Account & data

Local AI

Have a language model on your own machine write up a crawl or help with Competitor Intelligence, with nothing uploaded and the exact brief shown before it is sent.

Why local

CrawlStat's promise is that crawl data never leaves your machine. A report written by a cloud model would break that promise in the most literal way possible: it would post your client's site structure, page titles and problems to someone else's computer.

So the model runs on your machine, through Ollama. The default endpoint is http://127.0.0.1:11434. That's a loopback address, which means the request cannot leave the host.

Setting it up

  1. Install Ollama from ollama.com.
  2. Pull a model: ollama pull qwen3:8b
  3. In CrawlStat, open Settings → Local AI and press Test. CrawlStat lists the models you have installed; pick one.

Then open a crawl's Plan tab and click Write the report.

For Competitor Intelligence

Competitor Intelligence can use Ollama too, in two ways. Both are optional.

  • Matching by meaning. A small embedding model reads what each page is about, so pages that say the same thing in other words are matched up. Pull it with ollama pull all-minilm.
  • Naming topics and drafting briefs. A local chat model, like the one above, gives each topic a short name and helps draft content briefs.

Both run on your machine, like everything else here.

Using what it writes

A written report produced by a local model: a headline, a fix this first recommendation, a numbered plan, quick wins and what can wait
What a local model writes from your crawl: a headline, the one thing to fix first, a plan, quick wins and what can wait. Copy it, save it or send it to a client.

The written report stays with the crawl until you quit CrawlStat. From the Plan tab you can copy it, save it as Markdown, or click Export PDF to put it in a client report, where it becomes the summary page straight after the cover. The Export report dialog also offers Include the written report whenever one exists for the crawl.

Nothing here is required

The ranked plan is computed in full without any model involved. The only thing this adds is prose on top of numbers that already exist. With Ollama absent, you lose a paragraph, not a feature. Competitor Intelligence works without it too. And CrawlStat never needs a network connection to crawl or to audit.

What is actually sent

A compact brief: the ranked jobs, their counts and scores, and one example URL each. Not page bodies, not the crawl database.

The What is sent button shows you that brief, in full, before you send it. A privacy claim nobody can check is worth very little.

If you point it somewhere else

The endpoint is editable, because some people run Ollama on a machine under their desk. If you set it to anything that is not loopback, CrawlStat says so plainly in Settings. At that moment crawl summaries do leave your computer, and you should know it.