It runs on Monday. You read one paragraph.

Give it a list of pages and a morning. It checks them while you are asleep, ignores the build ids and the rotating banners, and tells you the prices that moved — or that nothing did.

Set up a watchFree to try. Nothing to install.

How it works

Three steps, and none of them is “configure”.

01
Give it a list and a morning
A markdown table of pages and what you care about on each. Then say when — “every Monday at nine” — and it books the schedule itself and tells you what unattended means before it does.
02
It runs with nobody there
Each run fetches every page, reduces it to comparable text, and compares it against the snapshot from last time. The first run is a baseline and reports nothing, which is the honest answer when there is nothing to compare against.
03
You get what moved, and only that
A dated report in reports/, written to be read on a phone by someone who has forgotten they set this up. Old reports are never deleted, so the history is the point.

What it can do

It knows the difference between a change and churn
A deploy hash, a cache-busting query string, a testimonial that rotates on every load — these differ on two fetches ten seconds apart. It says once that a page is like that and then stops reporting it, because a feed that cries wolf every week is a feed nobody opens.
A quiet week gets one line
If nothing moved, the report says nothing moved. It will not pad a page with rows saying “no change” to look like work happened — the short report is the honest form of a quiet week.
It tells you when your own list is wrong
A row with no address, the same page listed twice, a homepage that does not carry the prices you asked it to watch. It reports these by row and keeps going, rather than failing the run or quietly editing your file.

How it is set up

The mechanics, so you know what you are getting before you sign in.

Environment: A sandbox with a network, reached through Mume’s egress proxy — every request is resolved and logged. Scheduled runs get a short-lived credential scoped to one schedule, not a copy of your session.
Tools: bash for fetching and reducing pages, readFile, writeFile, editFile, glob, grep for the workspace, getCurrentTime so a weekly report knows which week, and scheduleRun, listSchedules and cancelSchedule.
Model: Any chat model in the catalogue. The judgement here is editorial — deciding which of forty differences is worth a sentence.
Starting files: Seeded with a watchlist and no snapshots, so the first run is a real baseline run. The list carries five planted problems: a duplicate row, a row with no URL, a page that changes on every fetch, a vague brief and a request it has no tool for.
What persists: Snapshots and every report stay with the session, which is what makes week two mean anything. Nothing is deleted between runs.

Things people ask it

“Check everything on the watchlist and set up a weekly run on Monday mornings.”
“What changed since last week?”
“What is wrong with my watchlist?”
“Stop the weekly run.”

What it will not do

Every one of these is a real constraint we have hit, not a roadmap item.

It cannot send you anything. There is no email or notification tool, so the report is a file in the workspace — it says so on the first run rather than letting you wait for a message that is not coming.
It reads what the server sends. A page that renders entirely in the browser arrives nearly empty, and it will tell you that rather than reporting an empty page as unchanged forever.
It watches pages, not products. A price that changed only inside a logged-in dashboard, or in a PDF behind a form, is not something it can see.
Scheduled runs cannot ask you anything. Nobody is there to answer, so they act on the instruction as written — which is why it asks you to approve the instruction before it books it.

Which model does this best

Measured over real runs of this agent, per model. A run counts as finished when it produced the answer on its own — nothing failed, nobody was asked to approve anything, and it did not run out of steps. Read all three columns together: a model that finishes fast by giving up scores badly on the first, and one that completes everything by grinding scores badly on the second.

Model	Completion rate	Median cost	Median time	Runs
claude-sonnet-5anthropic	93%	$0.066	43s	15

Completion rate: Runs that finished the job without you having to step in.
Median cost: What a typical run costs, in credits.
Median time: Wall clock, from the first message to the answer.

Medians over the runs behind each row. A model appears once it has 3 runs on this agent, and the run count is shown so you can judge how much a figure rests on.