Write a Blog Post

A form captures a title, brief, and tone. An HTTP node fetches research. Two LLM nodes draft and then edit the post.

form blog_request {
  label: "Blog Request"
  enabled: true
  schema: @json {
    {
      "type": "object",
      "required": ["title", "brief"],
      "properties": {
        "title": { "type": "string", "title": "Title" },
        "brief": { "type": "string", "title": "Brief or outline" },
        "tone": { "type": "string", "title": "Tone", "enum": ["professional", "casual", "technical"] }
      },
      "additionalProperties": false
    }
  }
}

graph write_blog {
  label: "Write Blog Post"
  description: "Research, draft, and edit a blog post"

  root {
    type: code
    label: "Normalize"
    inputSchema: @json {
      {
        "type": "object",
        "required": ["title", "brief"],
        "properties": {
          "title": { "type": "string" },
          "brief": { "type": "string" },
          "tone": { "type": "string" }
        },
        "additionalProperties": false
      }
    }
    outputSchema: @json {
      {
        "type": "object",
        "required": ["title", "brief", "tone"],
        "properties": {
          "title": { "type": "string" },
          "brief": { "type": "string" },
          "tone": { "type": "string" }
        },
        "additionalProperties": false
      }
    }
    code: @ts {
      const { title, brief, tone } = context.nodes.root.input
      return {
        title: title.trim(),
        brief: brief.trim(),
        tone: (tone ?? "professional").trim(),
      }
    }
  }

  node research {
    type: http
    label: "Fetch research"
    method: "POST"
    url: @ts {
      return "https://api.tavily.com/search"
    }
    headers: {
      "Content-Type": "application/json"
    }
    body: @ts {
      return JSON.stringify({
        api_key: context.secrets.TAVILY_API_KEY,
        query: context.nodes.root.output.title,
        max_results: 5
      })
    }
    schema: @json {
      { "type": "object", "properties": { "results": { "type": "array" } } }
    }
  }

  node draft {
    type: ai
    label: "Draft post"
    kind: object
    model: "anthropic/claude-3.5-sonnet"
    prompt: @ts {
      const { title, brief, tone } = context.nodes.root.output
      const research = JSON.stringify(context.nodes.research?.output?.results ?? [])
      return `Write a blog post.

Title: ${title}
Brief: ${brief}
Tone: ${tone}
Research context: ${research}

Structure with an introduction, subheadings, and conclusion.
Output in Markdown. Target 800-1200 words.`
    }
    temperature: 0.7
    maxTokens: 4000
    schema: @json {
      {
        "type": "object",
        "required": ["content"],
        "properties": { "content": { "type": "string" } }
      }
    }
  }

  node edit {
    type: ai
    label: "Edit and polish"
    kind: object
    model: "anthropic/claude-3.5-sonnet"
    prompt: @ts {
      const draft = context.nodes.draft.output.content
      const tone = context.nodes.root.output.tone
      return `Edit this blog post for clarity, flow, and grammar.
Tighten verbose sentences. Ensure the tone is ${tone}.

Draft:
${draft}`
    }
    temperature: 0.3
    schema: @json {
      {
        "type": "object",
        "required": ["content"],
        "properties": { "content": { "type": "string" } }
      }
    }
  }

  flow {
    root -> research
    research -> draft
    draft -> edit
  }
}

trigger on_blog_request {
  form:blog_request -> write_blog
  enabled: true
}

How it works

The form collects a title, brief, and optional tone (defaults to professional).
The root node trims whitespace and fills in the default tone.
The research HTTP node queries Tavily and returns up to five results as context.
The draft AI node writes an 800-1200 word Markdown post using the title, brief, tone, and research.
The edit AI node polishes the draft for clarity, flow, and grammar at a lower temperature.
The trigger fires the graph on every form submission.

Run it

swirls login
swirls secret set TAVILY_API_KEY <your-key>
swirls deploy

Open the Portal, navigate to the form, and submit a request. Check the Executions view to inspect each node's output.

Customize

Add a review step between draft and edit to collect human feedback before the editing pass.
Swap Tavily for another search API by updating the research node URL and body.
Add an HTTP node after edit to publish the finished post to your CMS.
Add persistence to build a library of generated posts.
Add a schedule trigger for recurring content generation.

Write a Blog Post

How it works

Run it

Customize

On this page