Recipes vs scraping: how is Wield different?
Scraping usually means a program downloads pages and parses them, often from a server. A Wield recipe does something narrower. It makes the few requests the site's own page makes, from a tab in your browser, with the session you already have, at human pace. No model runs during the recipe. The result returns to your AI app, and the run stops at fixed limits.
Comparison table
| Factor | Typical scraper | Wield recipe |
|---|---|---|
| Where it runs | Often a server or a script | Your browser, on your computer |
| Whose session | None, unless the scraper adds one | Yours. Cookies never leave your computer |
| Requests | As many as the job needs | The few requests the site's own page makes |
| Pace | Set by the scraper | Human pace, paced per site |
| Model | Often none | None during the run |
| Limits | Set by the scraper | 12 steps, 40 requests, 5 MB per response, 60 s per run |
| What it is | A program | A small signed JSON file |
Pace and limits
Every recipe run is bounded. A run stops after 12 steps, 40 requests, 5 MB per response or 60 seconds. Pacing is set per site, so a run cannot send a burst of requests to one site. Any write, such as a booking or a message, stops before its first write request and shows you every request for an Allow. Deny is the default.
Where it runs and whose session
A recipe runs in a tab of the site, inside your own browser. The request carries your normal session. The results go from the extension to wield-mcp and then to your AI app. They never pass through wield.run. Logs stay on your computer. They record the method, host, path, status and timing, but never the bodies or query values.
What a recipe is (and isn't)
A recipe is a public JSON file that describes the requests and steps for one task on one site. It is signed with Ed25519, and the extension refuses any recipe that is unsigned or changed. Public recipes are reviewed against each site's terms.
A recipe is not a crawler and not a dataset. It runs one task that you start, such as a train search or a slot lookup. It is not built to collect data in bulk, and it does not run in a loop to watch a site. The benchmark ran 3 fixed queries per task, not bulk collection.
FAQ
Is a Wield recipe scraping?
It reads data from a site, so it is a form of data access. The difference is in how it works. A recipe acts as you, in your browser, for one task you start, with fixed limits and no model in the loop.
Can a recipe crawl a whole site?
No. Recipes run one task at a time. A run stops after 12 steps, 40 requests, 5 MB per response or 60 seconds. Recipes are not designed for crawling or bulk collection.
Who checks the recipes?
Public recipes are reviewed for each site's terms. The registry signs every index and recipe with Ed25519. The extension and wield-mcp refuse unsigned or changed recipes, and no JavaScript from the registry runs in the page.
Do I need to write code to use a recipe?
No. Your AI app calls the recipe as a tool. You ask for a result in plain words, and the agent chooses the recipe. You can see how recipes are written if you want to read one.