--- title: "Monitors" id: "1956" type: "page" slug: "monitors" published_at: "2026-09-30T22:50:44+00:00" modified_at: "2026-10-01T00:28:30+00:00" url: "https://xedant.com/agents/news/docs/monitors" markdown_url: "https://xedant.com/agents/news/docs/monitors.md" excerpt: "Collection answers the question “what new has appeared”. A monitor answers a different question: “what…" --- # Monitors [https://xedant.com/agents/news/docs/monitors.md](https://xedant.com/agents/news/docs/monitors.md) Collection answers the question “what new has appeared”. A monitor answers a different question: “what has changed in the thing I am watching”. The practical difference is huge: a pricing page can be rewritten every week, but when you care about one line — the base plan’s price — you want to hear only about it. ## What a monitor is A monitor is an assignment: “watch this value on this page and tell me on that kind of change”. It is created in words and lives in a settings file: *“Watch this product’s price and availability. Warn me when the price drops more than 5% and when the product runs out.”* The agent then decides by itself which fields to take and which threshold to set, and you look at the result on the [Monitors](/agents/news/docs/views) screen. ## The watch region Before comparing, it must be decided where to look. There are three regions: - **The page as a whole** — all meaningful parts of the page are compared. Good for small pages, and when you do not yet know what exactly will change. - **List blocks** — comparison inside a block where records follow one another: products, listings, table rows. - **Individual values** — the most exact: you name a specific element of the page, and from then on only it is compared. ## Fields and their values A field is what you named and watch: price, availability, title, date, average score. There can be as many fields as you like — one or several at once. For every field the agent keeps a **history of values** and shows it as a chart: you see not only “it got cheaper” but also how the price behaved over the last months. ## Values by your own code Sometimes the needed value cannot be taken the usual way: it is substituted onto the page later, or sits in an unusual place. Then your own Python code extracts the value: - **Your own value lives like an ordinary one** — it has the same history, the same chart and the same thresholds as the built-in ones. You can both add a new value and replace a built-in one. - **The page can be cleaned before comparing** — remove random inserts and banners that cause false alarms. - **A code failure is an honest error** — the check is marked as errored and the values are not saved: better a skip than a wrong number. Details — in the [Custom Scripts](/agents/news/docs/scripts) section. ## Comparison modes and the threshold The threshold answers the question “which changes count as significant”. Without it you get an alarm on every penny: prices on sites move with currency recomputation and small layout edits. - **Any change** — always report. Fits product availability: it is either there or not. - **A percentage threshold** — “tell me when it gets more than 5% cheaper”. This is the one needed most often. - **An absolute amount** — “tell me when the price moves by more than 500 dollars”. Alongside there is the **baseline** — the starting value, the reference point. It can be pinned by hand (“we count from today’s price”) or cleared when comparing should start over. ## “Before and after” snapshots For every trigger, two snapshots are kept: what the value was and what it became. Old snapshots are deleted by retention — so the archive does not grow forever. The snapshots are the evidence you later check against: “was it really 12,900?” ## Schedule and health A monitor has a schedule: how often to check. Twice a day is enough for a price; for a hot product’s availability — once an hour. The source’s state is visible right on the card: “active”, “a streak of failures”, “error”, “paused”, “waiting for the first check”. When a site stops answering, you will not be guessing why it is silent. ## Managing a monitor There are no editing forms in the interface — the monitor card carries command buttons. Each prefills a ready phrase into the chat with your agent, and when the chat is not connected — copies a ready command to send by hand: - **Check now** — an out-of-schedule check, without waiting for the plan. - **Change the schedule** — check more often or less often. - **Pause** — the product was taken off sale; there is nothing more to watch. - **Archive** — the monitor is no longer needed, but the value history stays. ## Peculiarities - **Whole-page comparison** works without fields too, but yields noisier events: any block reshuffle looks like a change. - **Visual comparison by screenshots** needs an external service and is off by default. Until it is connected, only text and value comparison honestly works. Next → [Events & Evidence](/agents/news/docs/events) [← Back to the documentation index](/agents/news/docs)