About Findish
Curious how it’s made? Read how Findish is built, the technical version with diagrams.
What it is
I cook from a handful of sites I trust, and searching them one at a time got old. Findish looks through all of them at once, by dish, by what’s in the fridge or by who made it, and sends you to the creator’s own page for the recipe.
Not sure what to search for? Browse by cuisine, course or ingredient.
How to use it
Type a dish, an ingredient or a creator, or pick one from the suggestions to make it a tag in the box. Plain phrases work too. Filler like “easy” or “how to make” is ignored, and likely typos are fixed (it says so, and you can search exactly what you typed instead). “Under 30 minutes” or “quick” sets the time, “with tofu and eggs” adds those ingredients, “no pork” leaves pork out, and “from Maangchi” narrows to one creator: each becomes a tag you can remove.
Pick a cuisine from the suggestions (or type “thai food”) to see that cuisine’s recipes, with a link to explore that cuisine on World cuisine.
After a search, the words under the box narrow it down: what to show, the course (mains, sides, soups, desserts and so on), the cuisine, the creator and the time. Under a main dish, “Thai sides to go with it” (or whichever cuisine it is) finds sides from the same kitchen. The course comes from how each site files its recipes, or from a plainly named dish (“soup”, “cookies”), so a recipe no site labels only shows up when no course is picked. Diet (vegetarian or vegan) and the order (best match, quickest first or newest first) are there too. “Narrow it” under the count names the choices that would do something for this search, and “Share this search” gives you its link. In the footer, “New this week” lists what the latest weekly check found.
“Save” under a recipe keeps it for later, and “Saved” in the footer shows what you kept. Click in the empty search box for your recent searches and the recipes you opened last. All of it is kept only on your device, and each one can be removed. Press / to jump to the search box.
How it works
It’s a real search engine, just a small one: the same three parts as any other, built for one subject and a hand-picked list of cooks instead of the whole web.
1. A weekly program collects recipes
Where it runs: GitHub Actions, every Monday.
Once a week, a small program visits each site. It reads the site’s robots.txt and follows its rules, reads the sitemap (the site’s own list of its pages), and fetches only pages that are new or changed, one at a time, a few seconds apart. From each recipe it keeps the name, link, photo link, ingredients, time and cuisine. Never the method or the writing around it: every result sends you to the creator’s page. Videos come from YouTube’s own API, without Shorts. Last updated November 16, 2026.
2. The site rebuilds with a fresh index
Where it runs: Cloudflare, after each weekly run.
The new results start a rebuild of the site, which packs them into one small file, the index, with every ingredient matched to a shared list of ingredient names, so “scallions” and “green onions” are the same thing and “pork” covers pork belly.
There’s no database. The pages and the index are plain files.
3. Your browser runs the search
Where it runs: Your phone or computer.
The index is small enough (a few hundred kilobytes) to download once, so the search runs right in your browser, with no search server. Your words are matched against recipe names, ingredients, cuisines and creators, with matches in the name counting most. With ingredients added, recipes that use the most of them and need the fewest other things come first. Recipes people open more often get a small boost (once a day, the site counts how often each recipe was opened in the last 90 days: counts only, nothing about who), and when two recipes are about as good a match, creators take turns so one site doesn’t fill the page.
Where the data comes from
6,031 recipes and 5,700 videos from 15 creators. Their recipes are free because they put in the work. If one becomes a favorite, back them: every creator has a page with their ways, from memberships to cookbooks. Book links go to Bookshop.org, which pays the author and an independent bookstore.
Serious Eats and Red House Spice are offered as a link to their own search instead, since their owners don’t allow automated collection.
Good to know
“Vegetarian” and “vegan” leave out recipes with meat, seafood or things like fish sauce in their name or their ingredients (and for vegan, dairy, eggs and honey). “No dairy”, “no nuts”, “no eggs”, “no shellfish” and “no pork” work the same way. A recipe with no ingredient list is left out of all of them. A recipe that leaves the choice to you (“chicken or vegetable stock”) stays in, and a video is judged by its name alone, so “Katsudon” and “Smash Burgers” are out. That’s a best guess from names and ingredient lists, not an allergy check, so read the recipe if it matters.
New recipes show up once a week, after the weekly run. A recipe a creator takes down leaves the search on the next run.
Ask your AI assistant
Claude and other AI assistants can search this same index for you: by dish, by what you have, or by creator. Like the page, it only finds recipes. It answers with names and links, never the recipe itself, so you still read it on the creator’s site. It’s free, with no account, and keeps nothing about what you ask.
Opens Claude with the name and address filled in. Nothing is added until you say so there.
For other assistants, things to ask and what it can look up, see Findish for AI assistants.
For site owners
The weekly program calls itself FindishBot. It follows robots.txt, reads one page at a time a few seconds apart, and only fetches pages that are new or changed. To keep it out, addUser-agent: FindishBot and Disallow: / to your robots.txt, and your recipes leave the search on the next weekly run. Until November 16, 2026 it was called RecipeSearchBot, and rules written for that name still count.
Newsletters are read only through their public RSS feed, once a week, the way a feed reader does. The posts themselves are never fetched, nothing behind a paywall is read, and only posts that hold a recipe are listed.
Want your recipes taken out? Ask through kylepace.net and they will be gone within a week, no reason needed. The same link works for questions, or if you'd like to be added.
Privacy
What you type never leaves your device, and neither do your recent searches or saved recipes. Photos load from each creator’s site when they come on screen. Anonymous counts, like how often a recipe gets opened, go to the site’s analytics; see the privacy page.
One thing does ask this site’s server: the Seattle Grocery Deals app, when you tap for recipes that use several things on your shopping list. It sends only the ingredients it matched to that week’s deals, never what you typed, and the server answers and keeps nothing about it.
Get in touch
Questions, a correction or a site to suggest: email mail@kylepace.net. The code is public: github.com/kyle-pace/recipe-search.