AI Features

AI Workflow Generation

Describe what you want to automate in plain language and let the platform generate a complete, validated workflow graph for you.

How It Works

The AI workflow generator takes a natural language prompt, sends it to the configured LLM provider, and normalizes the response into a valid workflow schema. The generated workflow is validated against the platform's graph schema before being saved.

Natural Language Input

Type a description like 'Send me a daily email with the top HN stories'.

Schema Normalization

The AI output is normalized into steps, edges, and metadata that match the builder schema.

Graph Validation

The generated graph is validated for duplicate step IDs, missing edges, and invalid node types before import.

Generating a Workflow

Use the Generate AI button in the workflow builder or call the API directly.

POST /api/workflows/generate-ai

{
  "prompt": "Create a workflow that searches GitHub for my repos, summarizes the top 3, and saves the summary to a file."
}

Generated Workflow Example

The normalized output is a standard workflow object with steps and edges.

generated-workflow.json

{
  "name": "GitHub Repo Summarizer",
  "metadata": {
    "steps": [
      { "stepId": "s1", "type": "http", "method": "GET", "url": "https://api.github.com/user/repos" },
      { "stepId": "s2", "type": "llm", "prompt": "Summarize these repos: {{steps.s1.output}}" },
      { "stepId": "s3", "type": "file", "action": "write", "path": "./repos.txt", "content": "{{steps.s2.output}}" }
    ],
    "edges": [
      { "id": "e1", "source": "s1", "target": "s2" },
      { "id": "e2", "source": "s2", "target": "s3" }
    ]
  }
}

Limitations

• Generation is rate-limited to 10 requests per minute.
• Complex multi-branch graphs may require manual adjustment in the builder.
• Always review generated workflows before running them in production.