Work That Happens While You Sleep

Why only six

Autonomy is scoped, not assumed

Fred's default posture everywhere else is "you approve, Fred posts": a draft reply waits in your inbox tab, a report waits for a read-through, an invoice waits for a send click. That's not a limitation bolted on top — it's the permission model working as intended. Anything that touches the outside world (sending an email, posting to social media, spending money) sits behind an approval gate until you say otherwise.

The six scheduled presets are different because their scope is narrow and their output is safe by default: they read, summarise, scan, and report. None of them sends money, posts publicly, or deletes anything without a separate step. That's what makes handing them a schedule and walking away a reasonable thing to do, rather than a risk you're quietly accepting.

The six presets

What each one actually does

• Morning Briefing. Your inbox triaged, your calendar summarised, and the news that matters to you condensed — finished and waiting before you sit down.
• Security Scan. The Security agent audits your machine for known vulnerabilities, exposed secrets, risky dependencies, and open ports, and hands you a plain-English list of what needs attention.
• Code Review. Codex Coder runs a pass over your projects looking for issues worth flagging — before you sit down to work, not after you've shipped something.
• AI Newsletter. A draft newsletter assembled on a schedule, ready to read over before it goes anywhere.
• Workspace Backup. Your working files backed up on a schedule, so "I forgot to back up" stops being a sentence you say.
• Research Task. A standing research question — a market, a competitor, a topic you're tracking — re-run on a schedule so the answer stays current without you re-asking it.

Each one is a preset you switch on from the dashboard and give a schedule — daily, weekly, whatever fits — not a feature you have to configure from scratch.

The rest of the week

What "unattended" doesn't mean

Turning on a preset doesn't turn off the permission system underneath it. Prompt-injection protection with active blocking still runs on everything Fred reads, exec approval gates still apply to anything that executes on your machine, and the allowlist for sensitive actions still governs what any agent is allowed to touch — scheduled or not. A Security Scan can tell you a port is open; it doesn't close it without you saying so. A Research Task can hand you a report; it doesn't email it to anyone on its own. The six presets are unattended in the sense that no one has to press "go" — they're not unattended in the sense of unsupervised.

Everywhere outside these six, the ordinary rhythm holds: you ask Fred to chase an invoice, draft a reply, or post an update, and a draft waits for your approval before anything leaves the machine.

Where the autonomy stops

The honest limits

• Only these six run fully unattended. Every other job Fred does — email, social posts, invoices, most research — stops at a draft until you approve it. That's deliberate, not a missing feature.
• Scheduled runs need the machine on and online. Fred's thinking happens at your chosen AI provider over the internet; a Morning Briefing can't run while your computer is asleep or offline.
• Reports use the memory Fred has. Recall of earlier conversations can occasionally be imperfect, so a Research Task re-run weeks apart is a fresh look, not guaranteed continuity with what came before.
• Usage still costs what it costs. A daily Morning Briefing and a nightly Security Scan are real AI calls on your own key — light schedules land at a few dollars a month, heavier ones add up, and the dashboard shows the real number.

Pricing

All six presets, included

FAQ

Questions people ask

Before you go

Keep reading

More about Fred