Pattern Intelligence

A pattern is a group of posts that look and feel alike. Patterns come from image and video similarity, not tags, so they capture what content actually shows (an aesthetic, a product shot, a format) even when creators tag it differently.

The map is rebuilt weekly and topped up every 48 hours. Pattern reads take no window.

The GET reads below are free. To match a question to the map, run a patterns job (10 credits per call).

Browse the map

GET /intelligence/patterns?limit=25

curl -H "x-api-key: $API_KEY" "$API_BASE/intelligence/patterns?limit=25"

Each pattern has a patternKey (pat_…), label, summary, and post, creator and engagement counts. patternKey stays the same across rebuilds, so you can store it.

To filter, pass both elementType (theme, aesthetic, emotion, product or visual_community) and elementValue:

GET /intelligence/patterns?elementType=aesthetic&elementValue=minimalist

curl -H "x-api-key: $API_KEY" \
  "$API_BASE/intelligence/patterns?elementType=aesthetic&elementValue=minimalist"

One pattern

GET /intelligence/patterns/pat_…

curl -H "x-api-key: $API_KEY" "$API_BASE/intelligence/patterns/pat_…"

Adds:

• recurrentElements[]: the themes, aesthetics, emotions and products its posts share, with how many posts carry each (support).
• lateShare: the share of its posts from the later part of its window. Rising means it’s growing.
• trajectory[]: its size at each previous rebuild.

The posts in a pattern

GET /intelligence/patterns/pat_…/content?limit=25

curl -H "x-api-key: $API_KEY" "$API_BASE/intelligence/patterns/pat_…/content?limit=25"

Returns its posts, most representative first, with title, platform, contentUrl, publishedAt and engagement.

Lookalikes

GET /intelligence/patterns/pat_…/lookalikes?limit=10

curl -H "x-api-key: $API_KEY" "$API_BASE/intelligence/patterns/pat_…/lookalikes?limit=10"

Returns patterns that look alike but are tagged differently: two communities making the same kind of content under different names. Keyword search won’t find these connections.

• centroidCosine: how visually similar the two patterns are.
• tagJaccard: how much their tags overlap. High visual similarity with low tag overlap is the interesting case.
• sharedElements: what they actually have in common.

Match a question

When you start from a question rather than a pattern, submit a patterns job:

POST /intelligence/jobs
{
  "kind": "patterns",
  "input": {
    "query": "cozy home office setups",
    "limit": 8
  }
}

curl -X POST "$API_BASE/intelligence/jobs" \
  -H "x-api-key: $API_KEY" -H "content-type: application/json" \
  -d '{ "kind": "patterns", "input": { "query": "cozy home office setups", "limit": 8 } }'

The result lists the best-matching patterns, each with its elements, representative posts and lookalikes, plus a written summary. See Intelligence Jobs for inputs and polling.