@activeset/schema-gen

Generate Schema.org JSON-LD recommendations for a Webflow site's static pages using a local Ollama model (Gemma by default). Output is a portable JSON file you can upload via the ActiveSet dashboard's Import schema analyses button on the Webflow Pages tab.

Everything runs on your machine — your Webflow content and the model output never leave it until you upload the file.

Prerequisites

1. Node.js ≥ 20
2. Ollama running locally with a Gemma model:

   ollama serve
   ollama pull gemma4:e4b   # default — ~5 GB
   # or, smaller/faster:
   ollama pull gemma3:4b

3. A Webflow API token with read access to the target site.

Usage

# Discover what'll be processed
npx @activeset/schema-gen list \
  --site <webflowSiteId> \
  --token <webflowApiToken>

# Generate recommendations for every published static page
npx @activeset/schema-gen generate \
  --site <webflowSiteId> \
  --token <webflowApiToken> \
  --domain www.example.com \
  --out schema-output.json

Then in the dashboard: Webflow Pages → Import schema analyses → pick schema-output.json. Every analyzed page now shows a "Cached" badge in the Schema panel.

Auth via env

Place these in .env.local next to where you run the command and you can drop the flags:

WEBFLOW_API_TOKEN=...
OLLAMA_BASE_URL=http://127.0.0.1:11434   # optional
OLLAMA_MODEL=gemma4:e4b                  # optional

Useful flags

| Flag | Default | Notes | |---|---|---| | --only home,about | — | Restrict to specific slugs or publishedPaths | | --model gemma3:4b | gemma4:e4b | Smaller and noticeably faster | | --ollama-host <url> | http://127.0.0.1:11434 | Override if Ollama is on a different port | | --concurrency 2 | 1 | Pages analyzed in parallel — keep low; Ollama is single-GPU | | --out path/to/file.json | schema-output.json | Output path |

What's in the output file

{
  "version": 1,
  "generatedAt": "2026-04-15T...",
  "model": "gemma4:e4b",
  "siteId": "...",
  "domain": "www.example.com",
  "entries": [
    {
      "pageId": "...",
      "pageTitle": "Home",
      "url": "https://www.example.com/",
      "contentHash": "...",
      "result": {
        "pageType": "webpage",
        "summary": "...",
        "existing": [...],
        "recommended": [
          {
            "type": "Organization",
            "reason": "...",
            "confidence": "high",
            "jsonLd": { "@context": "https://schema.org", "@type": "Organization", ... }
          }
        ]
      }
    }
  ]
}

The pageId + contentHash pair is the cache key the dashboard uses, so re-running on unchanged pages is instant (the importer skips matching docs).

Why a CLI and not in-app?

Browser → local Ollama hits CORS restrictions, and a deployed server can't reach localhost on your machine. Running the analysis on the same machine that runs Ollama avoids both problems entirely. The dashboard becomes a viewer.

License

MIT