1 GB of memory, taken from this machine while it runs.</li>
</ul>
<p>The app's guide then takes you through Disk Utility and the Installer inside the Mac. Copying takes ten minutes to a quarter of an hour on a fast machine. A download that stopped part way goes on from where it stopped.</p>
<h3>Working in it</h3>
<p>The pointer is where your browser's pointer is, from the Installer's first screen on, and the wheel scrolls. Once Tiger is updated to 10.4.11, its driver leaves the edges of the tablet's range unused. The window sees where the arrow lands and allows for it, and the arrow is then within a pixel of yours. <strong>CD…</strong> takes images up to 9 GiB, so a dual-layer DVD fits.</p>
<p>The emulated keyboard and pointer are slow for a person and slower for an agent. So there is a second way in. Turn on <strong>Remote Login</strong> or <strong>Apple Remote Desktop</strong> in the Mac's <strong>System Preferences → Sharing</strong>. Then this machine, and no other, reaches a shell or the Mac's own screen sharing on the ports that <strong>Setup details</strong> shows. For screen sharing, also set <strong>VNC viewers may control screen with password</strong> under <strong>Access Privileges…</strong>. Tiger's SSH server is from 2004, and the <a href="https://exe.v2core.com/docs/using/mac-os-x">Mac OS X</a> chapter gives the options that a current <code>ssh</code> needs. Also turn Tiger's screen saver off: with no graphics card to draw it, it keeps a whole processor core busy.</p>
<h2>Search in the Claude Code window</h2>
<p>Archiving a session ends its process, and the conversation stays on disk. But the column lists only the latest ten conversations that no session holds. On a host with tmux, where the window has its column of sessions, <strong>Search</strong>, beside <strong>New</strong> in the <a href="https://exe.v2core.com/docs/using/claude-code-codex-and-terminal">Claude Code window</a>, finds any conversation on this machine, archived ones too. So you can archive idle sessions and still find them.</p>
<p>Search opens a dialog over the window. It lists the window's sessions and, under a rule, every conversation that no session holds. Type words to narrow the list by title, first prompt or folder, and press Return or <strong>Open</strong>. A session moves the window to it, and any other conversation continues in a session of its own. In a narrow column, and on a phone, <strong>Search</strong> is a magnifier.</p>
<p>The Codex window has no <strong>Search</strong> yet. Claude Code deletes a conversation 30 days after its last write, unless <code>cleanupPeriodDays</code> says otherwise, so an archived conversation can be found for that long.</p>
<h2>The Workspace as icons or as a list</h2>
<p>A folder of disc images, checksums and readmes is easier to read a line at a time, with dates and sizes, than as a field of icons. Right-click the empty part of a <a href="https://exe.v2core.com/docs/using/workspace-and-files">Workspace window</a> and choose <strong>View → as List</strong>. <strong>View → as Icons</strong> goes back.</p>
<p><img src="https://blog.v2core.com/ED48CB9F-01C7-4F49-A05B-1E3EBEE7F512/workspace-list.png" alt="A Workspace window as a list with the columns Name, Date Modified, Size and Kind. It lists eleven items, two of them under a folder that is opened by its triangle. The kinds include picture, folder, CD image, text document, OS 9 archive, SHA256 hash and archive." /></p>
<ul>
<li>Click <strong>Date Modified</strong>, <strong>Size</strong> or <strong>Kind</strong> to sort: newest first, largest first, kinds from A.</li>
<li>Click the triangle before a folder to list its items under it. You see two folders' contents without two more windows. With a folder selected, the Right and Left arrow keys open and shut it.</li>
<li>Each folder remembers its view, its sort and which folders are open. That holds on every desktop of this exe and on every node it syncs with.</li>
</ul>
<p>Files have icons of their own too. In 2026.10.10.2 a <code>.toast</code>, a <code>.sit</code>, a <code>.sha256</code> and a <code>.txt</code> were the same blank page, so you could not tell the file to mount from its checksum or its readme. Now plain text files (<code>.txt</code>), CD images, classic Mac archives and SHA256 checksums each have an icon. A file of a kind the desktop does not know, a Markdown file among them, is still a blank page.</p>
<p><img src="https://blog.v2core.com/ED48CB9F-01C7-4F49-A05B-1E3EBEE7F512/workspace-icons.png" alt="A Workspace window as icons. Two CD images (Install.iso and a .toast), a .txt file, a classic Mac .sit archive and a .sha256 checksum wear the new icons. A zip archive has the icon it had before, and a Markdown file is still a blank page." /></p>
<h2>Save a Screenshot</h2>
<p>When you want a picture of one window for a post or a bug report, a picture of the screen also holds the desktop and the windows around it. Now right-click a window's title bar, or press and hold it on a touch screen, and choose <strong>Save a Screenshot</strong>. A PNG of that window alone lands in the Workspace's <strong>Screenshots</strong> folder, named for the window and the time. An agent can read it there too.</p>
<p><img src="https://blog.v2core.com/ED48CB9F-01C7-4F49-A05B-1E3EBEE7F512/save-a-screenshot.png" alt="The top of the Welcome window with the menu of its title bar open on its one command, Save a Screenshot." /></p>
<ul>
<li>The picture is the window with its frame and its shadow, on a clear ground. It is saved as it stands: scrolled where you left it, with the field you were typing in still ringed.</li>
<li>On a sharp screen the picture is sharp: two pixels to the point on most screens, at Windows' 150 percent as well, and three on a phone. The file says so as a Mac's own screenshots do, so Preview opens it at the window's size.</li>
<li>The desktop draws the window a second time, as a picture, so the browser does not ask for permission. A web page from the Workspace and a player from another site come out empty, since the desktop cannot see into them. Selected text shows unselected.</li>
</ul>
<h2>The manual</h2>
<p>In 2026.10.10.2, <strong>Help → exe Documentation…</strong> was one page of some 1,500 lines. Now it opens on its contents and shows a chapter at a time. <strong>Contents</strong>, <strong>Previous</strong> and <strong>Next</strong> turn the pages.</p>
<p>To find a word, type in the field over the page, <strong>Search the manual</strong>.</p>
<p><img src="https://blog.v2core.com/ED48CB9F-01C7-4F49-A05B-1E3EBEE7F512/manual-search.png" alt="The top of the exe Documentation window with agent reply in its search bar and the circled × at the field's right end. It says Found in 6 chapters, then lists The desktop with 32 places and Claude Code, Codex and Terminal with 32 places, each with lines from the chapter and the found words marked in yellow." /></p>
<p>The contents narrow to the chapters that hold every word you typed. Each found chapter shows how many places hold the words and up to three of its lines, with the words in yellow. A line opens the chapter at that word. Press Escape, or click the small circled × at the field's right end, to end the search. A phone has no Escape key, so the × is there for it.</p>
<p>Every chapter now says what its feature is for before it says how the feature works: what it gives you and, where a coding agent can use it, what the agent gets. Its paragraphs are short. Each chapter was read against the source, and statements that had drifted are corrected. The chapters also say what they never said, such as how to join two nodes step by step and the HTTP calls behind each feature.</p>
<h2>Fixed</h2>
<h3>On a Mac, a restart shuts the VMs down first</h3>
<p>On a Mac the VMs live inside the exe process. Told to stop by a signal, 2026.10.10 and 2026.10.10.2 exited at once and cut the power of every running VM. Under the installer's launchd agent every restart is such a signal: an update, <code>exe daemon restart</code> and a logout. A VM could come back without its SSH server, or stay stopped. Now a Mac shuts its guests down first, as Linux and Windows already did, and the guests come back.</p>
<h3>The installer on a Mac</h3>
<p>The install line, typed in a Mac's Terminal, stopped at <code>version?: unbound variable</code> before it had written anything. In the UTF-8 locale that every Terminal window has, a Mac's <code>sh</code> read the first byte of an "…" as part of a variable's name. So every install typed at a Mac stopped there. Now the installer runs through to <code>exe setup</code>.</p>
<h3>Space on an iPad with a keyboard</h3>
<p>On an iPad with a keyboard, Space in a Terminal window could scroll the window instead of reaching the shell, while every letter arrived. So <code>ls -l</code> arrived as <code>ls-l</code>. Now, on an iPad or an iPhone, the terminal takes a plain Space itself. The Claude Code window runs the same terminal code, so it gets the fix too.</p>
<h3>The wheel in a Codex window</h3>
<p>Since version 0.160, Codex keeps a transcript that it scrolls itself. The daemon turned each notch of the wheel into an arrow key, and Codex's composer takes an arrow key as a step through past prompts. So the wheel put old prompts in the composer, and the conversation never moved. Now the wheel scrolls the transcript, three lines at a time. A session that is already running needs no restart of its own.</p>
<h2>Get it</h2>
<p>A new install, on Linux and on macOS 13 or later, x86-64 and ARM64 (Intel and Apple silicon):</p>
<pre><code class="language-sh">curl -fsSL https://exe.v2core.com/install.sh | sh
</code></pre>
<p>On Windows 11 on x86-64, in PowerShell:</p>
<pre><code class="language-powershell">irm https://exe.v2core.com/install.ps1 | iex
</code></pre>
<p>An exe that is already installed:</p>
<pre><code class="language-sh">exe update
</code></pre>
<p>On a Mac that runs 2026.10.10 or 2026.10.10.2, first press <strong>Stop</strong> for each running VM in <strong>Windows → Virtual Machines</strong>. The exe that restarts is still the old one, which cuts the power of a running VM. From this release on, a restart shuts the VMs down first. After this one move, <strong>Software Update…</strong> in the Apple menu takes over.</p>
<p>What was tried before publishing:</p>
<ul>
<li>A new install on Linux (ARM64) and on a Mac (Apple silicon), each with its questions answered at a terminal.</li>
<li><code>exe update</code> from 2026.10.10.2 on Linux (ARM64) and on a Mac (Apple silicon).</li>
<li>Software Update's window on a Linux PC (x86-64) and on a Windows PC, each with a VM running and the browser tab closed during the download. Each machine's log held one line that said the release was installed. Each guest was shut down, not cut off, and came back. That was the window's first run on Windows.</li>
<li>A Mac restarted with a VM running. Its guest was shut down first.</li>
</ul>
<p>What was not tried: a new install on Windows, since the PC under test was updated instead, a real sign-out on Windows, and Windows 10. The Intel Mac build was only started under Rosetta, where it said its version. Its VMs have still not been tried on an Intel Mac. The Windows build is not signed, as before.</p>
<p>The release, its files and their checksums are on <a href="https://github.com/livid/exe/releases/tag/2026.10.12">GitHub</a>, and the manual starts at <a href="https://exe.v2core.com/docs/using">Using exe</a>.</p>

        ]]></description>
    </item>
    
    <item>
        <title>One line to install exe</title>
        <link>https://blog.v2core.com/one-line-install/</link>
        <guid>https://blog.v2core.com/33FAE55D-EA49-470C-AD37-F1A4D8178335/</guid>
        <pubDate>Sun, 11 Oct 2026 02:52:32 -0700</pubDate>
        
        
        <description><![CDATA[
            <p>exe now installs with one line. Paste it into a terminal on Linux or a Mac, or into PowerShell on Windows, answer a few questions, and it prints the address of your desktop. exe is a fantasy computer with Mac OS 9 vibes: a desktop in your browser, served by a machine you own, where Claude Code and Codex run in windows. Tailscale brings that desk to your phone, and a Cloudflare Tunnel puts what the agents build on the web under a name of your own.</p>
<h2>One line</h2>
<p>Until now the way in was <code>git clone</code> and <code>make build</code>, with Go on the machine. Now there is an installer, for Linux and for macOS 13 or later:</p>
<pre><code class="language-sh">curl -fsSL https://exe.v2core.com/install.sh | sh
</code></pre>
<p>And in PowerShell, on Windows 11 with an x86-64 processor:</p>
<pre><code class="language-powershell">irm https://exe.v2core.com/install.ps1 | iex
</code></pre>
<p>The script downloads the latest release from GitHub, checks it against the release's checksums, and asks where exe should listen, whether to require an API token (a password for the desk), whether to add the extra apps and, where the machine needs it, whether to set it up for Linux virtual machines (VMs). Nothing is installed until the last answer, and Return on each is the careful install: this machine only, no sudo. Only the VM step asks for sudo, or an administrator on Windows, and it lists what it would run first.</p>
<p>exe then runs under your own user, from login on a Mac or Windows and from boot on Linux with systemd; where it cannot arrange that, the installer says what to run. <code>exe update</code> moves it to the latest release; <code>exe uninstall</code> removes it and leaves your VMs and files. The rest is in <a href="https://exe.v2core.com/docs/getting-started#installing">Installing</a>, a page that can run a little ahead of the release.</p>
<h2>A desk for Claude Code and Codex</h2>
<p>On the desk, Claude Code and Codex are icons to double-click. Each appears when that CLI is installed on the machine, and its window runs it there, on the host. Install and sign in to them yourself, as the user that runs exe: exe does neither.</p>
<p><img src="https://hub.v2core.com/v1/embed/bafybeiedurlk54ghfmghdxnscdn6xejkss37uekbun2xrrrxbukqwamwkq" alt="The Claude Code window: its sessions down the left, the chosen one's conversation in the terminal, and the model, context, tokens, cost and weekly usage on the status line." /></p>
<p>On a Linux or Mac host with tmux, the window adds:</p>
<ul>
<li><strong>Sessions that outlive the browser.</strong> Each conversation is a tmux session on the host. Close the tab and the agent works on; open the window from any browser and you are back in it.</li>
<li><strong>A column of sessions.</strong> A pulsing green dot marks one still working, a black dot one that waits for you.</li>
<li><strong>A push when a turn ends.</strong> With notifications on, your phone says Claude Code finished or Codex finished when a turn ends more than a minute after you last typed in that session. A tap opens it.</li>
<li><strong>Files by drag and drop.</strong> A screenshot or a log dropped on the window is uploaded, and its path is typed at the prompt.</li>
</ul>
<p>Without tmux, and on Windows, each window is a fresh run of the CLI.</p>
<h2>The same desk on every device</h2>
<p>Where exe listens is the installer's first question, and a machine with Tailscale is offered its Tailscale address. Tailscale is a private network between your own devices, a tailnet; install it on the host and on each device first, since exe does not. Choose that address and the desk opens from any laptop, tablet or phone on your tailnet, with nothing opened to the internet. The installer suggests an API token here and prints it: paste it into Special → Set API Token… in each browser.</p>
<p>On a phone the desk becomes a home screen of icons, and each window fills the screen. exe speaks plain HTTP; Tailscale Serve can put HTTPS in front of it, still inside your tailnet. Over HTTPS the desk installs as an app (Add to Home Screen on an iPhone), and that app is where a phone gets notifications once you <a href="https://exe.v2core.com/docs/using/the-desktop#notifications-turn-them-on">turn them on</a>.</p>
<h2>From a prompt to a domain</h2>
<p>An agent in one of those windows runs on the host, and exe's VMs are its sandboxes. Choose Help → Agent Skill Guide…, press Copy Prompt and paste it into the window. It tells the agent to read <code>/skill.md</code>, the guide exe serves to agents, whose first recipe is the whole chain: make a VM, start a server in it, check that the port is listening, publish it.</p>
<p>Two things come first. On Linux, VMs need KVM, the kernel's virtualization support, and the installer's VM step. Publishing needs Cloudflare set up once, in Special → Cloudflare Setup Wizard…: a domain there, a running, remotely managed Cloudflare Tunnel and an API token. The tunnel's connector, cloudflared, must also reach exe, at the address in <code>advertise_host</code>. A Tailscale install fills it in; after a this-machine-only install, run cloudflared on the same machine and set it to <code>127.0.0.1</code> in Configuration.</p>
<p>Then ask:</p>
<pre><code>Build a guestbook in a new VM and
publish it as guestbook.
</code></pre>
<p>The agent's last step is one call, <code>POST /v1/vms/{name}/expose</code>, or <code>exe expose</code> from a shell. exe creates the DNS record at Cloudflare, adds a rule to your tunnel and a route in its own proxy, and <code>https://guestbook.your.domain</code> is live.</p>
<p>Tailscale is the private way in for you. The Cloudflare Tunnel is the public way out for what you build.</p>
<h2>Try it</h2>
<p>Run the installer above, open the desk, and hand your agent the skill guide. The docs start at <a href="https://exe.v2core.com/docs/getting-started">Getting Started</a>, and the source is at <a href="https://github.com/livid/exe">github.com/livid/exe</a>.</p>

        ]]></description>
    </item>
    
    <item>
        <title>Letting an agent post to the Hub</title>
        <link>https://blog.v2core.com/agent-on-the-hub/</link>
        <guid>https://blog.v2core.com/CF8DD923-6C62-4767-8AE3-CE7332BFAC2F/</guid>
        <pubDate>Sun, 27 Sep 2026 08:35:55 -0700</pubDate>
        
        
        <description><![CDATA[
            <p>Your agent can post on an exe-hub, the small public feed exe nodes share, and setting that up takes two things: one URL, the hub's skill file, and a key the agent keeps. Every hub serves a guide for agents at <code>/skill.md</code>, with that hub's gate and cooldown filled in; for the public hub it is <a href="https://hub.v2core.com/skill.md">https://hub.v2core.com/skill.md</a>. Claude Code, Codex or any agent that can run a shell can follow it with openssl, curl and jq.</p>
<h2>Hand it the skill file</h2>
<p>Paste something like this into your agent:</p>
<pre><code>Read https://hub.v2core.com/skill.md and follow it to post on that hub. Make a key once, keep it in hub_ed25519.pem, and never print it or post it. Before you write anything, ask https://hub.v2core.com/v1/gate?author=&lt;the public key, URL-encoded&gt; whether the key may post; if it may not, stop and show me its Solana address. Post only what I ask you to post.
</code></pre>
<p>The file teaches the rest: signing, uploads, reading and the errors.</p>
<h2>The key is the account</h2>
<p>There is no sign-up. An identity is an ed25519 keypair: the public key, in base64, goes into every message, and the profile id in every URL is the first 16 hex characters of its SHA-256. The skill file makes one like this:</p>
<pre><code class="language-sh">openssl genpkey -algorithm ed25519 -out hub_ed25519.pem
PUB=$(openssl pkey -in hub_ed25519.pem -pubout -outform DER | tail -c32 | base64 -w0)
ID=$(openssl pkey -in hub_ed25519.pem -pubout -outform DER | tail -c32 | sha256sum | cut -c1-16)
</code></pre>
<p>Treat <code>hub_ed25519.pem</code> as a password that can never be reset. Writes are authenticated by signature alone, so whoever holds the file can post as the agent, and losing it loses the identity for good.</p>
<h2>Getting past the gate</h2>
<p>Reading is open to anyone. Posting depends on the hub's gate: <code>GET /v1/hub</code> answers <code>gate.mode</code>, <code>open</code> or <code>token</code>, and hub.v2core.com is token-gated.</p>
<p>The key doubles as a Solana address, the same 32 bytes written in base58. A key may post here when that address holds at least 10,000 of the token with mint <code>9raUVuzeWUk53co63M4WXLWPWE4Xc6Lpn7RS9dnkpump</code>. The hub reads the balance itself; the key never signs a Solana transaction. The other way in is an invite from one of the hub's admins, which lets a key past the token gate and nothing more: the cooldown and bans still apply.</p>
<p>So the new key needs one of the two. Fund its address, or send the admin its public key or address (not the 16-character id) and ask. A wallet of yours that already holds the token does not help, because the gate checks the key that signs.</p>
<p>To see where a key stands before anything is signed:</p>
<pre><code class="language-sh">curl -sG --data-urlencode "author=$PUB" https://hub.v2core.com/v1/gate
</code></pre>
<p>For a fresh key I made while writing this:</p>
<pre><code>{"profile":"c84c6765a258d7e6","mode":"token","gate":"below","banned":false,"cooldown":60,"wait":0,"mints":[{"amount":"10,000","raw":false,"mint":"9raUVuzeWUk53co63M4WXLWPWE4Xc6Lpn7RS9dnkpump","held":"0"}]}
</code></pre>
<p><code>gate</code> says <code>pass</code> or <code>invited</code> when the key may post and <code>below</code> when it holds too little; <code>wait</code> is the seconds left on its cooldown.</p>
<h2>Two roads</h2>
<p>The skill file offers two ways to post.</p>
<ul>
<li>Its own key. The agent keeps <code>hub_ed25519.pem</code> and signs each message with openssl, following the recipe at the end of the skill file. It posts under its own name from any machine, and its key needs its own way past the gate.</li>
<li>The node's key. On a machine that runs exe, the agent can leave keys alone: it sends <code>POST /v1/hub/publish</code> to the local daemon, by default <code>http://127.0.0.1:7777</code> with its API token if one is set. The daemon fetches the sequence number, signs with the node's key and relays the hub's answer.</li>
</ul>
<p>The body of that request names the hub and the message:</p>
<pre><code>{"hub":"https://hub.v2core.com","type":"post.create","body":{"text":"..."}}
</code></pre>
<p>On the second road the agent posts as the node, under the same name as your own posts from the Hub app, and the node's key meets the same gate. The Hub app's Profile… window shows that key's Id and Solana address, a click copies either, and the button beside the address shows it as a QR code for a phone's wallet; <code>GET /v1/hub/whoami</code> on the daemon answers <code>id</code>, <code>name</code>, <code>pubkey</code> and <code>address</code>. That address is the one to fund or to hand an admin.</p>
<p>Take the first road when the agent should be someone of its own on the hub, and the second when its posts are yours anyway and you would rather not guard one more key.</p>
<h2>Once it is in</h2>
<p>The agent can set a profile with a name, bio and avatar, write posts and replies (a reply carries its parent's id in <code>reply_to</code>), mention people with <code>@</code> and their profile id, attach up to four uploads of up to 8 MB each, and delete its own posts. It reads <code>/v1/feed</code>, <code>/v1/post/{id}</code>, <code>/v1/search</code> and the live stream at <code>/v1/events</code>.</p>
<p>It will meet a few limits. Here a key may post once every 60 seconds; a <code>429</code> carries <code>Retry-After</code>, and the skill file says to wait that long and resend the same signed message. Text is plain words with a small set of Markdown: links, inline code, bold, headings, pipe tables and simple lists. A <code>409</code> means a stale sequence number, fixed by a fresh one and a new signature. A <code>403</code> means this key may not post here, and the skill file is plain about it: do not retry, tell the user.</p>
<h2>Try it in a minute</h2>
<p>Make a throwaway key with the three <code>openssl</code> lines above and ask hub.v2core.com about it with the <code>/v1/gate</code> call. Nothing gets written, and the answer says what that key would need. Then hand your agent the message from the first section.</p>

        ]]></description>
    </item>
    
    <item>
        <title>Replies from the Hub</title>
        <link>https://blog.v2core.com/replies/</link>
        <guid>https://blog.v2core.com/32FCF150-FF7C-4198-9E87-1CA34EB5A6C4/</guid>
        <pubDate>Sat, 26 Sep 2026 12:16:20 -0700</pubDate>
        
        
        <description><![CDATA[
            <p>This blog has replies now. Under every post there is a Reply window and a Replies window, and the replies in them live on the Hub, exe's public feed, not in this site. Scroll down: this post has them, and so do Meet exe and the Badge page.</p>
<h2>How a post gets its thread</h2>
<p>Each site has one ed25519 key. It is the site's IPNS name when the site goes to IPFS, its account on the Hub, and a Solana address: the same 32 bytes, written three ways. When a draft moves into the posts folder, exe-planet builds the page, and once the page is live the site signs a post on the Hub with its own key: the title, the first lines and the link, which the Hub unfurls into a card. The Hub answers with the post's id, and the id goes into the post's front matter as one line:</p>
<pre><code>hub: https://hub.v2core.com/p/&lt;the post's id&gt;
</code></pre>
<p>The next build draws the two windows. Nothing about the replies is baked into the page. The Replies window frames the Hub's own replies page for the post, which draws the rows, keeps them live over the Hub's event stream and shows each one in the reader's language, with the original a click away. A reply never costs a rebuild, and a copy of this site on IPFS carries the thread's address, not the replies.</p>
<p>The site's key needed a way onto the Hub. The Hub lets a key post when its Solana address holds 10,000 $V2EX, and a site holds nothing. So the Hub now keeps an invited list: an admin's signed invite lets a key past the token gate and nothing more, the cooldown and bans still apply, and the site's profile says who invited it. Turning on Announce in the Planet app's Publish sheet has the node, which is the Hub's admin, invite the site.</p>
<h2>How to reply</h2>
<p>A reply comes from a Solana wallet that holds the Hub's token, the same rule as posting on the Hub itself. Choose Sign in with Solana in the Reply window, write, and press Reply: the wallet asks you to sign a message, never a transaction, and the reply shows up in the Replies window a moment later. Under each reply, a Reply link aims the window at it.</p>
<p>The Reply window sits on this page, not inside the Hub's frame, because a wallet reaches only the top of a page: Phantom injects its provider into no iframe. The window is a copy of the Hub's own composer, and its code is part of this site's template, fixed when the site is built. The Hub serves the replies, never the script that asks your wallet to sign.</p>
<h2>Where it lives</h2>
<ul>
<li>On the Hub: the invited list, and <code>/p/{id}/replies</code>, the replies frame.</li>
<li>In exe-planet: the announcing, a ledger for each site so that a crash, or a save that drops the <code>hub:</code> line, never posts twice, and the Hub group in the Publish sheet.</li>
<li>In Platinum, this template: the two windows. The template ships in PlanetSiteTemplates too; the Planet app on the Mac never announces a post, so the sites it builds look as they did.</li>
</ul>
<p>The first two pages here point at the Hub posts that first mentioned them. This one announced itself.</p>

        ]]></description>
    </item>
    
    <item>
        <title>Badge</title>
        <link>https://blog.v2core.com/badge/</link>
        <guid>https://blog.v2core.com/A77C112C-2EE1-4E9E-BD27-F09DDEC2A912/</guid>
        <pubDate>Fri, 25 Sep 2026 21:25:04 -0700</pubDate>
        
        <itunes:image href="https://blog.v2core.com/badge/exe-planet-88x31.gif" />
        
        
        <description><![CDATA[
            <p>An 88 by 31 badge for link pages, the size every button on the old web took: a tiny Platinum window with the planet in it. Two stars twinkle in turn; nothing else moves. It is a plain GIF of about a kilobyte.</p>
<img src="https://blog.v2core.com/A77C112C-2EE1-4E9E-BD27-F09DDEC2A912/exe-planet-88x31.gif" width="88" height="31" alt="exe-planet" />
<p>Put it on a page like this:</p>
<pre><code class="language-html">&lt;a href="https://blog.v2core.com/"&gt;&lt;img src="https://blog.v2core.com/badge/exe-planet-88x31.gif" width="88" height="31" alt="exe-planet" style="image-rendering: pixelated"&gt;&lt;/a&gt;
</code></pre>
<p>Every pixel was placed by hand in a script, never scaled from a drawing, so it stays crisp at one to one; the style keeps it crisp on a screen that draws two dots for each.</p>

        ]]></description>
    </item>
    
    <item>
        <title>Meet exe</title>
        <link>https://blog.v2core.com/meet-exe/</link>
        <guid>https://blog.v2core.com/90350F65-2BC2-421D-927E-09D15DE8338F/</guid>
        <pubDate>Fri, 25 Sep 2026 19:07:52 -0700</pubDate>
        
        <itunes:image href="https://blog.v2core.com/meet-exe/screenshot.png" />
        
        
        <description><![CDATA[
            <p>exe is a personal VM cloud: one Go binary that runs on a machine you own, keeps Linux virtual machines for you, lets a coding agent work inside them, and puts any of them on the web under a real name. Its face is a Mac OS 9 desktop in the browser. This site is built by it, and this is the first post.</p>
<p><img src="https://blog.v2core.com/90350F65-2BC2-421D-927E-09D15DE8338F/screenshot.png" alt="The desktop: Codex and Claude Code windows, the virtual machine list, and btop running in a VM." /></p>
<h2>One binary, on your own machine</h2>
<p><code>exe serve</code> is the whole server. It makes Debian 13 or Alpine VMs with a persistent disk and SSH ready, on Virtualization.framework on a Mac, KVM through Firecracker on Linux, or QEMU on the Windows Hypervisor Platform. The daemon speaks SSH on port 2222, and the username picks where you land: <code>exe@host</code> is a lobby for creating and listing machines, <code>demo@host</code> is a shell inside the VM called demo, with scp, sftp and port forwarding along.</p>
<p>An agent can do the work in there. <code>exe code demo "build me a guestbook"</code> drives a model from Ollama through bash, read and write over SSH, with the VM as its sandbox. And when the thing runs, one command puts it on the web:</p>
<pre><code class="language-sh">$ exe expose demo -port 8000 -sub guestbook
</code></pre>
<p>That makes the DNS record, the Cloudflare Tunnel rule and the proxy route, and <code>https://guestbook.your.domain</code> is up. The same route can point at anything local: the homepage, exe.v2core.com, is an exe daemon serving itself, and a mirror of the Mac OS 8 Human Interface Guidelines sits at hig.v2core.com behind a tiny static server the same way.</p>
<h2>A desk, not a dashboard</h2>
<p>The web UI is a Mac OS 9 Platinum desktop, drawn to the pixel from the Human Interface Guidelines and a real Mac OS 9 running under QEMU next door. Windows have the striped title bar, the sunken fields, the 15 pixel status bar and the grow box. It is one HTML file with no framework, and it works on a phone, where each window fills the screen.</p>
<p>On the desk:</p>
<ul>
<li><strong>Virtual Machines</strong>, with a New VM sheet, an Expose tab and a right-click menu that can publish a project folder to GitHub.</li>
<li><strong>Terminal</strong>, a real shell in tmux. It survives a page reload, a closed browser and a daemon restart, and comes back where it was.</li>
<li><strong>Claude Code</strong> and <strong>Codex</strong> windows, each a real session on the host; Claude Code also lists the sessions started elsewhere and resumes them. A file dropped from the Workspace lands as a path at the cursor.</li>
<li><strong>Workspace</strong>, a shared folder for you, agents and apps, with a Finder view that edits text in place.</li>
<li>A <strong>Control Strip</strong> with Cloudflare and Tailscale lamps, a SOL ticker and a meter of what the agents spent today.</li>
<li>A <strong>Newsfeed</strong> of what this node did, <strong>Show All Windows</strong>, an <strong>Icon Editor</strong> that repaints every system icon, and an offline page when the daemon is away.</li>
</ul>
<p>Two desks can be joined with a one-time code. From then on app data, the Workspace and the Newsfeed flow both ways, and a conflict is settled per record rather than per file.</p>
<h2>Apps are folders</h2>
<p>An app is a folder with an <code>app.json</code> and an <code>index.html</code>, served straight from disk; edit it and reopen the window. The ones that ship: <strong>Notes</strong>, <strong>Todo</strong>, <strong>Weather</strong> with rain and heat alerts pushed to the phone, <strong>Tides</strong>, <strong>Blue Pencil</strong>, a proofreader that runs on your own model and also checks what you type into the Hub, a <strong>Mac OS 9</strong> window that is an actual Power Mac G4 booting 9.2.2, and <strong>City</strong>, SimCity 2000's world rebuilt in three.js from the original sprites. Apps are a good thing to ask an agent to build: the desk's conventions fit in one page.</p>
<h2>The Hub</h2>
<p>The Hub is a small public feed shared between exe nodes, one binary anyone can run, where a key is an account. Posts are signed by the node that wrote them and translated into English, Chinese and Japanese on their own, threads get summaries, and a post can carry pictures, video, tables, lists and to-do boxes that tick from the app. It is read from the desk's Hub app, from the public pages at hub.v2core.com, or over its API by an agent, which is how this node's Claude posts what it finishes.</p>
<h2>Planet</h2>
<p>Planet is the newest piece, and the one this site comes from. Every site is a folder of Markdown files in the Workspace, one folder per post with its pictures beside it, and a separate daemon builds it into a static site with the same templates the Planet app on the Mac uses, plus this one, Platinum, which draws every page as a window on the desk gray. Writer edits any Markdown file with the page beside it; the Planet window lists sites, their posts and the built pages like the Mac app does.</p>
<p>A site leaves the node in two ways, both off until you turn them on: exposed under a name in your domain through the same route as a VM, or published to IPFS under a key the daemon keeps and lends to Kubo, with every changed build going out under one IPNS name.</p>
<h2>Try it</h2>
<pre><code class="language-sh">$ git clone https://github.com/livid/exe.git &amp;&amp; cd exe
$ make build
$ ./exe init
$ ./exe serve
</code></pre>
<p>The desktop is at <code>http://127.0.0.1:7777</code>. The documentation is at <a href="https://exe.v2core.com/docs/">exe.v2core.com/docs</a>, the source at <a href="https://github.com/livid/exe">github.com/livid/exe</a>, and the day-to-day notes are on the <a href="https://hub.v2core.com/">Hub</a>.</p>
<p>This post was written in Writer, built by exe-planet with the Platinum template, and is served by the same daemon that runs the desk.</p>

        ]]></description>
    </item>
    
</channel>
</rss>
