Setting up the loom agent

The right setup order

• +1. Connect your X account (for publishing)
• +2. Train "my voice" — loom will write in your voice
• +3. Add at least 3–5 sources: 2–3 high-signal X accounts in your field + 1–2 RSS/news sources (a single source produces single-type drafts; a mixed set brings variety)
• +4. Choose your mode — start with "approval"; turn on autonomous as trust grows
• +5. Set the daily draft limit and, if you want, a daily credit budget
• +6. Connect Telegram — so drafts land in your pocket with their content (optional)

Task center: assign tasks, lean back

On the Tasks page you see every standing instruction you've given loom on one screen: which accounts are watched, how many drafts each task produced in the last 7 days, how many you approved, when it last ran. You can paste a list of @accounts into the box and assign dozens of watch tasks at once.

Three modes, gradual trust

• +manual — loom only generates when you ask
• +approval — every draft comes to your approval first; publish, edit, or reject with one click
• +autonomous — loom schedules 75+ scoring drafts to your best time and publishes them itself

Even if you turn on autonomous, it doesn't start right away: up to the first 10 approved drafts loom behaves like approval mode (trust ramp), its authority expands gradually, and quotes always go to approval.

Safety and the brake

• +dual safety gate: sensitive topics and unverified claims aren't posted automatically — they go to your approval; the same happens if the check system is unreachable (the if-unsure-don't-post principle)
• +pause: with one click, pause for 1 hour or until the end of the day — scanning, generation, autonomous posting, and the posts loom scheduled all stop; the ones YOU scheduled manually are unaffected
• +decision log: from which source, with which memory records, with which score, and with which decision each draft was generated is recorded step by step

How loom learns you

Every approval, rejection, and edit is written to memory; if you state a reason with one click when rejecting (off-topic · wrong tone · cliché...), the learning sharpens. Patterns you reject repeatedly go on the "avoid" list, your real X performance shapes generation, and every Monday morning you get a weekly letter from loom: what I learned, where I'm weak, what I suggest.