ai-extract-keywords Edge Function — Ai

ai-extract-keywords — Extract top keywords from text using TF-IDF scoring

Extracts the most relevant keywords from any text using TF-IDF inspired scoring with built-in English stopwords — no external API calls.

---

API

POST /api/ai-extract-keywords

Request body

Field	Type	Required	Default	Description
text	string	✅	—	Text to extract keywords from
topN	number	❌	10	Maximum keywords to return
minLength	number	❌	3	Minimum word length to consider
includeScores	boolean	❌	false	Return {keyword, score} objects instead of plain strings
stopwords	string[]	❌	English defaults	Custom stopword list to override built-in list

Success response (200)

{
  "success": true,
  "data": {
    "keywords": ["machine", "learning", "neural", "network", "training"],
    "count": 5
  }
}

With includeScores: true:

{
  "success": true,
  "data": {
    "keywords": [
      { "keyword": "machine", "score": 0.0423 },
      { "keyword": "learning", "score": 0.0381 }
    ],
    "count": 2
  }
}

Error responses

Code	HTTP	When
INVALID_INPUT	400	Missing or empty text
INTERNAL_ERROR	500	Unexpected error

---

Usage

cURL

curl -X POST "$FUNCTION_URL" \
  -H "Content-Type: application/json" \
  -d '{"text": "Machine learning and AI are transforming the software industry.", "topN": 5, "includeScores": true}'

TypeScript / JavaScript (HTTP)

const response = await fetch(FUNCTION_URL, {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ text: articleBody, topN: 10 }),
});
const { data } = await response.json();
console.log(data.keywords); // ["machine", "learning", ...]

Direct import (Node / Bun / Deno)

import { aiExtractKeywords } from '@aerostack/functions/ai-extract-keywords';

const { keywords } = aiExtractKeywords({ text: document, topN: 15, includeScores: true });

---

Use Cases

• Auto-tagging blog posts or knowledge base articles for search indexing.
• Extracting topics from user messages to route them to the correct support category.
• Building a keyword-based content recommendation engine without a full NLP pipeline.

---

Notes

• Scoring is TF-IDF inspired: words that appear frequently but less uniformly score higher.
• Pure numbers are excluded from keyword candidates.
• The built-in stopword list covers ~120 common English function words.
• Pass stopwords: [] to disable all stopword filtering.