<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Oatnil</title>
    <description>The latest articles on DEV Community by Oatnil (@oatnil).</description>
    <link>https://dev.to/oatnil</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3901364%2F169a5621-3dec-4111-904e-f8be3a4db579.jpeg</url>
      <title>DEV Community: Oatnil</title>
      <link>https://dev.to/oatnil</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/oatnil"/>
    <language>en</language>
    <item>
      <title>Shoot together, ask once: everyone's photos in one place, a travel journal from one sentence, shared with one code</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Mon, 16 Nov 2026 13:32:20 +0000</pubDate>
      <link>https://dev.to/oatnil/shoot-together-ask-once-everyones-photos-in-one-place-a-travel-journal-from-one-sentence-3n6c</link>
      <guid>https://dev.to/oatnil/shoot-together-ask-once-everyones-photos-in-one-place-a-travel-journal-from-one-sentence-3n6c</guid>
      <description>&lt;p&gt;Three people spend a day on the coast and come home with three phones and three camera rolls. Getting the photos into one place usually means a group chat that shrinks every picture, or a shared album that everyone has to sign up for, and pay for once it fills up. Once the photos are all in one place, someone still has to sort them and someone has to write it up.&lt;/p&gt;

&lt;p&gt;This post walks through one way to do it in &lt;a href="https://udctl.com/" rel="noopener noreferrer"&gt;udctl&lt;/a&gt; (UnDercontrol): photos sync on their own, everyone's shots end up together in a group, one sentence in the evening gets an agent to write the trip up, and a short code lets family read it without an account. All of it can run on your own machine.&lt;/p&gt;

&lt;p&gt;We made a 3-minute video of this: &lt;a href="https://www.youtube.com/watch?v=qYvkCBpgg7I" rel="noopener noreferrer"&gt;watch it on YouTube&lt;/a&gt;. This post is the written version. The people and photos in it are made-up demo data.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fd6gt4c4o2hnfk08vjirc.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fd6gt4c4o2hnfk08vjirc.jpg" alt="The phone's photo library, with photos three people shared to the group side by side, each carrying a group badge" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Photos land in one place without anyone collecting them
&lt;/h2&gt;

&lt;p&gt;Turn on photo sync in the udctl mobile app and the photos you take upload to your own library. No uploading them one at a time.&lt;/p&gt;

&lt;p&gt;When everyone uses udctl, each person long-presses in the phone's photo library to start selecting, picks the shots they want the others to see, taps Share to group and chooses the group. Photos are stored as originals, not run through a chat app's compression. Tap In group in the filter bar and you see every photo shared into the group, each with a group badge.&lt;/p&gt;

&lt;p&gt;Adding someone to the group takes one invite link. They open it, accept, and they are in.&lt;/p&gt;

&lt;h2&gt;
  
  
  Every photo comes with its time, place and a one-line description
&lt;/h2&gt;

&lt;p&gt;When a photo arrives, the server stores when it was taken and its GPS location, then has an AI model look at the thumbnail and write a short description into the file's description field.&lt;/p&gt;

&lt;p&gt;So later, nobody has to remember what a photo shows, or when and where it was taken. Below is one photo as the agent reads it from the command line: time taken, coordinates, camera model, and the description the AI wrote.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fm85a6ep816s70q9frqts.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fm85a6ep816s70q9frqts.jpg" alt="One photo's details on the command line: time taken, coordinates, camera info, and the description written by an AI model" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  No sorting photos, no writing it up yourself
&lt;/h2&gt;

&lt;p&gt;In the evening, say one sentence to Alfred in chat: "turn today's photos into a travel journal." Alfred opens a card and hands it to your travel writer agent. That agent can be Claude Code, Codex, OpenCode, or any terminal-based agent.&lt;/p&gt;

&lt;p&gt;The agent lists the day's photos with the ud CLI, in time order, and reads each one's place and description. When it needs detail, it downloads the original and looks at it. The journal lands as a note on the card, with each photo set under the paragraph it belongs to. Anything it could not verify (say, a place name worked out only from coordinates) is marked to confirm, so you check it and fix it.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcxajtnuex64735mjsie2.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcxajtnuex64735mjsie2.jpg" alt="The journal note on the card: one section per time of day, each followed by its photo" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The journal is an ordinary Markdown document. Tasks, notes, expenses and accounts in udctl all use the same editor, so changing a word or swapping a photo works the way it does everywhere else.&lt;/p&gt;

&lt;p&gt;Want the others on the trip to edit it too? Share the card with the group, and every member can read and edit it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Anyone without an account can read it with a short code
&lt;/h2&gt;

&lt;p&gt;When the journal is ready, click Share to Public in the card menu. That makes a link and a short access code. Your parents, or friends who don't use udctl, open the share page, type the code, and read the journal with all its photos. No sign-up, no login.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffmhn3uleoaajb3q3m96y.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffmhn3uleoaajb3q3m96y.jpg" alt="A visitor in a phone browser, not logged in, reading the journal after entering the short code" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Your photos sit on storage you choose, and the bill doesn't count gigabytes
&lt;/h2&gt;

&lt;p&gt;The photo library lives where you put it: this computer's disk, a server at home, or your own S3 or R2 bucket.&lt;/p&gt;

&lt;p&gt;Self-hosted udctl licenses are priced by team-size band. One person is free, a team of up to three is free, and no line on the bill charges by capacity (see the &lt;a href="https://udctl.com/docs/pricing" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt;). What storage costs you depends on the disk or bucket you pick.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8xszffoe0ytl97xnnwws.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8xszffoe0ytl97xnnwws.jpg" alt="Where the library lives: your own disk, your own server, or your own S3 / R2 bucket" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The whole service runs on your own machine
&lt;/h2&gt;

&lt;p&gt;You can deploy udctl yourself. One command starts the whole server (Docker image, a single npm-installed binary, or Homebrew; see the &lt;a href="https://udctl.com/docs/self-deployment" rel="noopener noreferrer"&gt;deployment docs&lt;/a&gt;). Rather not run a server? The desktop app ships its own backend and runs fully local. The mobile app connects to that server of yours.&lt;/p&gt;

&lt;p&gt;Three details about your data:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A new instance has sign-up closed by default, so only people you invite get in.&lt;/li&gt;
&lt;li&gt;The mobile, desktop and web apps carry no analytics.&lt;/li&gt;
&lt;li&gt;The AI model that describes photos and writes the journal is the one you connect.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F6skbsjhocc86lggg7wow.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F6skbsjhocc86lggg7wow.jpg" alt="One command in a terminal starts the server; the ready banner shows the database, the storage location, and sign-up closed by default" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Typical scenarios
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;A weekend trip with friends&lt;/strong&gt;: everyone shoots on their own phone, picks a few to share to the group afterwards, one sentence in the evening gets the journal written, and anyone in the group can add a line.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Traveling with your parents&lt;/strong&gt;: you take the photos, then drop the short code into the family chat. Your parents read the journal without installing an app or signing up.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Your kid's weekends&lt;/strong&gt;: one short write-up with photos each week. The photos are already in time and place order, and you only fix the spots the agent marked to confirm.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Site visits and business trips&lt;/strong&gt;: on-site photos carry their own time and location. Have the agent turn them into a visit report along the timeline, then share it with the team.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Getting started
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Self-host: read the &lt;a href="https://udctl.com/docs/self-deployment" rel="noopener noreferrer"&gt;deployment docs&lt;/a&gt; or start from the &lt;a href="https://udctl.com/self-hosting" rel="noopener noreferrer"&gt;self-hosting page&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;Rather not run a server: install the &lt;a href="https://udctl.com/download#desktop" rel="noopener noreferrer"&gt;desktop app&lt;/a&gt;, which runs its own backend locally.&lt;/li&gt;
&lt;li&gt;On your phone, turn on photo sync and point it at your own server. New photos go straight into your library.&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>selfhosted</category>
      <category>productivity</category>
      <category>ai</category>
      <category>photography</category>
    </item>
    <item>
      <title>Let Your Agent Pick the Model: Light Jobs to Fast Models, Hard Ones to the Strongest</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Thu, 12 Nov 2026 05:55:31 +0000</pubDate>
      <link>https://dev.to/oatnil/let-your-agent-pick-the-model-light-jobs-to-fast-models-hard-ones-to-the-strongest-57om</link>
      <guid>https://dev.to/oatnil/let-your-agent-pick-the-model-light-jobs-to-fast-models-hard-ones-to-the-strongest-57om</guid>
      <description>&lt;p&gt;Fixing a typo and designing a whole interface shouldn't cost the same, or take the same wait. But if you have to decide "which model this time?" before every job you hand out, that decision becomes a job of its own.&lt;/p&gt;

&lt;p&gt;This post shows one way to handle it in &lt;a href="https://udctl.com" rel="noopener noreferrer"&gt;udctl&lt;/a&gt; (UnDercontrol): write your model-choice rule once and let Alfred pick and dispatch by it. You skip the judgment call before every dispatch, light jobs stop queuing behind your strongest model, and hard jobs don't land on a model that can't handle them.&lt;/p&gt;

&lt;p&gt;We made a 100-second video of this. The dispatches in it come from a real run. This post is the written version.&lt;/p&gt;

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/j3Pvh3PWjiM" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;p&gt;&lt;a href="https://www.youtube.com/watch?v=j3Pvh3PWjiM" rel="noopener noreferrer"&gt;Watch on YouTube&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqnea4jixecso33ndcmth.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqnea4jixecso33ndcmth.jpg" alt="You say &amp;quot;Dispatch these.&amp;quot; Alfred reads the rule and sends fix a typo, tidy the tags, build the form, design the UI and final review to claude-haiku, claude-sonnet, claude-opus, claude-fable and codex" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Each model is one command, so you can see what runs
&lt;/h2&gt;

&lt;p&gt;In udctl, each model is its own agent CLI, with the model right in the command. The &lt;code&gt;claude-fable&lt;/code&gt; row is &lt;code&gt;claude --dangerously-skip-permissions --model fable&lt;/code&gt;; &lt;code&gt;claude-haiku&lt;/code&gt; is the same command with &lt;code&gt;--model haiku&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;It's not just Claude. Codex, Pi: whatever is installed on your machine, you can add, with "New CLI..." at the bottom of the list. Claude Code, Codex, OpenCode, or any terminal-based agent plugs in the same way.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftq4uzpvox9dhun9fn2ob.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftq4uzpvox9dhun9fn2ob.jpg" alt="The Agent CLI list in settings: Codex, Pi, and four Claude Code rows for Fable, Haiku, Opus and Sonnet, with the --model flag at the end of each command underlined" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Set the usual split once and stop thinking about it
&lt;/h2&gt;

&lt;p&gt;Each agent has its own default: Opus for the coder, Fable for the designer, Codex for the reviewer. In the agent's edit form, the CLI you pick under "Runs with" is its default.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcfz47vqy036idg7owgrn.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcfz47vqy036idg7owgrn.jpg" alt="The reviewer's edit form: Runs with is set to Codex, marked available on one of your machines" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  No more picking a model every time you hand out a job
&lt;/h2&gt;

&lt;p&gt;A default only covers what an agent normally uses. The same coder might fix a typo in a button label today and build a whole form tomorrow, and one model for both is a poor fit. So every job you hand out means picking a model again.&lt;/p&gt;

&lt;p&gt;Write that judgment down as a rule, once. Global Instructions in settings go into the prompt of every workspace session: yours, and those of the agents you trigger. The rule in the video reads:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Choosing a model: default to claude-opus; simple jobs use claude-sonnet; tiny, repetitive jobs use claude-haiku; complex jobs that need strong design use claude-fable; for code review, have codex take a look as well.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fjjlgr72nf49is9zssler.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fjjlgr72nf49is9zssler.jpg" alt="The Global Instructions box holding the model-choice rule" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  One sentence, and each job lands on the right model
&lt;/h2&gt;

&lt;p&gt;With the rule written, you can hand the model decision to Alfred too. In the video, the demo account says one sentence to Alfred: "Dispatch the four jobs on the "Checkout page redesign" card."&lt;/p&gt;

&lt;p&gt;Alfred reads the rule, splits the work into child cards and dispatches it himself. Here is what that real run produced:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Job on the child card&lt;/th&gt;
&lt;th&gt;Dispatched to&lt;/th&gt;
&lt;th&gt;CLI&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;New layout and interaction draft&lt;/td&gt;
&lt;td&gt;designer&lt;/td&gt;
&lt;td&gt;claude-fable&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Build the form and its validation&lt;/td&gt;
&lt;td&gt;coder&lt;/td&gt;
&lt;td&gt;claude-opus&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Fix the typo in the button label&lt;/td&gt;
&lt;td&gt;coder&lt;/td&gt;
&lt;td&gt;claude-haiku&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Review before merge&lt;/td&gt;
&lt;td&gt;reviewer&lt;/td&gt;
&lt;td&gt;claude-opus&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Second-opinion review&lt;/td&gt;
&lt;td&gt;reviewer&lt;/td&gt;
&lt;td&gt;codex&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The last row is a card Alfred added on his own, because the rule says to have codex take a look at code review as well. Each child card got its own session; with Alfred's, that is six.&lt;/p&gt;

&lt;p&gt;Every dispatch he sends names one CLI next to the agent mention:&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fkc0oukda9cvg8h6fqdd8.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fkc0oukda9cvg8h6fqdd8.jpg" alt="Alfred's dispatch on a child card: @designer followed by @claude-fable" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;And the session starts with exactly that command. Open the designer's session and the Command shows &lt;code&gt;--model fable&lt;/code&gt;:&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fdr1cmczhnnmajq8sy8ec.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fdr1cmczhnnmajq8sy8ec.jpg" alt="The designer's session running, with --model fable in the launch command" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;If your Alfred mentions the agent but no CLI, the session starts on that agent's default. Add one line to Alfred's prompt, "When you dispatch, put the CLI that the model-choice rule in Global Instructions calls for next to the agent mention", and his dispatches will carry it from then on.&lt;/p&gt;

&lt;h2&gt;
  
  
  When you want to choose, it's one switch
&lt;/h2&gt;

&lt;p&gt;Sometimes you want to pick it yourself. When you dispatch, switch the CLI in the dropdown above the input box, say claude-haiku for the coder on a typo fix.&lt;/p&gt;

&lt;p&gt;It applies to that one message only. The agent's default stays as it is, and the next message goes back to its usual CLI.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2lammakftxn0ghiytyq1.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2lammakftxn0ghiytyq1.jpg" alt="The CLI dropdown above the input: claude-haiku selected, with the footer &amp;quot;For this message only. coder's own defaults are set under Manage coder.&amp;quot;" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Set it up yourself: three steps
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Add each model as a CLI.&lt;/strong&gt; One row per model under Agent CLI in settings, with the model in the command.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Give each agent a default.&lt;/strong&gt; "Runs with" in the agent's edit form.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Write a model-choice rule in Global Instructions.&lt;/strong&gt; Say which kind of job gets which CLI, then hand the work to Alfred with one sentence: "Dispatch these."&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;You don't have to click through these yourself either. Tell your agent the split you want, let it do the setup, and adjust it with the agent in conversation until it fits.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://udctl.com" rel="noopener noreferrer"&gt;udctl.com&lt;/a&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>agents</category>
      <category>productivity</category>
      <category>devtools</category>
    </item>
    <item>
      <title>Server or Daemon: Where Should Your UnDercontrol (udctl) Agent Run?</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Fri, 06 Nov 2026 22:02:16 +0000</pubDate>
      <link>https://dev.to/oatnil/server-or-daemon-where-should-your-undercontrol-udctl-agent-run-bog</link>
      <guid>https://dev.to/oatnil/server-or-daemon-where-should-your-undercontrol-udctl-agent-run-bog</guid>
      <description>&lt;p&gt;An agent in UnDercontrol (udctl) can now run in two places. Pick &lt;strong&gt;server&lt;/strong&gt; when the job lives inside udctl, such as answering from your cards, creating a card, writing a note or posting a scheduled rundown. Your machine can be off for all of these. Pick &lt;strong&gt;daemon&lt;/strong&gt; when the job needs your code, your shell, or a coding CLI: Claude Code, Codex, OpenCode, or any terminal-based agent.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcgfhvhdzxpd3kxbmljzo.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcgfhvhdzxpd3kxbmljzo.png" alt="Two panels. Server runs inside the udctl backend: it is mentioned or scheduled, runs one turn, works with tasks, cards and notes, and replies in the same conversation, while your machine can be off. Daemon runs on a machine of yours, with Claude Code, Codex, OpenCode or any terminal-based agent, your repo, your shell and a terminal you can watch" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;We made a 58-second video of this: &lt;a href="https://www.youtube.com/watch?v=4mr1paOVkFo" rel="noopener noreferrer"&gt;watch it on YouTube&lt;/a&gt;. This post is the written version.&lt;/p&gt;

&lt;h2&gt;
  
  
  Get an answer while your laptop is closed
&lt;/h2&gt;

&lt;p&gt;A daemon agent starts on a machine of yours, so it needs one online. A server agent runs inside the udctl backend. You install nothing on your computer, and your computer does not have to be on.&lt;/p&gt;

&lt;p&gt;Mention the agent in a conversation and it runs one turn. In that turn it can look up a task, search your tasks, create a card, and add a note to a card. Its final message is the reply, in the same conversation. Each step is recorded, shown live, and kept with the card and the thread. The model calls use your own key.&lt;/p&gt;

&lt;p&gt;A typical case: you are away from your desk and your laptop is shut. You open the web app, mention the agent, ask which release cards are still open, and tell it to add a card for the thing you just remembered.&lt;/p&gt;

&lt;p&gt;New agents run on the server by default. Agents you already had keep running where they ran before.&lt;/p&gt;

&lt;h2&gt;
  
  
  Get a morning rundown without leaving a machine on
&lt;/h2&gt;

&lt;p&gt;Open a conversation, choose &lt;strong&gt;More&lt;/strong&gt;, then &lt;strong&gt;Set as resident conversation&lt;/strong&gt;, and give it a schedule and a message. When the time comes, the server posts that message into the conversation, the agent runs one turn, and the answer stays there. The same job appears under &lt;strong&gt;Settings &amp;gt; Automation&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;In our promo film the schedule is every day at 9 AM and the message asks for the open cards tagged &lt;code&gt;release&lt;/code&gt;. At 09:00 the agent answers with all five, and which one to do first, with no machine online.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F598z6s6fgxgyjqrj5s5m.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F598z6s6fgxgyjqrj5s5m.png" alt="A timeline with three separate turns: a mention, a 09:00 schedule, and a follow-up mention. Each one starts one turn, and each turn ends with a reply in the thread. Between turns nothing is running and the conversation holds the history" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;A server agent works in turns like these. It starts when it is mentioned or when its schedule fires, and it finishes when the reply is written. So set the schedule to the moments you want a report, and read the result when you next open the conversation.&lt;/p&gt;

&lt;h2&gt;
  
  
  Keep coding work next to your code
&lt;/h2&gt;

&lt;p&gt;A daemon agent starts a CLI on your machine, so it has your repository, your shell, and your local tools. Tests run where they always run. Usage counts against the account that CLI is already logged in to. In a daemon conversation, the header's &lt;strong&gt;Stream&lt;/strong&gt; tab shows the terminal, so you can watch the work and step in.&lt;/p&gt;

&lt;p&gt;A daemon agent also has the full &lt;code&gt;ud&lt;/code&gt; CLI. Notes, expenses and accounts share one Markdown editor and one store with your tasks, so a daemon agent can pull a receipt and a spec into the same job. A server agent's tools work with tasks and their notes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Pick by what the job has to touch
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1mtqnjr4piz7o1ie8meg.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1mtqnjr4piz7o1ie8meg.png" alt="A decision chart. Jobs that touch cards, notes and the conversation go to server. Jobs that touch your code, shell or local tools go to daemon. When it depends, you pick server or daemon as you send the message" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;The job&lt;/th&gt;
&lt;th&gt;Pick&lt;/th&gt;
&lt;th&gt;Why&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;"What is still open on the release?"&lt;/td&gt;
&lt;td&gt;server&lt;/td&gt;
&lt;td&gt;It only reads your cards&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Turn a message into a card, or add a note to one&lt;/td&gt;
&lt;td&gt;server&lt;/td&gt;
&lt;td&gt;Both are udctl tools&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A rundown of your cards every morning&lt;/td&gt;
&lt;td&gt;server&lt;/td&gt;
&lt;td&gt;No machine has to be online when it fires&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Fix a bug in a repository&lt;/td&gt;
&lt;td&gt;daemon&lt;/td&gt;
&lt;td&gt;It needs your code and a coding CLI&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Run the tests or a build&lt;/td&gt;
&lt;td&gt;daemon&lt;/td&gt;
&lt;td&gt;It needs your shell&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Keep using your Claude Code or Codex login&lt;/td&gt;
&lt;td&gt;daemon&lt;/td&gt;
&lt;td&gt;The CLI on your machine does the work&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Watch the terminal and step in&lt;/td&gt;
&lt;td&gt;daemon&lt;/td&gt;
&lt;td&gt;The Stream tab exists for daemon conversations&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;If you are unsure, ask one question: does this job need a file on my computer? If yes, use daemon. If no, start with server.&lt;/p&gt;

&lt;h2&gt;
  
  
  Change your mind on a single message
&lt;/h2&gt;

&lt;p&gt;Right after you mention an agent, you can choose where it runs for this message. The chip under the input shows the current choice, such as "runs on server by default", and a click changes it.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A choice outside the agent's declared range is refused, and the conversation tells you why.&lt;/li&gt;
&lt;li&gt;If you pick daemon and no machine is online, you get a receipt at once instead of a silent wait. When the agent's range includes the server, the receipt offers a ready-made line that sends the question there.&lt;/li&gt;
&lt;li&gt;An agent you own can run on the server for one conversation even if its settings list only daemons. Its settings stay as they were, and its other conversations keep using a daemon.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Slack is coming soon
&lt;/h2&gt;

&lt;p&gt;Coming soon: mention the agent in a Slack channel and it replies in the thread. Scheduled channel digests are coming soon too.&lt;/p&gt;

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

&lt;p&gt;We tested this on the hosted site, and both paths work there: mentioning a server agent for an answer, and a job that fires on schedule.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Open &lt;a href="https://app.udctl.com" rel="noopener noreferrer"&gt;app.udctl.com&lt;/a&gt;. Registering a new account needs an invite code. If you do not have one, choose &lt;strong&gt;Visitor (No signup)&lt;/strong&gt; on the sign-in page. Visitor accounts are temporary, so use a model key you can revoke.&lt;/li&gt;
&lt;li&gt;Add your own model key in &lt;strong&gt;Settings&lt;/strong&gt;. A server agent needs it to answer, on either kind of account.&lt;/li&gt;
&lt;li&gt;Create an agent. New agents run on the server by default.&lt;/li&gt;
&lt;li&gt;Start a conversation, type &lt;code&gt;@&lt;/code&gt; and the agent's name, and ask about your cards.&lt;/li&gt;
&lt;li&gt;To put it on a schedule, choose &lt;strong&gt;More &amp;gt; Set as resident conversation&lt;/strong&gt;.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;On your own instance, "server" means your own backend: see &lt;a href="https://udctl.com/self-hosting" rel="noopener noreferrer"&gt;Self-Hosting&lt;/a&gt; and &lt;a href="https://udctl.com/docs/getting-started" rel="noopener noreferrer"&gt;Getting Started&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>agents</category>
      <category>productivity</category>
      <category>devtools</category>
    </item>
    <item>
      <title>DeepSeek Says the Agent Bottleneck Is Human Attention: Watch Fewer Terminal Windows with udctl</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Tue, 03 Nov 2026 15:32:05 +0000</pubDate>
      <link>https://dev.to/oatnil/deepseek-says-the-agent-bottleneck-is-human-attention-watch-fewer-terminal-windows-with-udctl-e44</link>
      <guid>https://dev.to/oatnil/deepseek-says-the-agent-bottleneck-is-human-attention-watch-fewer-terminal-windows-with-udctl-e44</guid>
      <description>&lt;p&gt;On October 31, DeepSeek's Harness team published "Super-agent design exploration: toward more autonomous agent systems" and open-sourced a research prototype, dsh-superagent. The starting point of the article: models need a human to step in less and less to finish a single task, so the bottleneck has moved to human attention. The prototype's README puts the principle in one line: take as little human attention as possible, and give as much value as possible.&lt;/p&gt;

&lt;p&gt;If you keep several Claude Code, Codex or OpenCode windows open at once, you have probably hit this bottleneck already. Past a few windows, you can no longer tell which one is waiting for your answer and which one finished long ago.&lt;/p&gt;

&lt;p&gt;udctl (UnDercontrol) has no feature called super-agent. But each piece the article breaks out has a counterpart in udctl that is already shipped: requests come from events, each task gets its own session, follow-up messages go straight back to that session, the agent speaks up only when it needs a person, and it runs on your own machine. The sections below are organized by what you no longer have to keep track of, and each one names the feature behind it.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftli3gerfxicbf6v040ih.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ftli3gerfxicbf6v040ih.png" alt="Left column: six parts of DeepSeek's Super-agent design. Right column: the udctl feature already shipped for each" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  No more watching a row of terminal windows
&lt;/h2&gt;

&lt;p&gt;In DeepSeek's design, a Worker is a session with its own isolated context, and one Worker does one job.&lt;/p&gt;

&lt;p&gt;In udctl, one card plus one agent is one session. When you @-mention an agent on a card, the backend opens a session for it on that card; mention the same agent on another card and that is a separate session, with no shared context. To see which sessions are running and which are waiting for input, run &lt;code&gt;ud get sessions&lt;/code&gt;. You don't have to tell windows apart from memory.&lt;/p&gt;

&lt;p&gt;As of November 3, our own workspace has had 5,451 of these sessions.&lt;/p&gt;

&lt;h2&gt;
  
  
  No typing a prompt every time
&lt;/h2&gt;

&lt;p&gt;The article calls an agent's input a Trigger: a group-chat message, an email, a calendar reminder, a code review comment, a CI result, rather than a person typing a prompt into a box.&lt;/p&gt;

&lt;p&gt;udctl has three kinds of entry points that wake an agent:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;an @mention in a card's description, a note or a comment;&lt;/li&gt;
&lt;li&gt;a message to Alfred through Telegram, Discord or the desktop app's quick input;&lt;/li&gt;
&lt;li&gt;a scheduled job: at the set time it creates a card whose @mention wakes the agent, or it uses &lt;code&gt;wake_agent_session&lt;/code&gt; to wake an existing session.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The scheduled job that prepares grammar lessons in our previous post, &lt;a href="https://udctl.com/blog/2026/10/25/hand-off-the-small-daily-decisions/" rel="noopener noreferrer"&gt;Hand Off the Small Daily Decisions&lt;/a&gt;, had fired 88 times by October 24, and its owner never typed a prompt for it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Replies go back to the session doing the work
&lt;/h2&gt;

&lt;p&gt;In DeepSeek's design, follow-up messages on the same Thread (a PR, an IM topic, an email chain) bypass the Router and go straight to the Worker subscribed to it.&lt;/p&gt;

&lt;p&gt;udctl's comment threads work the same way. Once an agent has posted in a thread, it is subscribed to it; when you reply in that thread later, the message goes straight into its session on that card. If the session drops halfway, the next one starts by reading that same thread, with everything said in between.&lt;/p&gt;

&lt;h2&gt;
  
  
  Called in only when you have to decide
&lt;/h2&gt;

&lt;p&gt;The agent in the article speaks up in IM only when it needs a person to decide or sign off, and the team also tried capping each reply at 100 characters.&lt;/p&gt;

&lt;p&gt;In udctl 0.181.0, the sidebar's "Needs you" lists only threads that @-mention you. An agent finishing a turn no longer counts as "waiting for you"; a thread shows up there only when the agent actually @-mentions you. Every other thread stays in Conversations for when you want to look.&lt;/p&gt;

&lt;h2&gt;
  
  
  One sentence in, handed out to a team of agents
&lt;/h2&gt;

&lt;p&gt;The article's Router layer dispatches and does no work itself: it takes in all messages and hands tasks to Workers.&lt;/p&gt;

&lt;p&gt;udctl has Agent Teams. Give a lead agent a roster of members; the roster is injected into every session it starts, and it @-mentions the right member for each job. This post came about that way: the author wrote one line on a card ("see if we can ride this trend"), Alfred @-mentioned the content team's lead, and the lead did the research and the draft and handed them back in the same thread.&lt;/p&gt;

&lt;h2&gt;
  
  
  Runs on your own machine
&lt;/h2&gt;

&lt;p&gt;The article mentions BYOM (bring your own machine): when a device drops and comes back, the work picks up again.&lt;/p&gt;

&lt;p&gt;udctl agent sessions run on a daemon. The desktop app registers itself as a daemon as soon as it starts; on a machine without it, run &lt;code&gt;ud daemon start&lt;/code&gt;. When an agent is woken, the backend picks one of your own devices first, and uses one someone shared with you only when none of yours is online. If everything is offline, the message stays in the thread, and the next message after a machine comes back opens a new session. An agent can also declare that it runs on the server, with no online device needed.&lt;/p&gt;

&lt;h2&gt;
  
  
  What agents write sits next to your own records
&lt;/h2&gt;

&lt;p&gt;The card descriptions, notes and comments an agent reads and writes use the same Markdown editor as your expense notes and account descriptions. The progress and conclusions an agent writes live alongside your own records, and you find them by path, tag and query.&lt;/p&gt;

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

&lt;ol&gt;
&lt;li&gt;Install the desktop app (it registers itself as a daemon), or run &lt;code&gt;ud daemon start&lt;/code&gt; on any machine.&lt;/li&gt;
&lt;li&gt;On a card, @-mention an agent and write one sentence about what you want done. The agent can be Claude Code, Codex, OpenCode, or any terminal-based agent.&lt;/li&gt;
&lt;li&gt;Go do something else. When it needs you, that thread shows up in "Needs you".&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Reference: DeepSeek's prototype repository, &lt;a href="https://github.com/deepseek-ai/dsh-superagent-prototype" rel="noopener noreferrer"&gt;https://github.com/deepseek-ai/dsh-superagent-prototype&lt;/a&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>productivity</category>
      <category>devtools</category>
      <category>cli</category>
    </item>
    <item>
      <title>Run Your Team's Server From Your Desktop</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Mon, 05 Oct 2026 04:15:45 +0000</pubDate>
      <link>https://dev.to/oatnil/run-your-teams-server-from-your-desktop-4j3c</link>
      <guid>https://dev.to/oatnil/run-your-teams-server-from-your-desktop-4j3c</guid>
      <description>&lt;p&gt;The udctl desktop app has a full backend inside it. Install it and your computer is already a server: it listens on &lt;code&gt;localhost:24816&lt;/code&gt; and keeps tasks, notes and ledgers in a SQLite file on your own disk. Add a license and that same server becomes a multi-user team backend. Add a Cloudflare Tunnel and your phone, your laptop and your teammates can reach it from anywhere. No VPS, no public IP, no router port forwarding.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxekafolootgf3nju6qzu.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fxekafolootgf3nju6qzu.jpg" alt="Diagram: your computer runs the udctl desktop app with its built-in backend on localhost:24816 and SQLite data on disk; a cloudflared tunnel connects it to the public internet, and your phone and your team reach it over HTTPS" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;You do not need to know Docker either. Most self-hosting guides start with a container, environment variables and a reverse proxy. Here the server is a desktop app you install like any other, and it runs as soon as you open it. The terminal commands in this post are all for Cloudflare's tunnel, not for the server.&lt;/p&gt;

&lt;p&gt;The 78-second video below walks the whole path. The rest of this post is the same path as commands you can copy.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://www.youtube.com/watch?v=xp3N_cU-_zw" rel="noopener noreferrer"&gt;Watch the 78-second walkthrough on YouTube&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: Install the desktop app
&lt;/h2&gt;

&lt;p&gt;Download it from &lt;a href="https://udctl.com/download/" rel="noopener noreferrer"&gt;udctl.com/download&lt;/a&gt;: macOS (Apple Silicon and Intel), Windows x64, or Linux x64 AppImage. The desktop app is not a web page in a window. It ships the Go backend, and the backend starts with the app. There is no Docker image to pull and no database to set up. Once the app is installed, the server is installed.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fsvezd7321iw1973b8hfn.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fsvezd7321iw1973b8hfn.jpg" alt="The udctl desktop app showing a kanban board with To do, In progress and Done columns, served by the backend running on the same computer" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Check that it is up:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl http://localhost:24816/health
&lt;span class="c"&gt;# {"status":"healthy", ...}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;24816 is a fixed port. If something else already holds it, the app shows the backend log instead of quietly moving to another port, so a tunnel pointed at 24816 never ends up talking to nothing. (Older guides, including our July post, used 8888. That is gone.)&lt;/p&gt;

&lt;p&gt;Used alone, this is the Personal tier: one user, free, no license key. The app signs you in automatically with a random password generated for this install. You can see it, and reset it, on your profile page under account details; the CLI and the phone use those same credentials.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: Activate a license to make it a team backend
&lt;/h2&gt;

&lt;p&gt;Skip this step if the server is only for you.&lt;/p&gt;

&lt;p&gt;For a team, open your profile page, go to &lt;strong&gt;System &amp;amp; Language → License&lt;/strong&gt;, and fill in four fields:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;License token&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Host secret&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Admin email&lt;/strong&gt;, which becomes your admin login&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Admin password&lt;/strong&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Click &lt;strong&gt;Activate&lt;/strong&gt;. The backend restarts, which takes a few seconds, and the License section shows tier &lt;strong&gt;Pro&lt;/strong&gt;. On the first activation the data already on this machine moves to the admin account you just entered, and the built-in personal account is removed. Nothing is lost: your tasks, notes and records change owner, not content. The form warns you before you click, because this step happens once and cannot be undone.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F6fmaj2wky3spx2r10rgz.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F6fmaj2wky3spx2r10rgz.jpg" alt="The License section under System &amp;amp; Language after activation: tier shows Pro (Active) with an expiry date, while a notice at the top reads " width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;From here you are a server admin. Create accounts for teammates under &lt;strong&gt;Admin → Access → Users → Create User&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Where the license comes from: per the &lt;a href="https://udctl.com/docs/pricing/" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt;, a team of up to 3 people is free, with one free license for the team; &lt;a href="https://udctl.com/contact" rel="noopener noreferrer"&gt;contact us&lt;/a&gt; to get it. Larger teams are priced by band: US$49/month up to 20 people, US$149/month for any size.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: Put it on the internet with one command
&lt;/h2&gt;

&lt;p&gt;Cloudflare Tunnel works by reverse connection: &lt;code&gt;cloudflared&lt;/code&gt; on your computer dials out to Cloudflare's edge, and requests from outside come back through that connection. Because the connection goes out from your machine, you need no public IP and you never touch the router.&lt;/p&gt;

&lt;p&gt;Install cloudflared:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# macOS&lt;/span&gt;
brew &lt;span class="nb"&gt;install &lt;/span&gt;cloudflared

&lt;span class="c"&gt;# Windows&lt;/span&gt;
winget &lt;span class="nb"&gt;install &lt;/span&gt;Cloudflare.cloudflared

&lt;span class="c"&gt;# Linux (Debian/Ubuntu): add the repo from https://pkg.cloudflare.com first&lt;/span&gt;
&lt;span class="nb"&gt;sudo &lt;/span&gt;apt &lt;span class="nb"&gt;install &lt;/span&gt;cloudflared
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;cloudflared tunnel &lt;span class="nt"&gt;--url&lt;/span&gt; http://localhost:24816
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A few seconds later it prints a random &lt;code&gt;https://&amp;lt;something&amp;gt;.trycloudflare.com&lt;/code&gt; address. That is your public endpoint, with HTTPS, and it needs no Cloudflare account. Check it from your phone on cellular data (Wi-Fi off, so you know the request really comes in from outside): open &lt;code&gt;https://&amp;lt;something&amp;gt;.trycloudflare.com/health&lt;/code&gt; and look for &lt;code&gt;healthy&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fn5nhp8lfxbbjoh84sxaf.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fn5nhp8lfxbbjoh84sxaf.jpg" alt="A terminal running cloudflared tunnel --url http://localhost:24816, which reports the tunnel was created and is routing traffic to localhost:24816. The address shown is a placeholder; a real quick tunnel prints a trycloudflare.com URL" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Anyone who has that URL reaches your sign-in page, so give the admin account a real password before you share it.&lt;/p&gt;

&lt;h3&gt;
  
  
  For daily use: a fixed address
&lt;/h3&gt;

&lt;p&gt;The quick tunnel's hostname changes every time &lt;code&gt;cloudflared&lt;/code&gt; restarts. For a server your team relies on, use a named tunnel. It needs a free Cloudflare account and a domain on Cloudflare.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;cloudflared tunnel login
cloudflared tunnel create ud
cloudflared tunnel route dns ud ud.example.com
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;~/.cloudflared/config.yml&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight yaml"&gt;&lt;code&gt;&lt;span class="na"&gt;tunnel&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;ud&lt;/span&gt;
&lt;span class="na"&gt;credentials-file&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;/Users/you/.cloudflared/&amp;lt;tunnel-id&amp;gt;.json&lt;/span&gt;

&lt;span class="na"&gt;ingress&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
  &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;hostname&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;ud.example.com&lt;/span&gt;
    &lt;span class="na"&gt;service&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;http://localhost:24816&lt;/span&gt;
  &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;service&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;http_status:404&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;cloudflared tunnel run ud        &lt;span class="c"&gt;# run it by hand first&lt;/span&gt;
&lt;span class="nb"&gt;sudo &lt;/span&gt;cloudflared service &lt;span class="nb"&gt;install&lt;/span&gt; &lt;span class="c"&gt;# then install it as a service that starts at boot&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then tell the desktop app its public address: profile page, &lt;strong&gt;System &amp;amp; Language → Public access address&lt;/strong&gt;, enter &lt;code&gt;https://ud.example.com&lt;/code&gt;, Save. The backend restarts and builds file download links from that address, so attachments open on remote devices too. Without it, remote clients get &lt;code&gt;localhost&lt;/code&gt; links for attachments. A trycloudflare address is fine for a demo but not here, because it changes on every run and old file links break.&lt;/p&gt;

&lt;p&gt;Last thing: stop the computer from going to sleep in your power settings. It is a server now.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: Connect your phone
&lt;/h2&gt;

&lt;p&gt;The iOS app is in public beta on &lt;a href="https://testflight.apple.com/join/st2TnaBF" rel="noopener noreferrer"&gt;TestFlight&lt;/a&gt;. Android installs from the APK on the &lt;a href="https://udctl.com/download/" rel="noopener noreferrer"&gt;download page&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;On the iOS sign-in screen, find the &lt;strong&gt;Backend&lt;/strong&gt; section and tap &lt;strong&gt;Change&lt;/strong&gt;. Enter your public address with &lt;code&gt;/api/v1&lt;/code&gt; at the end:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;https://ud.example.com/api/v1
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The app does not add &lt;code&gt;/api/v1&lt;/code&gt; for you. As you type, a status line checks the address; when it reads &lt;code&gt;connected · Pro&lt;/code&gt;, the phone has reached the server on your desk. Save, then sign in with your admin email (or the account an admin created for you). On the Personal tier, use the username and password from the desktop profile page instead.&lt;/p&gt;

&lt;p&gt;The phone now shows the same tasks, notes and ledgers as the desktop. Edit a task on the phone and the desktop has the change.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0008rllya7zddro75y1k.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0008rllya7zddro75y1k.jpg" alt="The iOS app showing the Shipping v1.4 task list with the status line " width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Not on a desktop?
&lt;/h2&gt;

&lt;p&gt;The desktop app is one way to run the server. There are two more, both covered in the &lt;a href="https://udctl.com/docs/self-deployment/" rel="noopener noreferrer"&gt;Self-Deployment Guide&lt;/a&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Docker&lt;/strong&gt;: an all-in-one image with the frontend and backend in one container, published for linux/amd64 and linux/arm64.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Raw binary&lt;/strong&gt;: &lt;code&gt;ud-server&lt;/code&gt;, published on npm with the web UI compiled into the binary, for macOS, Linux and Windows. It needs Node.js 18+ and no Docker.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What it costs
&lt;/h2&gt;

&lt;p&gt;The server is your desktop and Cloudflare's quick and named tunnels are free, so the setup itself costs nothing, and none of it asks you to learn Docker. The one optional purchase is a domain for a fixed address. udctl is free for one person, with no license key, and free for a team of up to 3.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Desktop app: &lt;a href="https://udctl.com/download/" rel="noopener noreferrer"&gt;udctl.com/download&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Pricing: &lt;a href="https://udctl.com/docs/pricing/" rel="noopener noreferrer"&gt;udctl.com/docs/pricing&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;License activation guide: &lt;a href="https://udctl.com/docs/subscription-tiers" rel="noopener noreferrer"&gt;udctl.com/docs/subscription-tiers&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>selfhosted</category>
      <category>cloudflare</category>
      <category>productivity</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>We only take what you hand us</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Sat, 19 Sep 2026 06:24:42 +0000</pubDate>
      <link>https://dev.to/oatnil/we-only-take-what-you-hand-us-49kc</link>
      <guid>https://dev.to/oatnil/we-only-take-what-you-hand-us-49kc</guid>
      <description>&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzv2dnmqb6lun0cslbgtc.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzv2dnmqb6lun0cslbgtc.png" alt="We only take what you hand us" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;On 18 September, someone reverse-engineered ZCode, Zhipu's desktop coding assistant, and found that once you were signed in it packed your entire working directory and uploaded it in the background. Not just the code you were editing — the full Git commit history, the LFS cache, the local branch records too. For the commercial project that got posted, one snapshot came to roughly 313 MB across forty-two thousand files, with &lt;code&gt;.git&lt;/code&gt; accounting for 86.6% of it.&lt;/p&gt;

&lt;p&gt;Zhipu apologised that afternoon. The problem came from a repository-indexing feature that shipped enabled by default, it was fixed, and they promised to open-source the client, bring in a third-party audit, and give every user an extra weekly quota reset. That is a decent response, and a fast one.&lt;/p&gt;

&lt;p&gt;But there are two more layers here that matter more than "enabled by default".&lt;/p&gt;

&lt;h3&gt;
  
  
  The gap
&lt;/h3&gt;

&lt;p&gt;When you open a coding assistant, what are you handing it? This file. This stack trace. This function, how should I change it. What ZCode took was the entire repository from the first commit onward: the bad commit you deleted long ago, the branch you force-pushed away that the reflog still remembers, the key that used to sit in a config file before you moved it out.&lt;/p&gt;

&lt;p&gt;The problem isn't whether data ends up on a server — the data of every cloud tool ends up on a server. The problem is &lt;strong&gt;the distance between the small thing you handed over and the whole thing that got taken&lt;/strong&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  The switches
&lt;/h3&gt;

&lt;p&gt;The second thing is worse than the default.&lt;/p&gt;

&lt;p&gt;ZCode's settings had two switches that looked relevant. According to the reverse-engineering write-up, "experience optimisation" controlled whether your data was used to train models, and "repository snapshot indexing" controlled whether the server indexed snapshots &lt;strong&gt;it had already received&lt;/strong&gt;. Turn both off and the local packing and uploading happened anyway. One user reproduced it with everything switched off and still found a 699 MB local snapshot sitting on disk, with a trail of upload logs.&lt;/p&gt;

&lt;p&gt;Shipping enabled by default is a bad decision, and a bad decision can be reversed. A switch whose name describes one thing while it governs another is a different category: the user takes the correct action, gets the wrong result, and never finds out. As far as they know, they already dealt with it.&lt;/p&gt;

&lt;p&gt;And a third: the private key lives only on the server. Content is encrypted with AES, the key is wrapped with an RSA public key, and the decryption key is never handed down. So even if you catch the upload on your own machine, you can't open it — you cannot audit what was taken. Zhipu says the uploaded data is "destroyed immediately and not retained". That may well be true, and a user has no way at all to check it.&lt;/p&gt;

&lt;h3&gt;
  
  
  Our side
&lt;/h3&gt;

&lt;p&gt;UnDercontrol keeps tasks, notes, expenses and accounts in one set of data, with the same Markdown editor across all of them, and the CLI, desktop app, web app and iOS client all working on that same content. For something shaped like that, saying "we take privacy seriously" means nothing. Here is the specific version.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;We never scan your disk.&lt;/strong&gt; In the shipped &lt;code&gt;ud&lt;/code&gt; binary, the number of times the &lt;code&gt;git&lt;/code&gt; command is invoked is zero. It doesn't read &lt;code&gt;.git&lt;/code&gt;, doesn't snapshot, doesn't index, and uploads nothing in the background. There are exactly four places in the whole CLI that walk a directory, and every one of them walks a path you named: the directory you point &lt;code&gt;ud pull&lt;/code&gt; / &lt;code&gt;ud push&lt;/code&gt; at, &lt;code&gt;ud sync resource &amp;lt;dir&amp;gt;&lt;/code&gt;, installing a skill package, and its own state directory. &lt;code&gt;ud sync&lt;/code&gt; also skips every file and directory beginning with &lt;code&gt;.&lt;/code&gt;, so even when you deliberately ask it to sync a repository, &lt;code&gt;.git&lt;/code&gt; is never enumerated.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Agent status reporting is one word.&lt;/strong&gt; When you run Claude Code, Codex, OpenCode or any other terminal agent inside UnDercontrol, the hook we install tells the server what state the session is in. The entire request body is a single field whose value is one of &lt;code&gt;pending&lt;/code&gt;, &lt;code&gt;running&lt;/code&gt;, &lt;code&gt;planning&lt;/code&gt;, &lt;code&gt;awaiting_input&lt;/code&gt;, &lt;code&gt;stopped&lt;/code&gt;, &lt;code&gt;exited&lt;/code&gt;. The hook script does read the working directory the agent CLI hands it, but that path is only used locally, to match a file on your own machine. It never goes into the request.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;There is no analytics SDK in any of the three clients.&lt;/strong&gt; Not in the web app, the desktop app or the iOS app: no sentry, posthog, mixpanel, amplitude, segment, firebase, bugsnag, datadog. The web page's HTML loads no third-party script either.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The desktop app makes no background connections.&lt;/strong&gt; Checking for updates is a single GET, sent only when you click the button, which fetches a small static file from a CDN and compares version numbers. It doesn't download, it doesn't install, and the request has no body. The external hostnames that appear anywhere in the desktop source fall into four groups: the API address you configured, that release manifest, the documentation site, and the official installer script of an agent CLI at the moment you choose to install one.&lt;/p&gt;

&lt;h3&gt;
  
  
  You can check all of this yourself
&lt;/h3&gt;

&lt;p&gt;You have no reason to believe any of the above simply because I wrote it.&lt;/p&gt;

&lt;p&gt;There is no channel here you cannot read. Every request the client makes goes to the address you configured — in the web app you change &lt;code&gt;custom_api_url&lt;/code&gt; in settings, and the CLI takes &lt;code&gt;UD_API_URL&lt;/code&gt; as an environment variable or &lt;code&gt;ud config set-context&lt;/code&gt;. Point it at your own proxy and every byte is visible to you. There is no second hard-coded reporting host, and no encryption layer that only the server can open, which is exactly what ZCode's users did not have.&lt;/p&gt;

&lt;h3&gt;
  
  
  Five places content leaves your machine
&lt;/h3&gt;

&lt;p&gt;Having said what we don't take, I should say what we do. The earlier sentence only means something if this list is complete.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Cloud sync itself.&lt;/strong&gt; If you use our hosted backend, the tasks, notes, comments, expenses and attachments you write live on the server. That is what the product does.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;AI features.&lt;/strong&gt; When you use AI, what you submit goes to a model provider. Two cases, and the difference is large: &lt;strong&gt;with your own provider configured, the browser talks to the provider directly and the conversation never passes through our backend&lt;/strong&gt;; with the provider preconfigured on our cloud, the key can't be handed to the browser, so the request is proxied by us. Server-side features — creating tasks from an image, reading a receipt, summarising a task — always go through the backend. We don't train any model on your content; what a provider does with what it receives is governed by that provider's own terms.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The Workspace remote terminal.&lt;/strong&gt; When you run an agent session through UnDercontrol, its terminal output is sent to the server so you can see where it has got to from your phone or another machine. That output lives only in the server process's memory, is gone the moment the process restarts, and is never written to the database — the reason is recorded in the code: the last few screens of a terminal routinely contain passwords and deploy scripts. What is persisted is session metadata only: the working directory path, the launch command, the first prompt you typed, the status, the process id.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Voice input on iOS.&lt;/strong&gt; This uses the system's own speech recognition. The audio is never sent to us, we only receive the transcribed text — but with Apple's Speech framework on its default settings, the audio goes to Apple's servers to be transcribed.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Website analytics.&lt;/strong&gt; udctl.com loads Google Analytics with IP anonymisation, to count page views. The apps don't, and there is no advertising or cross-site tracking anywhere.&lt;/li&gt;
&lt;/ol&gt;

&lt;h3&gt;
  
  
  Two ways to keep it from leaving at all
&lt;/h3&gt;

&lt;p&gt;&lt;strong&gt;The desktop app with a local database.&lt;/strong&gt; Your data sits on your own machine and touches no server.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Self-hosting.&lt;/strong&gt; Run your own instance, on your own infrastructure. Then "what left my machine" is a question you answer by reading your own logs.&lt;/p&gt;

&lt;p&gt;Both of these have always been there. Neither was added for this post.&lt;/p&gt;

&lt;h3&gt;
  
  
  The full policy
&lt;/h3&gt;

&lt;p&gt;It is at &lt;a href="https://udctl.com/privacy" rel="noopener noreferrer"&gt;udctl.com/privacy&lt;/a&gt;: what we store, who else touches it, how long it is kept, how to delete it — including why audit logs are not deleted along with your account, which is that an audit log the subject can erase is not an audit log.&lt;/p&gt;

&lt;p&gt;If there's a line in here you'd like to check for yourself, point the API address at your proxy and run it for a day. If it doesn't match what I wrote, write to me.&lt;/p&gt;

</description>
      <category>privacy</category>
      <category>security</category>
    </item>
    <item>
      <title>Your context shouldn't live only on someone else's machine</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Sat, 12 Sep 2026 14:52:40 +0000</pubDate>
      <link>https://dev.to/oatnil/your-context-shouldnt-live-only-on-someone-elses-machine-46d2</link>
      <guid>https://dev.to/oatnil/your-context-shouldnt-live-only-on-someone-elses-machine-46d2</guid>
      <description>&lt;p&gt;&lt;em&gt;Reading back everything you said to an agent today sounds about as much fun as writing your own standup notes.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;But if you don't, tomorrow it opens with "Hi, what project are you working on?" all over again.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Rather than re-explaining yourself every morning, let the work you finished land in a library you own, gather it once a day, and let the next agent pick up from there.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fighn4i9n5th3s226d4tw.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fighn4i9n5th3s226d4tw.jpg" width="799" height="460"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;For the past year I've been moving between Claude Code, Codex and OpenCode, and I've written the same set of conventions three times: one copy in this agent's directory, one in that agent's config, one in my own head. Every time I thought that was finally settled. Then I'd open a different terminal and start explaining again.&lt;/p&gt;

&lt;p&gt;Something else was going on at the same time. The work we actually finished together was scattered. Some of it in session logs, some in scratch files, some in a vendor's database, the copy I can read but can't take with me.&lt;/p&gt;

&lt;p&gt;One of those looks like an efficiency problem and the other like an ownership problem. They're two ends of the same thing.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;You work with agents all day, and when you stop, the chat box is the only thing that remembers.&lt;/strong&gt;&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;So I started building my own knowledge base. It began because Obsidian sync kept getting in my way, and because I wanted the data to stay with me: a pile of markdown files I could take anywhere, in a format no single app owned. It grew into a whole workbench, called udctl (formerly UnDercontrol).&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3hq2hs848aird3is4m0r.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3hq2hs848aird3is4m0r.jpg" alt="udctl's Explorer: tasks and files in the same tree." width="799" height="460"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Fig 1 · udctl's Explorer: tasks and files in the same tree.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  1. More agents every month, and you're still the only one who remembers
&lt;/h2&gt;

&lt;p&gt;Start with the problem. If you also work through terminal agents, you've probably hit the same few things.&lt;/p&gt;

&lt;p&gt;Claude Code, Codex, OpenCode, Gemini CLI, Cline. Terminal agents have been arriving faster than anyone can adapt to them.&lt;/p&gt;

&lt;p&gt;The more of them you run, the harder these get to answer:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;That conclusion an agent dug up yesterday, where is it now?&lt;/li&gt;
&lt;li&gt;I wrote the same convention into three agents' directories. Which copy is current?&lt;/li&gt;
&lt;li&gt;What did my agents and I actually finish this month? Is there a list?&lt;/li&gt;
&lt;li&gt;New machine, different agent, and the context starts from zero again. Why?&lt;/li&gt;
&lt;li&gt;I worked this out last week. Why am I working it out again today?&lt;/li&gt;
&lt;li&gt;Whose database is my work sitting in, and if I stop using it one day, can I take all of it out?&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The second-to-last one is the expensive one. &lt;strong&gt;What repeated work costs you isn't the extra half hour. It's that the second time through, you have no idea you're doing it again.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fkk5yhpv148n6nilcoi6a.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fkk5yhpv148n6nilcoi6a.png" alt="Before: an agent's memory ended when the session did." width="800" height="347"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Fig 2 · Before: an agent's memory ended when the session did.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  2. The five things this is trying to do
&lt;/h2&gt;

&lt;p&gt;One thing first. There is no shortage of tools that let you write things down: note apps, wikis, a hundred markdown editors.&lt;/p&gt;

&lt;p&gt;Most of them stop in the same place. They store it, format it, and put it in front of a human. That's the end of the job.&lt;/p&gt;

&lt;p&gt;We care about the other half. Can an agent read that back and write to it directly, with no copy-paste step in the middle?&lt;/p&gt;

&lt;p&gt;udctl in one line: a self-hosted workbench that runs on your own machine. People and agents share one copy of the data and read the same bytes.&lt;/p&gt;

&lt;p&gt;The chain it's trying to complete has five steps. &lt;strong&gt;One container holds it, any agent can read it, it gathers once a day, it settles into long-term memory, and it lands in a directory tree you can open.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbgd07wzpl6e6locsjae9.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbgd07wzpl6e6locsjae9.png" alt="People and agents share one container: the same card, the same note, the same bytes." width="800" height="347"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Fig 3 · People and agents share one container: the same card, the same note, the same bytes.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;First: see what your agents are doing&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;udctl manages agents as first-class objects. You can see:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;which sessions are running, and which card each one is attached to&lt;/li&gt;
&lt;li&gt;which agents are registered in the fleet, and what skills each one carries&lt;/li&gt;
&lt;li&gt;whether the last scheduled run succeeded&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;code&gt;ud get sessions&lt;/code&gt;, &lt;code&gt;ud get agents&lt;/code&gt;, &lt;code&gt;ud get scheduled-jobs&lt;/code&gt;. Three commands.&lt;/p&gt;

&lt;p&gt;It looks at the work surface, not the bill. Usage and cost is a layer we don't do, and don't need to, because agents start from your own login shell, on your own machine, on your own subscription. There's no second billing system, so there's no second set of books to reconcile.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Second: pull scattered conventions into one copy&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The hard part isn't managing tools. It's managing the conventions inside them.&lt;/p&gt;

&lt;p&gt;One rule can be sitting in three agents' directories at once. A skill you spent an afternoon tuning has to be rewritten when you switch tools. udctl puts them in one library instead. Write your review conventions into a skill of your own called &lt;code&gt;code-reviewer&lt;/code&gt;, and &lt;code&gt;ud describe skill code-reviewer&lt;/code&gt; hands the same bytes to Claude Code, to Codex and to OpenCode. All three are reading the one copy on your server.&lt;/p&gt;

&lt;p&gt;Change it once, and every agent picks it up on its next run.&lt;/p&gt;

&lt;p&gt;What that builds is a shared context layer above the individual agents. Write it in one place, and it applies everywhere.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fheuk3x78xdkm1r7zj5tg.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fheuk3x78xdkm1r7zj5tg.png" alt="Reuse across agents: skills, conventions and context go into udctl, then flow back out to each agent." width="800" height="387"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Fig 4 · Reuse across agents: skills, conventions and context go into udctl, then flow back out to each agent.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Third: a card is more than a to-do&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;In udctl a task is a general-purpose container. It can be a to-do. It can also be a document, a decision record, meeting notes, or a draft you haven't finished.&lt;/p&gt;

&lt;p&gt;Under that it splits into three layers. The card body is a living document you edit whenever. Notes are an append-only timeline of how it became what it is today. Comments are short conversation, with people and agents both talking in them.&lt;/p&gt;

&lt;p&gt;There's one editor. Tasks, notes, expenses, accounts: everywhere you can type, it's the same Markdown editor.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fo2whe9x18exntkj0sx15.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fo2whe9x18exntkj0sx15.jpg" alt="Card body / notes / comments, three layers, with one editor everywhere you can type." width="799" height="460"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Fig 5 · Card body / notes / comments, three layers, with one editor everywhere you can type.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fourth: a daily gather, and a pipeline that throws things out&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;This one needs stating precisely. The setup below is one we built ourselves, not something udctl ships with. The product gives you the container and the scheduled jobs. The arrangement is ours.&lt;/p&gt;

&lt;p&gt;On our ledger card, one entry lands per day. On the fourth day the oldest entry moves out of the near pile and is archived into that month's card, and its key points are squeezed into a single monthly compaction.&lt;/p&gt;

&lt;p&gt;The moving out is the part that does the work. Without it, "accumulating" is a pile that grows until nobody reads it, which is not far from keeping nothing at all.&lt;/p&gt;

&lt;p&gt;Here are the numbers from my own ledger. For the full 31 days of August 2026, the raw daily entries came to 919,538 characters. The monthly compaction that stays in the near pile and gets re-read every day is 13,506 characters. Not a word of the original is gone. It's in the month card, and I can go read it when I need to.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8dmnomqyiwnp1ekpqnmo.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8dmnomqyiwnp1ekpqnmo.png" alt="Daily entry, archived into the month card on day four, squeezed into a monthly compaction, with durable facts moving into the knowledge-base card." width="799" height="313"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Fig 6 · Daily entry → archived into the month card on day four → squeezed into a monthly compaction → durable facts into the knowledge-base card.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fifth: what settles has to land somewhere you can open&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;All of it needs a place to live. In udctl every card and every file hangs on a virtual path: &lt;code&gt;/content/blog&lt;/code&gt;, &lt;code&gt;/ud/docs/architecture/&lt;/code&gt;, &lt;code&gt;/agents/claude-cert-tutor/lessons/D1/&lt;/code&gt;. Together they make a directory tree. Even the built-in skills are on that tree, at &lt;code&gt;/system/skills/ud-cli/&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;The tree holds more than cards. Images, PDFs, docx, spreadsheets, CSVs, code, drawio diagrams and HTML all go under the same path, lying next to the cards.&lt;/p&gt;

&lt;p&gt;HTML is worth its own line. Upload a &lt;code&gt;.html&lt;/code&gt; and opening it in the app gives you the rendered page, not a screen of source. Styles apply, scripts run, and download and full-screen sit right beside it. Our own architecture doc about virtual paths is a &lt;code&gt;.html&lt;/code&gt; that has been sitting under &lt;code&gt;/ud/docs/architecture/&lt;/code&gt; all along, and we open it when we need it.&lt;/p&gt;

&lt;p&gt;The whole tree can also come down to disk. &lt;code&gt;ud init /Notes/&lt;/code&gt; binds a local folder to a virtual path, and &lt;code&gt;ud pull&lt;/code&gt; pulls the cards and files under that path down together, subfolder for subfolder. Cards land as &lt;code&gt;.md&lt;/code&gt;, with the filename as the title and a single &lt;code&gt;ud-id&lt;/code&gt; line of frontmatter. Files land as themselves. Your editor, your grep and your git all work on them directly.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frvyh26vu62fkjhyelx4t.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frvyh26vu62fkjhyelx4t.jpg" alt="Cards, images and .html side by side on one tree; opening the .html gives you the rendered page." width="799" height="460"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Fig 7 · Cards, images and &lt;code&gt;.html&lt;/code&gt; side by side on one tree; opening the &lt;code&gt;.html&lt;/code&gt; gives you the rendered page.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Whose disk these bytes sit on
&lt;/h2&gt;

&lt;p&gt;There's one line we decided not to cross while building this. Your working record shouldn't have to live with us.&lt;/p&gt;

&lt;p&gt;Your record of working with agents holds a lot: what you're building, where you're stuck, who the client is, how the money moves, the approach you used to get around something. That information is valuable to you, and it's valuable to other people too.&lt;/p&gt;

&lt;p&gt;So udctl is self-hosted. After you install it, your cards, notes, daily entries and books land in a SQLite file under &lt;code&gt;~/.ud-server/data&lt;/code&gt; on your own machine.&lt;/p&gt;

&lt;p&gt;On first launch it doesn't even ask for a password. You open it, and you're in your workbench.&lt;/p&gt;

&lt;p&gt;One thing to be straight about: every model call still sends that call's context to the model vendor, and nobody can skip that step. But the copy that stays behind, the cards, the notes, the daily entries, is in that file on your machine.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Your record is yours.&lt;/strong&gt; That's the whole of what "self-hosted" means here.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Who this is for
&lt;/h2&gt;

&lt;p&gt;If any of these describes you, it may be worth a look:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;you work through terminal agents every day, and more than one of them&lt;/li&gt;
&lt;li&gt;you wrote your conventions into a skill and ended up with a copy in three tools' directories&lt;/li&gt;
&lt;li&gt;you have a pile of "I'm fairly sure I solved this before" problems and can't get back to the record&lt;/li&gt;
&lt;li&gt;you'd rather not put your working record, your clients' names and your books in a third party's database&lt;/li&gt;
&lt;li&gt;you want the agent that starts tomorrow to already know where today got to&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  5. Getting started: three ways to install
&lt;/h2&gt;

&lt;p&gt;There isn't much to pick up. Take whichever one suits you.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Option one: a single command&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;npm i -g @oatnil/ud-server &amp;amp;&amp;amp; ud-server
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Six seconds to install, one to start. Open a browser and it's there, with the database created for you.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Option two: Docker.&lt;/strong&gt; Frontend and backend in one image, one command:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;docker run -d -p 3000:8080 \
  -e HOST_DOMAIN=http://localhost:3000 \
  -e JWT_SECRET=change-me-to-a-random-string \
  -e ADMIN_EMAIL=admin@example.com \
  -e ADMIN_PASSWORD=changeme \
  -v undercontrol-data:/app/data \
  lintao0o0/undercontrol:latest
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Open &lt;code&gt;http://localhost:3000&lt;/code&gt; and sign in with the email and password you set above. Data goes to SQLite by default. This is the one to leave running on a small machine at home.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Option three: the CLI on its own.&lt;/strong&gt; &lt;code&gt;npm i -g @oatnil/ud&lt;/code&gt; gives you both &lt;code&gt;ud&lt;/code&gt; and &lt;code&gt;udctl&lt;/code&gt;. This is what the agents use in the terminal.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where to start
&lt;/h2&gt;

&lt;p&gt;The ledger pipeline above is ours, and you'd probably build a different one. If you don't know where to begin, the smallest version is this: before you stop for the day, have your agent write what it did into a note on one card.&lt;/p&gt;

&lt;p&gt;Do that for a week. When the card gets too long to read, go think about archiving and compaction.&lt;/p&gt;

&lt;p&gt;Project: &lt;a href="https://udctl.com" rel="noopener noreferrer"&gt;https://udctl.com&lt;/a&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>productivity</category>
      <category>devtools</category>
      <category>cli</category>
    </item>
    <item>
      <title>Why a Ledger Gets Harder to Continue After One Missing Entry</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Thu, 30 Jul 2026 13:24:49 +0000</pubDate>
      <link>https://dev.to/oatnil/why-a-ledger-gets-harder-to-continue-after-one-missing-entry-1dpf</link>
      <guid>https://dev.to/oatnil/why-a-ledger-gets-harder-to-continue-after-one-missing-entry-1dpf</guid>
      <description>&lt;p&gt;When one entry is missing, the first thing that breaks is often not the monthly report. It is your trust in the ledger.&lt;/p&gt;

&lt;p&gt;You know it is already inaccurate. The next time you open it, adding that one entry is not enough. Did you miss others before it? Did the same purchase appear once in each of two channels? Was the original expense reversed after a refund arrived? If any of those questions remains unanswered, continuing feels like adding numbers to a crooked table.&lt;/p&gt;

&lt;p&gt;That is where many people stop. They still want a record of their spending, but they do not want to spend more time on a result they no longer trust.&lt;/p&gt;

&lt;p&gt;A recent real reconciliation covered four months, several payment and banking channels, and hundreds of existing records. There was no assumption about where the problems would be. The comparison surfaced omissions, duplicates, refunds, and old-data issues:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Some expenses appeared on statements but not in the ledger.&lt;/li&gt;
&lt;li&gt;Some purchases left a record in more than one channel.&lt;/li&gt;
&lt;li&gt;Some recorded expenses were later refunded in full but remained in the totals.&lt;/li&gt;
&lt;li&gt;Some old records had no date, so monthly reports could not see them.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Memory is a poor way to reconstruct any of these. Refunds and undated old records are especially difficult because they do not look like errors. They quietly change the result.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqpcezrunp4iqlipywz7i.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqpcezrunp4iqlipywz7i.png" alt="Cross-channel duplicate: one purchase leaves similar records in two channels and the existing ledger, so it is flagged as a possible duplicate for human confirmation" width="800" height="347"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Reconcile First, Then Change the Ledger
&lt;/h3&gt;

&lt;p&gt;The agent did not read the statements and immediately change the data.&lt;/p&gt;

&lt;p&gt;It first compared the statements with the existing records and produced an analysis: confirmed omissions, possible cross-channel duplicates, possible refunds, and cases that required a decision. At this stage, it presented evidence without writing or deleting anything.&lt;/p&gt;

&lt;p&gt;The human supplied the judgment. Whether a transfer counts as spending, how a refund should be classified, and which of two duplicate records should remain all depend on the owner's accounting rules. The agent wrote and corrected records only after the proposed treatment was approved.&lt;/p&gt;

&lt;p&gt;Writing was not the end. It read the ledger back and checked the resulting count and total against the approved list. For every deduplication or deletion, it retained the record identifier, the reason, and the outcome. A conclusion could be traced back to the original record, and the write itself could be checked for mistakes.&lt;/p&gt;

&lt;p&gt;Auditability matters more than a claim that the work was automatic. If automation says only, “I recorded it for you,” the person still does not know what it missed or merged. Analysis first, approval before writing, and a read-back afterward make the work safe to delegate.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fdrw9e763ynyhe2s4i7gn.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fdrw9e763ynyhe2s4i7gn.png" alt="Auditable reconciliation flow: analyze and show evidence, wait for human confirmation, write approved changes, read back and check, then retain the record and reason" width="800" height="300"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  A Missing Entry No Longer Means Starting Over
&lt;/h3&gt;

&lt;p&gt;The ledger does not require a perfect daily habit. At the end of the month, the agent can compare statements from different sources with the existing records and surface omissions. Forgetting one entry does not invalidate everything recorded before it.&lt;/p&gt;

&lt;p&gt;Duplicates do not have to be guessed by eye either. Similar times, amounts, and descriptions are only clues. Uncertain matches are listed for a human decision instead of being merged silently.&lt;/p&gt;

&lt;p&gt;The person's work is reduced to the few places that require judgment: defining the rules, reviewing evidence, and confirming the treatment. The agent handles the statement-by-statement comparison, cross-checking, data entry, and read-back.&lt;/p&gt;

&lt;p&gt;When a reconciliation can show what it inspected, what it proposed changing, and what it changed, a missing entry becomes one issue to resolve instead of a reason to abandon the entire ledger.&lt;/p&gt;

</description>
      <category>productivity</category>
      <category>ai</category>
      <category>personalfinance</category>
      <category>tooling</category>
    </item>
    <item>
      <title>No Server? Your Desktop Is the Server</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Sun, 26 Jul 2026 12:27:51 +0000</pubDate>
      <link>https://dev.to/oatnil/no-server-your-desktop-is-the-server-pf6</link>
      <guid>https://dev.to/oatnil/no-server-your-desktop-is-the-server-pf6</guid>
      <description>&lt;p&gt;Where your data lives is a fork in the road when you pick a tool. SaaS is the convenient path: sign up and go, nothing to maintain, at the cost of your data sitting on the provider's servers. UnDercontrol will offer that hosted option too. But if what you want is full data self-custody — tasks, notes, and ledgers stored only on your own devices, where backup means copying a file and migration means moving to another machine — this tutorial is for you.&lt;/p&gt;

&lt;p&gt;Self-custody usually means running your own server, and that is exactly where most people give up. Here is a zero-cost path that skips the server entirely: the UnDercontrol desktop app ships with a full backend and keeps your data in a SQLite file on your own disk; Cloudflare Tunnel exposes that machine to the internet for free; the iOS app connects straight to your own endpoint. Check tasks, jot notes, and log expenses from the subway, while the data itself never lives anywhere but your computer. No public IP, no router port forwarding, and no domain to buy (optional, only if you want a fixed address).&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2wg494tbwc3b029hnd0a.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2wg494tbwc3b029hnd0a.png" alt="Architecture: the desktop app embeds a server, cloudflared dials out to Cloudflare's edge, and iOS reaches it over HTTPS from any network" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: Install the desktop app, meet the embedded server
&lt;/h2&gt;

&lt;p&gt;Install the UnDercontrol desktop app (macOS / Windows / Linux) from the &lt;a href="https://oatnil.com/download" rel="noopener noreferrer"&gt;download page&lt;/a&gt;. It is not a web-page shell: a complete Go backend ships inside, and the moment the app starts, a local server is listening on port 8888 (it moves to the next free port if 8888 is taken), storing data as a SQLite file on your own disk.&lt;/p&gt;

&lt;p&gt;On first launch you will see "Welcome to UnDercontrol" and a Start button. One click and you are in: the app auto-creates a local account &lt;code&gt;personal@undercontrol.local&lt;/code&gt; with the default password &lt;code&gt;personal123&lt;/code&gt;, no registration, no invite code. Remember these credentials, you will need them on the phone in step 4. And change the default password before exposing anything to the internet; the Security section below shows how.&lt;/p&gt;

&lt;p&gt;Confirm the server is alive:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl http://localhost:8888/health
&lt;span class="c"&gt;# {"service":"ud-go-production","status":"healthy"}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;If you see &lt;code&gt;healthy&lt;/code&gt;, your "server" is ready. It is the computer you are using right now.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: One command to reach the internet
&lt;/h2&gt;

&lt;p&gt;Cloudflare Tunnel works by reverse connection: &lt;code&gt;cloudflared&lt;/code&gt; on your computer dials out to Cloudflare's edge, and outside traffic is relayed back through that connection. Because the connection is outbound, you need no public IP and never touch your router settings.&lt;/p&gt;

&lt;p&gt;Install cloudflared:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# macOS&lt;/span&gt;
brew &lt;span class="nb"&gt;install &lt;/span&gt;cloudflared

&lt;span class="c"&gt;# Windows&lt;/span&gt;
winget &lt;span class="nb"&gt;install &lt;/span&gt;Cloudflare.cloudflared

&lt;span class="c"&gt;# Linux (Debian/Ubuntu): add the repo from https://pkg.cloudflare.com first&lt;/span&gt;
&lt;span class="nb"&gt;sudo &lt;/span&gt;apt &lt;span class="nb"&gt;install &lt;/span&gt;cloudflared
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then one command:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;cloudflared tunnel &lt;span class="nt"&gt;--url&lt;/span&gt; http://localhost:8888
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fprc8vbdh6zqhip04l7cx.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fprc8vbdh6zqhip04l7cx.png" alt="Real terminal output: quick tunnel created, and curl against the public address returns healthy" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;A few seconds later the terminal prints a random &lt;code&gt;https://xxx.trycloudflare.com&lt;/code&gt; address: your public endpoint, HTTPS included, no Cloudflare account required. Verify from your phone (turn off Wi-Fi so you are on cellular, which proves the path is really public): open &lt;code&gt;https://xxx.trycloudflare.com/health&lt;/code&gt; in the browser. &lt;code&gt;healthy&lt;/code&gt; means the whole chain works.&lt;/p&gt;

&lt;p&gt;Quick tunnels are for trying the flow out: the address changes every time you restart &lt;code&gt;cloudflared&lt;/code&gt;, and Cloudflare makes no uptime promise for them. For daily use, spend five minutes upgrading to a named tunnel.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3 (recommended): A named tunnel with a fixed address
&lt;/h2&gt;

&lt;p&gt;A named tunnel needs a free Cloudflare account and a domain hosted on Cloudflare. The domain costs a few dollars a year and is the only place money can enter this setup.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# 1. Log in; authorize your domain in the browser&lt;/span&gt;
cloudflared tunnel login

&lt;span class="c"&gt;# 2. Create the tunnel&lt;/span&gt;
cloudflared tunnel create ud

&lt;span class="c"&gt;# 3. Point a subdomain at it&lt;/span&gt;
cloudflared tunnel route dns ud ud.example.com
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Write a config file at &lt;code&gt;~/.cloudflared/config.yml&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight yaml"&gt;&lt;code&gt;&lt;span class="na"&gt;tunnel&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;ud&lt;/span&gt;
&lt;span class="na"&gt;credentials-file&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;/Users/you/.cloudflared/&amp;lt;tunnel-id&amp;gt;.json&lt;/span&gt;

&lt;span class="na"&gt;ingress&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
  &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;hostname&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;ud.example.com&lt;/span&gt;
    &lt;span class="na"&gt;service&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;http://localhost:8888&lt;/span&gt;
  &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="na"&gt;service&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s"&gt;http_status:404&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Run it, then register it as a system service so it survives reboots:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;cloudflared tunnel run ud        &lt;span class="c"&gt;# run manually first&lt;/span&gt;
&lt;span class="nb"&gt;sudo &lt;/span&gt;cloudflared service &lt;span class="nb"&gt;install&lt;/span&gt; &lt;span class="c"&gt;# then install as a service&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;From now on &lt;code&gt;https://ud.example.com&lt;/code&gt; is your fixed endpoint. Remember to stop the computer from sleeping in the system power settings — it is your server now.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: Point the iOS app at your endpoint
&lt;/h2&gt;

&lt;p&gt;The iOS app is in &lt;a href="https://testflight.apple.com/join/st2TnaBF" rel="noopener noreferrer"&gt;public beta on TestFlight&lt;/a&gt; (install TestFlight from the App Store first).&lt;/p&gt;

&lt;p&gt;Open the app. At the bottom of the sign-in screen there is an &lt;strong&gt;API server&lt;/strong&gt; section showing the current server. Tap &lt;strong&gt;Change&lt;/strong&gt; and enter your public address, &lt;strong&gt;including the &lt;code&gt;/api/v1&lt;/code&gt; suffix&lt;/strong&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;https://xxx.trycloudflare.com/api/v1
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The status dot probes as you type: when it turns green with &lt;code&gt;connected · Personal&lt;/code&gt;, your phone has reached the server on your computer. Tap &lt;strong&gt;Save&lt;/strong&gt;, then sign in with the account from step 1: &lt;code&gt;personal@undercontrol.local&lt;/code&gt; and your password.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2hz1yn8in6ixbvwgwafu.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2hz1yn8in6ixbvwgwafu.png" alt="iOS sign-in screen: API server pointing at the trycloudflare address, status connected · Personal" width="800" height="1739"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;What you see on the phone is the data on your computer: tasks, notes, ledgers. Edit on the phone and the desktop has it.&lt;/p&gt;

&lt;p&gt;One detail: iOS requires HTTPS for public domains (Cloudflare Tunnel provides it, nothing to do). If you enter a LAN or private-mesh IP instead, write &lt;code&gt;http://&lt;/code&gt; explicitly; without a scheme the app assumes &lt;code&gt;https://&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Security
&lt;/h2&gt;

&lt;p&gt;Public exposure means anyone with the URL can reach your endpoint. Three things, in order of importance:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. Change the default password first.&lt;/strong&gt; Every data endpoint requires login, but the desktop's default password &lt;code&gt;personal123&lt;/code&gt; is public knowledge. Exposing it unchanged is taping the key to the door. The password comes from the &lt;code&gt;PERSONAL_TIER_PASSWORD&lt;/code&gt; environment variable; on macOS, launch from a terminal with it set:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nv"&gt;PERSONAL_TIER_PASSWORD&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s1"&gt;'your-strong-password'&lt;/span&gt; /Applications/UnderControl.app/Contents/MacOS/UnderControl
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;On Windows, add &lt;code&gt;PERSONAL_TIER_PASSWORD&lt;/code&gt; in System Environment Variables and restart the app. Phone and CLI then log in with the new password.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. A random URL is not a security mechanism.&lt;/strong&gt; The trycloudflare address is hard to guess, but do not treat that as protection; think before sharing the link.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. For an extra gate&lt;/strong&gt;, &lt;a href="https://developers.cloudflare.com/cloudflare-one/policies/access/" rel="noopener noreferrer"&gt;Cloudflare Access&lt;/a&gt; on the free plan can put an email-code check in front of a named tunnel. Configure it once and it applies to every device.&lt;/p&gt;

&lt;h2&gt;
  
  
  Don't want public exposure at all?
&lt;/h2&gt;

&lt;p&gt;If all you need is "my devices can reach each other", skip the public internet entirely: use a private mesh network tool to put your phone and computer on the same virtual LAN. Install the client on each device and they see each other; your computer gets a virtual-LAN IP, and the iOS app endpoint becomes &lt;code&gt;http://&amp;lt;virtual-lan-ip&amp;gt;:8888/api/v1&lt;/code&gt; (write &lt;code&gt;http://&lt;/code&gt; explicitly). Zero public exposure, and the free tiers of these tools are usually plenty for personal use. The mesh setup itself is out of scope here; every vendor's docs cover the phone-plus-computer case.&lt;/p&gt;

&lt;h2&gt;
  
  
  After it works
&lt;/h2&gt;

&lt;p&gt;The bill: VPS $0, tunnel $0, HTTPS certificate $0. The only optional cost is a domain at a few dollars a year.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Desktop download: &lt;a href="https://oatnil.com/download" rel="noopener noreferrer"&gt;https://oatnil.com/download&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;iOS TestFlight: &lt;a href="https://testflight.apple.com/join/st2TnaBF" rel="noopener noreferrer"&gt;https://testflight.apple.com/join/st2TnaBF&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>tutorial</category>
      <category>selfhosted</category>
      <category>cloudflare</category>
      <category>ios</category>
    </item>
    <item>
      <title>Pin the Discussion to the Line It's About</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Sat, 25 Jul 2026 07:07:48 +0000</pubDate>
      <link>https://dev.to/oatnil/pin-the-discussion-to-the-line-its-about-55mb</link>
      <guid>https://dev.to/oatnil/pin-the-discussion-to-the-line-its-about-55mb</guid>
      <description>&lt;p&gt;You send out a spec and the discussion immediately comes apart from it. Someone replies in Slack: "that field in the third section, is it nullable?" Now you scroll back to the doc, count down to the third section, and guess which field they meant. Read the same message a day later and the context is gone.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fw2efbngylbik6ilq73le.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fw2efbngylbik6ilq73le.png" alt="Anchored comments: discussion pinned to the line it is about" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Anchored comments in UnDercontrol keep the discussion next to the original text. Select a sentence in the task body, click Comment in the bubble menu, and a comment carrying that quote appears in the right sidebar while the sentence itself turns into a highlight. Click the highlight later and the sidebar jumps to the matching thread. The quote lives on the comment, so the markdown in the task body is never touched. Copy it out or pull it down from the command line and you get exactly what you wrote.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbouofc7o7j9xueg7lw0m.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbouofc7o7j9xueg7lw0m.png" alt="A highlighted sentence in the task body with its anchored comment thread open in the sidebar" width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;When a discussion is settled, hit Resolve. The highlight disappears from the body, the thread collapses into a single grey line, and you can reopen it whenever you want. Threads update on their own when someone replies, with no page refresh. On a phone there are no highlights and the comments sit below the body.&lt;/p&gt;

&lt;h3&gt;
  
  
  Reviewing a spec like you review code
&lt;/h3&gt;

&lt;p&gt;A reviewer no longer has to write "third section, second sentence". They select the line that says "supports bulk import", ask "what's the upper limit?", and the author answers in the same thread. Agree, resolve, move on.&lt;/p&gt;

&lt;p&gt;Once you have been through a spec, the highlights still lit on the page are the parts nobody settled. There is no separate write-up to do afterwards, because the unresolved threads are the to-do list.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fgke24639ynjw15b5hx0s.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fgke24639ynjw15b5hx0s.png" alt="The comments sidebar after a review pass: resolved threads collapsed, unsettled ones still open" width="776" height="1560"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Point at a line and let an agent change it
&lt;/h3&gt;

&lt;p&gt;This is the part that is not like a normal comment section. Select the requirement you want changed, leave a comment that @mentions your agent, and say what it should become. The mention starts an agent session, reusing one that is already running, and when the agent is done its reply is the next comment in that thread, sitting alongside your colleagues' replies with the commit hash in it. Click the highlight and you are back at the text that changed, side by side.&lt;/p&gt;

&lt;p&gt;People and agents talk in one thread, so you are not relaying messages between two tools.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1xn73vpqs78lnsjx4vys.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F1xn73vpqs78lnsjx4vys.png" alt="A human comment mentions an agent and the agent replies in the same thread with a commit hash" width="690" height="612"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Section-by-section discussion on long documents
&lt;/h3&gt;

&lt;p&gt;Write the task body as a design doc and record each iteration as a note. Discussions anchor to their own section: comments on the body in one group, comments on each note in a group of their own, and unanchored chatter in a third. Three months later, why a decision went the way it did is written next to the sentence it changed.&lt;/p&gt;

&lt;h3&gt;
  
  
  The same thread from your terminal
&lt;/h3&gt;

&lt;p&gt;Comments are just as readable and writable from a terminal. The ud CLI treats them as a resource like any other, so an agent can read the discussion a human left in the browser and reply to it:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# read every discussion on a task&lt;/span&gt;
ud get comments &lt;span class="nt"&gt;--task&lt;/span&gt; 02c137b3

&lt;span class="c"&gt;# reply to a thread&lt;/span&gt;
&lt;span class="nb"&gt;cat&lt;/span&gt; &lt;span class="o"&gt;&amp;lt;&amp;lt;&lt;/span&gt;&lt;span class="sh"&gt;'&lt;/span&gt;&lt;span class="no"&gt;EOF&lt;/span&gt;&lt;span class="sh"&gt;' | ud apply -f -
---
task_id: 02c137b3
parent_id: 7a1c9f2e
---
Per-user rate limiting is in, commit a1b2c3d4.
&lt;/span&gt;&lt;span class="no"&gt;EOF
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Claude Code, Codex, OpenCode, or any terminal-based agent works here without an adapter layer. Close a thread out with &lt;code&gt;ud patch comment &amp;lt;id&amp;gt; --status resolved&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;Anchored comments have been available since v0.88.0, in the browser, in the desktop app, and from the CLI.&lt;/p&gt;

</description>
      <category>productivity</category>
      <category>devtools</category>
      <category>ai</category>
      <category>markdown</category>
    </item>
    <item>
      <title>Turn a Task into a Parcel Locker</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Fri, 24 Jul 2026 10:51:06 +0000</pubDate>
      <link>https://dev.to/oatnil/turn-a-task-into-a-parcel-locker-1ic9</link>
      <guid>https://dev.to/oatnil/turn-a-task-into-a-parcel-locker-1ic9</guid>
      <description>&lt;p&gt;The computer at a print shop isn't yours, and you're not going to log into anything on it. The usual ways to move a file there are all awkward: messaging apps want a QR scan and a login session, cloud-drive share pages nag you to install their app, and the USB stick is at home.&lt;/p&gt;

&lt;p&gt;A task in UnDercontrol is already a container: body text, notes, and file attachments. Generate a public share link for it, and anyone who has the link can read the content and download the files. No account, nothing to install. It works like a parcel locker: you drop things in, get a pickup code, and whoever has the code takes them out.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F12tjqk67gxtgdsnhuqli.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F12tjqk67gxtgdsnhuqli.png" alt="Concept: a task working as a parcel locker" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 1: put the files on a task
&lt;/h3&gt;

&lt;p&gt;Create a task and upload the files in its attachments section. The task body is Markdown, so write the instructions right next to the files. Whoever opens the link sees your print requirements, not a bare file list.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fvxkdc9djr80s7g24lkrv.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fvxkdc9djr80s7g24lkrv.png" alt="Task detail: attachments and print notes" width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 2: generate the share link
&lt;/h3&gt;

&lt;p&gt;Open &lt;strong&gt;Share to Public&lt;/strong&gt; in the task menu. Two settings:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Expiry&lt;/strong&gt;: 1 hour / 1 day / 7 days / never&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Allow attachment downloads&lt;/strong&gt;: off by default; check it when the point is the files&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Click Generate Link and you get three things at once: the full link, a 6-letter pickup code, and a pre-composed message (Copy Share Message) that carries both, ready to paste into any chat.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fswtxx8sfqrsc5uk78zdd.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fswtxx8sfqrsc5uk78zdd.png" alt="Share dialog: link, pickup code, expiry" width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Step 3: pick up on any device
&lt;/h3&gt;

&lt;p&gt;Three ways in, for three situations:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;You have the link&lt;/strong&gt;: open it. It's a read-only page with the title, body, notes, and attachments, each with a download button.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;You have only the code&lt;/strong&gt;: open the share entry page and type the 6 letters. Handy when the code was read out loud, or on a public computer where you'd rather not type a long URL.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;You're standing next to the person&lt;/strong&gt;: the share page shows a QR code; scanning it lands on the same page.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fx0yhasu68f10igmo4388.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fx0yhasu68f10igmo4388.png" alt="Public share page: QR code, expiry countdown, downloads" width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fupu9j20gcrkx2eqpdvnh.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fupu9j20gcrkx2eqpdvnh.png" alt="Pickup code entry page" width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  The security details
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;The token in the link is 32 random bytes (64 hex characters), so it can't be enumerated&lt;/li&gt;
&lt;li&gt;Links expire on schedule. The dialog lists each link's remaining time, and any link can be revoked at any moment&lt;/li&gt;
&lt;li&gt;Attachment downloads go through presigned URLs generated per click and valid for 15 minutes; the file in storage is never public&lt;/li&gt;
&lt;li&gt;The shared page is read-only and carries only the title, body, notes, and attachments. No account info, and comments stay private&lt;/li&gt;
&lt;li&gt;Self-hosted instances can switch public sharing off entirely in system config&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  A parcel locker for documents too
&lt;/h3&gt;

&lt;p&gt;The task body and notes are Markdown, written in the same editor UnDercontrol uses everywhere else (notes, finance records). So the thing you share doesn't have to be a file. Meeting minutes, a PC-build checklist, a travel plan — the other side opens a formatted read-only page. Files and documents travel through the same link.&lt;/p&gt;

&lt;h3&gt;
  
  
  Typical scenarios
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Print shop&lt;/strong&gt;: attach the PDFs, type the pickup code on the shop computer, download and print, revoke the link when you get home&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sending files to someone without an account&lt;/strong&gt;: paste the Copy Share Message; one message carries the link and the code&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A shared meeting-room computer&lt;/strong&gt;: type the 6-letter code to open the document read-only; whoever wants a copy afterward scans the QR on the page&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deliverables for an outside collaborator&lt;/strong&gt;: pick the 7-day expiry and the link shuts off by itself&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Output from terminal AI agents&lt;/strong&gt; (Claude Code, Codex, OpenCode, or any terminal-based agent): the agent attaches its report and artifacts to a task; you review, then forward one link&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The entry point is Share to Public in the ⋮ menu on the task detail page. Self-hosting? Check that the public share switch in system config is on — it is by default.&lt;/p&gt;

</description>
      <category>productivity</category>
      <category>devtools</category>
      <category>webdev</category>
      <category>tooling</category>
    </item>
    <item>
      <title>Claude Plans, Codex Executes</title>
      <dc:creator>Oatnil</dc:creator>
      <pubDate>Sun, 19 Jul 2026 09:26:00 +0000</pubDate>
      <link>https://dev.to/oatnil/claude-plans-codex-executes-3mm1</link>
      <guid>https://dev.to/oatnil/claude-plans-codex-executes-3mm1</guid>
      <description>&lt;p&gt;Plenty of developers already work this way: hash out the approach in Claude Code, then copy the conclusion into Codex to implement. The reason is practical, too — you're paying for both subscriptions anyway. Claude is great at planning; letting Codex do the heavy lifting doesn't burn your Claude quota. Both subscriptions earn their keep. The only problem: the "copy-paste" role in the middle is &lt;em&gt;you&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;In UnDercontrol, this pipeline works out of the box&lt;/strong&gt;: Claude plans, Codex executes, and the task system is the whiteboard they share — you only step in to review at the key points.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F4gvbbcja2hg8uawfkxaa.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F4gvbbcja2hg8uawfkxaa.png" alt="Claude plans, Codex executes — tasks are the shared whiteboard between your AI agents" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Five steps to a working pipeline
&lt;/h3&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Configure two agents.&lt;/strong&gt; Claude Code, Codex, and OpenCode are built-in Agent CLI templates — just pick one (any custom command works too). Give the planner Claude Code and the executor Codex. Don't feel like configuring? Even this step can be outsourced — there's a built-in &lt;strong&gt;Agent Creator&lt;/strong&gt;: @mention it, describe the combo you want, and it sets the agents up for you.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Toss the requirement to the planner.&lt;/strong&gt; A rough task is fine — just @planner it. The daemon spins up a Claude Code session on your dev machine: it reads the description, clarifies the requirement, and splits it into subtasks with acceptance criteria.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The planner delegates to the executor.&lt;/strong&gt; It @executor-mentions each subtask — an agent-to-agent @mention automatically launches a new workspace session for the other agent, and the delegation chain preserves your access scope.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Codex executes.&lt;/strong&gt; Each session picks up one subtask: writes code, runs tests, commits, notes progress back on the task, and marks it &lt;code&gt;pending&lt;/code&gt; when done.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;You step in at exactly two points.&lt;/strong&gt; Approve the plan; review the output.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzwufhogqm86me9bbfqo8.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzwufhogqm86me9bbfqo8.png" alt="From a rough idea to reviewed commits — one pipeline" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The whole collaboration happens on the task itself — the description is the spec, comments are the conversation, notes are the progress log. Here's a real task detail: the planner's acceptance criteria on the left, the three-way comment thread on the right:&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fesg7jfgmiqbg51nukhsm.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fesg7jfgmiqbg51nukhsm.png" alt="Task detail — the planner writes the spec, the executor reports commits, everything on the record" width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Form a squad: Agent Teams
&lt;/h3&gt;

&lt;p&gt;Don't want to conduct two agents by hand every time? Organize them into a &lt;strong&gt;Team&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;A Team has one &lt;strong&gt;lead&lt;/strong&gt; and any number of &lt;strong&gt;members&lt;/strong&gt;, and each member carries a delegation hint — "what kind of work routes here." When the lead's session starts, its roster is injected into the prompt automatically — &lt;strong&gt;the lead is born knowing who reports to it, what each specialist is good at, and how to delegate&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmb81iz7kahyirmj9jzki.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fmb81iz7kahyirmj9jzki.png" alt="Agent Teams — a lead plus members with delegation hints, roster auto-injected into the lead's session" width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;"Claude plans, Codex executes" becomes a one-liner: @mention the dev-team's lead and it arranges the rest. Teams also nest — a member can itself lead another Team, so hierarchy grows naturally while each agent only ever needs to know its direct reports.&lt;/p&gt;

&lt;h3&gt;
  
  
  Watch everything live
&lt;/h3&gt;

&lt;p&gt;Open a board for the big picture: how much got planned, what's executing, what's waiting for review — one glance. Every agent session's terminal output also streams back to the web in real time — on your laptop, or your phone.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F58juox1ul0482cnrenfs.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F58juox1ul0482cnrenfs.png" alt="The AI team delivery board — Planning / Executing / Pending Review / Done" width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Typical scenarios
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Queue work before bed, review after breakfast.&lt;/strong&gt; Drop three feature requests on the planner at night; wake up to split subtasks and several &lt;code&gt;pending&lt;/code&gt; implementations, each with its own commits.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Slice a big refactor.&lt;/strong&gt; Claude writes the migration plan and cuts it into 20 small tasks; Codex executes them one by one — roll back exactly the step that goes wrong.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A bug-triage pipeline.&lt;/strong&gt; The planner reproduces, localizes, and writes up the root cause before delegating the fix — the executor gets "what to change, why, and how to verify," not "there's a bug, fix it."&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  The combo is yours to choose
&lt;/h3&gt;

&lt;p&gt;Claude + Codex is just one pairing. UnDercontrol is not tied to any AI tool — Claude Code, Codex, OpenCode, or any terminal-based agent can be a lead or a member. Roles are defined by prompts and skills; tools by Agent CLI configuration. Whatever subscriptions you hold and whatever each model is best at, the orchestration is in your hands.&lt;/p&gt;

&lt;p&gt;One more thing: the progress agents write back lives in the same Markdown system as everything you write by hand — tasks, notes, comments, even finance records share one editing experience.&lt;/p&gt;




&lt;p&gt;Want to try it? &lt;a href="https://oatnil.com/docs/download" rel="noopener noreferrer"&gt;Download UnDercontrol&lt;/a&gt;, start the daemon, create a Team, and @mention the lead on your first task.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>productivity</category>
      <category>devtools</category>
      <category>agents</category>
    </item>
  </channel>
</rss>
