Web watch console for managing live web monitoring rules

Build me an internal web watch console: a single app my marketing team opens to manage everything we monitor on the live web with Ahrefs Firehose, without anyone needing to learn Lucene query syntax. The people using it are marketers and content leads, not engineers, so every screen has to speak in plain English and the query language stays hidden behind the forms.

Main screen. Down the left, a list of our taps loaded with List Taps, each showing its display name, when it was created, and how many rules it currently runs; selecting one loads its detail with Get Tap. The right side is the working area: a table of every watch rule on the selected tap, loaded with List Rules and refreshed for a single row with Get Rule after an edit. Each row shows the rule's tag, a plain English description of what it watches (for example "pages mentioning 'pricing update' on competitor sites, English, last 7 days"), the number of matches recorded in the last seven days, when that tally was last updated, and row actions to edit, preview, or retire.

Quota meter. Ahrefs Firehose caps an organization at 25 rules total and returns a 422 if you exceed it, so put a visible quota meter at the top of the rules area reading something like "18 of 25 rules used" with a filling bar that turns amber past 20 and red at 25. At the cap, disable the Add Rule button and replace it with a prompt to prune first, with a shortcut that sorts the table by lowest seven day match count so the deadest rules surface at the top. Never let the user discover the limit by getting a save error.

Rule builder. A form with plain fields only: a tag for the rule, keywords or phrases to match, domains to include, domains to exclude, page category, language, and a recency window (last 24 hours or last 7 days). Handlers compile those fields into the Lucene query the API expects, mapping keywords to the default added field or to title where the user picks "page title", domains to domain terms joined with OR and excluded domains with NOT, category to page_category, language to language, and the recency window to recent:24h or recent:7d. Escape Lucene reserved characters in every user supplied value, colons and slashes especially, so a pasted URL or a time string can never produce a broken query. Show the compiled query in a collapsed, read only "advanced" disclosure for the curious, but never require anyone to touch it.

Live preview. Next to the builder, a preview panel with an explicit Preview button that opens one short bounded Stream read, stopping at whichever comes first: about 20 seconds elapsed or 25 events received. It renders the sample matching pages as cards with page title, domain, URL, detected category and language, and the snippet of added or removed text that triggered the match. This is the whole point of the console, so make it feel decisive: a rule that returns three relevant pages is worth saving and a rule that fills the panel in two seconds is a firehose of noise, and the panel should say so with a line like "25 matches in 4 seconds, this rule will be very noisy". Never fire the preview on keystroke or on field change, only on the explicit button, because Stream is limited to 30 connections per minute per tap. Disable the button with a short cooldown after each run.

Writes. Saving a new rule calls Create Rule, saving an edit calls Update Rule, and retiring a noisy rule calls Delete Rule behind a confirmation that names the rule's tag. Tap management lives in the same app: Create Tap to spin up a new tap, Update Tap to rename one, and Revoke Tap to retire one, with the revoke confirmation clearly warning that revoking permanently destroys the tap and every rule attached to it.

Slack sanity check. Every sample match card in the preview panel has a "Send to Slack" action that posts a formatted example into a channel with Slack Bot Send a Message: the page title as a link, the domain, the matched snippet, and a line naming the rule tag and its plain English description so the team knows which watch produced it. Populate the channel picker with Slack Bot List Channels and remember the last channel used as the default.

Persistence in app storage. The stream is Server Sent Events and real time only with no history endpoint, so the app has to keep its own records. Store a per rule tally of observed matches with timestamps so the seven day count column can be computed, incrementing it whenever the console observes stream events during a preview run or an explicit "collect samples" action that opens one bounded read across the whole tap. Show "no samples collected yet" rather than a misleading zero when a rule has never been sampled. Also persist the builder field values for each rule keyed by rule id, so reopening a rule for editing restores the plain form the user originally filled in instead of trying to reverse parse Lucene, and so the plain English description in the table stays accurate.

Platform constraints to respect. Ahrefs Firehose uses a two key model: the credential holds the management key with the fhm_ prefix, which only works on tap operations, while rule and stream operations need a tap token with the fh_ prefix obtained from Create Tap or List Taps. Every rule and stream handler must resolve the tap token server side before it does anything else. Tap tokens must never be rendered in the UI, never returned to the client in any handler response, and never included in the Slack message; the console shows tap names and ids only. Keep tap and rule calls comfortably under 60 requests per minute by loading rules once per tap selection and refreshing single rows after writes rather than refetching the whole table.