Always-On Agent Kit

Every big AI lab now has an agent that works while you sleep. Most of them don't work in Europe yet. This kit shows you where each one stands, and how I built my own morning agent with Claude Code from Switzerland.

#1. The 5 agents and Europe

Status as of 25 Oct 2026. These products change every few weeks, so check the official pages before you pay for a plan. The Dots region detail comes from news coverage, OpenAI's own page only says "eligible markets".

#2. They are all the same 5 parts

1. A model that plans and decides.
2. Its own computer: a cloud machine with a browser, so it keeps running when your laptop is closed.
3. Your apps: connectors to mail, calendar, docs, GitHub.
4. A schedule: it wakes up on its own, for example every morning at 07:30.
5. An ask-first rule: what it may do alone, what needs your OK, what it may never do.

That is a harness around a model. You can build a small version yourself.

#3. Build your own morning agent with Claude Code (Mac)

What you get: every morning at 07:30 your Mac runs Claude Code in the background, reads what you allow it to read, and leaves you a short briefing. It can't send, pay or delete anything, because those tools are simply not allowed.

#Step 1: Install Claude Code and log in

npm install -g @anthropic-ai/claude-code
claude

Log in once in the interactive session, then quit.

#Step 2: Connect the apps you want it to read

Add the MCP servers for the tools you use, for example GitHub, your calendar or your notes app. Each one is a single command:

claude mcp add <name> -- <command that starts the server>
claude mcp list

Start with one or two apps. You can always add more later.

#Step 3: Write the briefing prompt

Create a folder ~/agent and a file ~/agent/brief.md:

You are my morning agent. Read only, never change anything.

1. My calendar today: meetings, gaps, anything that moved.
2. Open pull requests and issues assigned to me.
3. Anything that needs a decision from me today.

Write at most 10 short lines. Start with the most urgent thing.
If you are unsure about something, say so instead of guessing.

#Step 4: Run it once by hand

cd ~/agent
claude -p "$(cat brief.md)" \
  --allowedTools "Read" "mcp__github" \
  --permission-mode dontAsk \
  > "briefing-$(date +%F).md"

• -p runs Claude Code once, prints the answer and exits.
• --allowedTools is your allow list. Replace mcp__github with the servers you added in step 2.
• --permission-mode dontAsk refuses everything that is not on the allow list, instead of asking. That is your ask-first rule: nothing outside the list can happen while you sleep.

#Step 5: Put it on a schedule with launchd

launchd is the Mac scheduler. Unlike cron, it also runs a missed job after your Mac wakes up.

Save this as ~/Library/LaunchAgents/ai.sangeeths.morning-agent.plist (replace YOURNAME and check the path of claude with which claude):

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
  <key>Label</key><string>ai.sangeeths.morning-agent</string>
  <key>ProgramArguments</key>
  <array>
    <string>/bin/zsh</string><string>-lc</string>
    <string>cd ~/agent &amp;&amp; claude -p "$(cat brief.md)" --allowedTools Read mcp__github --permission-mode dontAsk &gt; briefing-$(date +%F).md &amp;&amp; osascript -e 'display notification "Your briefing is ready" with title "Morning agent"'</string>
  </array>
  <key>StartCalendarInterval</key>
  <dict><key>Hour</key><integer>7</integer><key>Minute</key><integer>30</integer></dict>
  <key>StandardErrorPath</key><string>/Users/YOURNAME/agent/agent.log</string>
</dict>
</plist>

Load it:

launchctl load ~/Library/LaunchAgents/ai.sangeeths.morning-agent.plist
launchctl start ai.sangeeths.morning-agent

The second command runs it right away, so you can test it. Every morning you now get a notification and a new briefing-<date>.md in ~/agent.

#Step 6: Grow it slowly

• Add one app at a time to the allow list.
• Keep sending, paying and deleting out of the list. If you want drafts, let it write them into a file and send them yourself.
• Read agent.log when something looks off.

#4. My 3 questions before I trust any agent

1. Where does my data live? On my Mac, in the vendor's cloud, or in a VM I don't control?
2. What can it do without asking? Read only, draft, or send and pay?
3. Can I stop it in one tap? Pause button, revoked logins, a clear log of what it did.

#Credits and sources

Launch details from the official announcements and coverage by TechCrunch, BetaNews, TechRepublic, Notebookcheck and others, collected on 25 Oct 2026.

Made by @sangeeths.ai: real AI engineering from Switzerland. Share this kit: sangeeths.ai/agent · All free kits: sangeeths.ai