Set up a scraper in one sentence. It keeps itself running.
Describe the data you want. Everscraper finds the sites, builds a scraper for each one and refreshes your table every morning. When a site changes, it repairs itself. You just read the email with what changed.
No code, no URLs, no selectors. Now in invite-only beta.
You askDental clinics in Texas with email and phone
- texasdentists.exampleMember directory
- clinicfinder.exampleClinic listing
- smilemap.exampleSearch results
{
"strategy": "css",
"items": "li.clinic",
"fields": {
"name": { "selector": "h3" },
"phone": {
"selector": ".tel",
"transforms": ["trim"]
}
}
}- Marlowe DentalPhone changed512-555-0167
- Pinecrest Family Dentistry214-555-0178
- Fennick & Rowe Dental713-555-0119
- Ossington Dental CareNew817-555-0150
Set up like an AI scraper. Runs like a classic one.
Classic scrapers are cheap to run, but they take hours to build and break when a page changes. AI scrapers are easy, but they pay a model on every run. Everscraper uses AI once to build a plain scraper for each site, then runs it every day without it.
- 01
Job postings for data engineers in Barcelona
- title
- company
- location
- posted
- url
Describe the data
One sentence, no URLs. Everscraper proposes the columns and finds the sources, and you confirm both before the build.
- 02
- fetch_page /clinics?page=1
- test_recipe 24 rows
- save_recipe match 98%
We build the scrapers
Everscraper reads each site and writes its scraper, a plain JSON recipe: where the pages are and where each column lives. It is checked against the real pages before it goes live.
- 03
- MonNo changes
- TueNo changes
- Wed3 changed, 1 new
It runs and repairs itself
At 06:00 your table refreshes with no model in the loop, and the differences reach your inbox. If a site changes, its scraper is rebuilt automatically.
The table is the record. The difference is the news.
Every refresh is compared with the one before. A changed value keeps its old one beside it, and rows that appear or disappear are marked as such.
Dental clinics in Texas
Changes since the last refresh: 1 changed, 2 new, 1 removed.
- Marlowe Dental phone: "512-555-0142" → "512-555-0167"
- Ossington Dental Care New
Nothing moved on Tuesday, so the table reads the same as Monday's. Switch to Wednesday to see a morning with news.
| Clinic | Source | City | Phone |
|---|---|---|---|
| Marlowe Dental | texasdentists.example | Austin | 512-555-0142 |
| Pinecrest Family Dentistry | texasdentists.example | Dallas | 214-555-0178 |
| Fennick & Rowe Dental | clinicfinder.example | Houston | 713-555-0119 |
| Juniper Orthodontics | smilemap.example | San Antonio | 210-555-0196 |
| Alder Street Dental | clinicfinder.example | El Paso | 915-555-0133 |
| Clinic | Source | City | Phone |
|---|---|---|---|
| Marlowe DentalPhone changed | texasdentists.example | Austin | |
| Pinecrest Family Dentistry | texasdentists.example | Dallas | 214-555-0178 |
| Fennick & Rowe Dental | clinicfinder.example | Houston | 713-555-0119 |
| Juniper Orthodontics | smilemap.example | San Antonio | 210-555-0196 |
| Ossington Dental CareNew | smilemap.example | Fort Worth | 817-555-0150 |
| Larkspur Kids DentistryNew | clinicfinder.example | Lubbock | 806-555-0127 |
| Alder Street DentalRemoved | clinicfinder.example | El Paso | 915-555-0133 |
It repairs itself, and you can see it working.
You can see how each source is read, how well its recipe matched, whether it is healthy, and what every run cost.
- texasdentists.exampleRecipe, JSON99%Active
- clinicfinder.exampleRecipe, CSS98%Active
- smilemap.exampleRecipe, JSON-LD96%ActiveRepaired once
- dentalpages.exampleRecipe, CSSNo dataBrokenIts last data is kept
Fixes itself when a site changes
If a source returns far fewer rows or many more empty cells than its baseline, its scraper is rebuilt automatically, up to three times.
Keeps your last good data
A source that cannot be repaired is marked broken, keeps its last data and is named in your next email.
Checked before it counts
A recipe is compared with a model reading the same pages. Below a 95% match it is sent back with the objections.
The cheapest route first
Shops that expose a product API are read straight from it, with no agent. Everything else gets a compiled recipe.
If a public page lists it, it's one sentence away.
- Leads
Dental clinics in Texas with email and phone
name email phone city - Prices and stock
Espresso machine prices in UK online shops
name price currency in_stock - Job postings
Data engineer jobs in Barcelona
title company location url - Listings
Two-bedroom flats for rent in Valencia
name price area url
Built once, so it's cheap every day after.
Calling a model on every run costs the same every day. Everscraper uses AI once per source, to build the scraper. After that each refresh is plain requests with no model involved. You can see what every run cost.
Request an invite- A model on every run
- Everscraper
Model callRecipe run, no model
Questions before the first build?
Do I need to know how to code?
No. You describe the data in one sentence and approve the sources and columns. The recipe behind each source is there to read if you are curious.
Do I need to know which sites to scrape?
No. You describe the data and Everscraper looks for sources and suggests a few. You can drop any of them or add a URL of your own before the build starts.
Do I need to maintain anything?
No. Sources are checked every day and repaired automatically when a site changes. If one cannot be repaired, it keeps its last data and the email tells you.
Why is it cheap to run?
Because AI only builds the scraper, once per source. The daily runs are plain requests with no model involved.
What kind of data can it collect?
Any kind of record a public page lists: business contact details for a lead list, product prices and stock, job postings, property listings. One source is a page or listing that gives one or more rows.
What is a recipe?
A JSON document, never code. It says where the pages are, how to move through them and where each column lives. Because it is data, it runs the same way every day and you can read it.
How often is the data refreshed?
Once a day, at 06:00. You can also refresh a dataset yourself whenever you want a fresh read.
What happens when a site changes its layout?
Each source has a baseline. If a refresh returns fewer than half the usual rows, or far more empty cells, the source is compiled again automatically, up to three times. If that fails it is marked broken, its last data stays in the table and the email tells you.
Are there limits?
A refresh reads up to 60 pages and 5,000 rows per source.
How do I get the data out?
Read it in the table or export it as CSV. The change email is on by default and you can turn it off.