Docs/Features/Filters & tagging

Intake filters & tagging rules

Two kinds of rules shape what lands in your Dataset: intake filters decide whether an item is stored at all, and tagging rules label the records that match. This page documents both — every field, every operator, and the rule JSON.

Intake filters vs tagging rules

Intake filters are a hard gate. An item must pass every intake filter to be ingested at all — fail one and it's dropped before it reaches your Dataset or costs you a record. Use them to cut noise: minimum score, minimum length, English only, no NSFW.
Tagging rules are labels. They never block ingestion — items that don't match still ingest. Matching records get the rule's tags, which power the matched view, tag filters in search and GET /v1/records, and tag-based Alerts.

Rule of thumb

If you never want to see it, use an intake filter. If you want to see it sorted, use a tagging rule.

Where you edit them

Both kinds live on the thing they filter: open a Watcher or a Keyword Monitor keyword and you'll find an Intake filters section and a Tagging rules section. Each rule is a visual builder — match ALL or ANY of a list of conditions — with a Preview matches button that runs the rule against recent items before you save. A tagging rule with no conditions matches every record its Watcher ingests (that's how a keyword tags all of its finds).

Adding a new tagging rule to a community Watcher backfills matches from the last 7 days of ingested posts, so a new tag is never empty while you wait.

Fields you can filter on

Field	Type	Meaning
platform	string	Source platform — 'reddit', 'hackernews', …
channel	string	Subreddit name or query the item came from (lowercase)
createdAt	date	When the post was created on the platform
title	string	Post title
body	string	Post body text
author	string	Author username
score	number	Upvotes / points at crawl time
numComments	number	Comment count at crawl time
questionDetected	boolean	True when the post reads like a question ("how", "recommend", "vs", ends with ?)
hasLink	boolean	True when the post links out
linkDomainCategory	string	Linked domain class: 'social', 'dev', 'blog', or 'other'
freshnessBucket	string	Post age: '<1h', '<6h', '<24h', '<72h', or 'older'
lengthBucket	string	Body length: 'empty', 'short' (<280 chars), 'medium' (<1200), or 'long'
language	string	'en', 'other', or 'unknown' (lightweight detection)
nsfw	boolean	Reddit NSFW flag (Reddit items only)
flair	string	Reddit post flair (Reddit items only)
hnTags	string	Hacker News tags, comma-joined: story, ask_hn, show_hn, front_page, poll (HN items only)

Derived fields (questionDetected, freshnessBucket, lengthBucket, language, linkDomainCategory) are computed by Prowlo at ingestion, so you can filter on them without writing your own heuristics.

Operators

Operator	Meaning
eq / neq	Equals / not equals
in / nin	Value is in / not in a list
contains / not_contains	String contains / does not contain
regex_i	Case-insensitive regular expression match
gt / gte / lt / lte	Number comparisons (score, numComments)
after / before	Date comparisons (createdAt)
exists	Field has a value

The rule document (JSON)

Under the visual builder, every rule is a small JSON document: a match tree of conditions composed with all (AND), any (OR), and not, plus an optional tag list on tagging rules. A tagging rule that labels English question posts scoring 20+:

{
  "version": 1,
  "match": {
    "all": [
      { "field": "score", "op": "gte", "value": 20 },
      { "field": "language", "op": "eq", "value": "en" },
      { "field": "questionDetected", "op": "eq", "value": true }
    ]
  },
  "tag": ["hot-question"]
}

An empty all ({ "all": [] }) matches everything — that's the "tag every find" rule. On intake filters the tag list is ignored: intake only decides whether the item is stored.

Recipes

Drop low-effort posts — intake: score gte 5 and lengthBucket neq empty
English only — intake: language eq en
No bots or promo accounts — intake: author nin ["AutoModerator", …]
Only fresh threads — intake: freshnessBucket in ["<1h", "<6h", "<24h"]
Tag buying questions — tagging: questionDetected eq true + tag question
Tag competitor mentions — tagging: body contains "competitorname" (or regex_i for variants) + tag competitor
Only Show HN launches — intake on an HN Watcher: hnTags contains show_hn

Working with tagged records

Tagged records show up everywhere downstream: the matched view in the dashboard and in GET /v1/records (view=matched unlocks the tags filter), Alerts that fire when a record with a given tag lands, and your agent over MCP — the keyword_* tools manage tagging rules conversationally ("add a tag for 'onboarding frustration'").

← Keyword Monitor Alerts & Trends →