Recipe 03, AI content summarizer

Goal: POST an article URL to a webhook. Flero fetches the page, summarises it with an LLM, and emails you the result.

Difficulty: Beginner · Time: ~15 minutes · Connectors: OpenAI or Anthropic, Email

🎬 Gif needed: recipe-03__ai-run.gif, the end-to-end run lighting up: Webhook → Web Scraper → AI Agent → Email, each turning green. ~6 seconds.

Prerequisites

• An LLM provider credential (OpenAI or Anthropic).
• An Email credential or transactional email connector.

Finished workflow

┌─────────────┐    ┌──────────────┐    ┌──────────┐    ┌──────────────┐    ┌───────┐
│  Webhook    │───▶│ Web Scraper  │───▶│ AI Agent │───▶│  Transform   │───▶│ Email │
│  Trigger    │    │              │    │          │    │ (format msg) │    │       │
└─────────────┘    └──────────────┘    └──────────┘    └──────────────┘    └───────┘

Step-by-step

1. Webhook trigger

Replace Start with Webhook Trigger (POST). Expect a body like { "url": "https://..." }.

2. Web Scraper

Add a Web Scraper node:

• URL: {{ $trigger.body.url }}
• Render JavaScript: false (faster; toggle on for SPA pages)
• Selectors:
  • title → h1, attribute text
  • body → article, .post-content, main, attribute text

3. AI Agent

Add an AI Agent node:

• Provider: Anthropic (or OpenAI)
• Credential: your LLM credential
• Model: claude-opus-4-7 (or gpt-4)
• System instructions:

  You are a concise editor. Given an article, produce a 3-sentence summary
  and 3 key takeaways as a bulleted list. Don't make things up.

• Prompt:

  Title: {{ $node["Web Scraper"].json.title }}
  
  Article:
  {{ $node["Web Scraper"].json.body }}

• Max tokens: 500
• Temperature: 0.3 (more deterministic for summarisation)

4. Transform the result into an email body

Add a Transform node:

map:
  subject: "Summary: {{ $node["Web Scraper"].json.title }}"
  body: |
    <h2>{{ $node["Web Scraper"].json.title }}</h2>
    <p><strong>Source:</strong> {{ $trigger.body.url }}</p>
    <p>{{ $node["AI Agent"].json.response }}</p>

5. Send the email

Add an Email node:

• Credential: your email credential
• To: your email address (or {{ $trigger.body.recipient }} for dynamic)
• Subject: {{ $node["Transform"].json.subject }}
• Body, HTML: {{ $node["Transform"].json.body }}

6. Save and activate

Try it

curl -X POST <your webhook URL> \
  -H "Content-Type: application/json" \
  -d '{"url": "https://en.wikipedia.org/wiki/Workflow"}'

Within ~15 seconds you should receive an email with a tidy summary.

Variations

Multiple URLs at once

Change the trigger payload to { "urls": ["url1", "url2", ...] }. Add a Loop over urls; the rest of the workflow runs once per URL.

Different output formats

Modify the system instructions:

• "Output as JSON with fields summary and keyTakeaways" → use responseFormat: json_object on the AI Agent.
• "Output in French" → just say so.

Add guardrails

Insert an AI Guardrails node after AI Agent to filter PII or unsafe content before the email is sent.

Cost-cap with Token Cost Guard

Add a Token Cost Guard node at the start with a budget of 5 000 tokens per run. Runaway prompts can't bankrupt you.

Cache repeat requests

Add a Cache Storage node keyed on {{ $trigger.body.url }}. If the URL was summarised in the last 24 hours, skip the AI call and reuse the cached summary.

Troubleshooting

Next

• AI & LLM nodes
• Recipe 06, RAG document Q&A, answering questions over many documents

Found something out of date? This page lives in the Flero docs content set.