Prelaunch · early access open
Anything online.
In a table.
Posts, prices, people, papers, rates, listings. If a person can find it on the internet, describe it in a sentence. An AI agent checks the sources and tells you if it is possible and what it costs. You say yes. You get the table.
The idea
Computers can do anything again. Most people have not heard.
Developers have always known that with code, nothing on the web is out of reach. It just took the right person with the right tools and a free week. Agentic AI removed the week. Now an agent can write and run the collector while you watch. That realization has landed for technical people. It has not landed for everyone else.
Before
Possible, but only for some.
You needed a developer who knew the right libraries, proxies, and retries, and had the hours. Everyone else copied and pasted, or gave up.
Now
An agent does the developer’s week in minutes.
Agentic AI reads the sources, plans the collector, writes it, runs it, and recovers when a page misbehaves. The knowledge that anything is possible is now a thing a computer has.
The catch
You still have to know what to ask for.
If you do not know how the web works, you cannot picture what can be automated. That is the window we are here to open: describe the outcome, and we tell you what is possible.
Things people ask for
- “Every post about home espresso on Reddit and X this month, with the accounts behind them.”
- “The rainfall tables from 14 water authorities, one row per region per year, with the source.”
- “The posted mortgage rate from 30 lenders, every Monday.”
- “Every roaster in Oregon with a website and a public email.”
- “The products on these two stores, and what changed since yesterday.”
- “Public creator profiles about a topic, with follower counts and public contact details.”
Common shapes
Anything is a lot. Here are the shapes we see most.
Every tile is a kind of request people bring us. If yours is not here, describe it anyway. The status shows what is built first, what is planned, and what comes later.
Social and community search
Find the posts, people, and creators about a topic across Reddit, X, Instagram, LinkedIn, and forums. In one table, not forty tabs.
Research and obscure data
Pull scattered public data out of PDFs, government pages, registries, and archives into one clean table.
Financial and market data
Rates, prices, filings, and listings from many sites, in one dated table you can refresh on a schedule.
Website to spreadsheet
Any set of pages into rows and the columns you choose. The general-purpose version of everything else here.
Product data collection
Names, links, prices, currencies, availability, and any other field a store shows, from as many stores as you need.
Competitor price monitoring
Check a defined set of products every morning and see only what changed.
Lead lists and contact data
Companies or people matching your profile, with publicly listed contact details and the page each one came from.
Something else entirely
Describe it. We reply with possible or not, and roughly what it would cost. That is the whole point.
Describe your ask →How it works
Five steps. The agent does the hard three.
You describe and approve. The agent checks, builds, and collects. Code enforces the limit you set.
- 1
Say it
Describe what you want from the internet, like you would to a colleague. Paste any URLs you have.
- 2
We check it
An agent looks at the sources and tells you: possible or not, which fields exist, and what it costs.
- 3
You okay it
Pick the columns. Set a spending limit. Nothing runs until you say yes.
- 4
We get it
The agent builds and runs the collector. Code enforces your limit. Empty means the source did not say.
- 5
Keep it fresh
Download the table, or schedule it and read only what changed.
Columns and cost
You choose the columns. The price follows.
Renaming a column is free. Adding one that needs a visit to every page is not. The agent tells you which is which before you approve. Try it.
Estimate
13requests
- Rows, capped
- 300
- Listing pages
- 13
- Product-page visits
- 0
- Derived columns
- free
Illustrative units. The real estimate also covers model usage, compute, and a bounded retry allowance, and shows a price after terms are set. Nothing runs until you approve a limit.
Who asks
People who know the data is out there.
Different jobs, same realization: the collecting was never the interesting part.
Marketers
Stop searching four platforms by hand. Ask for the table.
Analysts and researchers
The data exists. It is just in fourteen places.
Finance and actuarial teams
Every public number you track, in one dated table.
Ecommerce
Know what competitors sell, charge, and have in stock.
Agencies
Do the research for the tenth client without redoing the work.
Outbound marketing
A lead list you can defend, with a source for every contact.
Keep it fresh
Ask once. Get the answer every week.
Save the collector. It reruns at your frequency, inside per-run and monthly limits, and shows you what changed instead of the whole table again.
Controls
- Frequency and timezone
- Per-run and monthly spending limits
- Fresh snapshot, append, or update by key
- Pause, resume, edit, delete
- Notification on completion or failure
Monthly projection
Before you activate a schedule you see an estimated cost per run and a monthly projection based on frequency. The projection states its volume assumptions.
| lendertext | rate_last_weeknumber | rate_todaynumber | changenumber · derived |
|---|---|---|---|
| Harbor Federal | 6.25 | 6.125 | -0.125 |
| Pearl Credit Union | 6.125 | 5.99 | -0.135 |
| Alder Bank | 6.25 | 6.25 | 0 |
Exports
Where the table goes.
CSV ships first. Google Docs reports and Airtable are planned. Google Sheets is proposed. They are different things and we do not swap one for another.
CSV
Structured records with your column names preserved.
Google Docs
A formatted report containing the requested tables and explanatory content.
Airtable
Records in a base and table you select.
Google Sheets
A spreadsheet dataset in a sheet and tab you select.
Pricing approach
First we say if it is possible. Then what it costs. Then you decide.
Exact prices are not set. The structure is.
One-time ask
A bounded collection
Scope, columns, and a row cap you approve. One estimate, one spending limit, one table.
Subscription
Saved collectors on a schedule
Hosted scheduling, change history, exports, and a defined usage allowance or usage-based billing.
Additional usage
Only with your approval
Work beyond your limit needs a new yes. Limits are enforced in code, not by asking the agent nicely.
Straight answers
How we say yes, and how we say no.
Yes, usually
If a browser can reach it, so can we.
Public pages, PDFs, directories, listings, posts, profiles, registries, archives. An agent collects it, normalizes it, and hands you the table with a source on every row.
No, sometimes
You hear it before you pay.
Sources that block automated access. Fields a site simply does not show. When that is the answer, you hear it at the estimate step, with what we can get instead. We never quietly swap your request for an easier one, and we never fill a gap with a guess.
Coming later
Pages you can only see when signed in.
The agent will browse them the way you do, in a real browser, using your own account and only where the site’s terms allow it. A higher tier, because a real browser costs more to run.
Questions
Common questions.
Can you really get anything?
If a person can find it on the public internet, we can almost always collect it. Every request gets a straight answer first: possible or not, and what it would cost. The rare no comes from sources that block automated access, and you hear it before you pay a cent. Pages you can only see when signed in are coming later, on a higher tier: the agent browses them in a real browser with your own account, where the site’s terms allow it.
How is this different from a scraping tool?
A tool asks you to know which tool, which selector, which proxy, and which retry strategy. We ask you what you want. An AI agent writes and runs the collector, the way a developer would if you had one on call. Agentic AI is what made this possible. Our job is to make it something you can hire in a sentence.
How do I know what it will cost?
Before anything runs you see an estimate with its assumptions: sources, row count or cap, and which columns add work. You set a spending limit. Application code enforces that limit during the run.
What happens when a source does not have a field?
The cell stays empty. We never fill a gap with a guess. Derived fields, such as a price change, are labeled and we show how they were calculated.
Do I need to understand code, APIs, or scraping?
No. You describe the data. An agent plans the collector, we run it inside limits you approve, and application code enforces the spending cap. If you are technical, you can later export the workflow as runnable code.
Can I keep the data current?
Yes. Save the workflow, choose a frequency and timezone, set per-run and monthly limits, and read only what changed between runs.
Where does the table go?
CSV export is in the first release. Google Docs reports and Airtable are planned. Google Sheets is a proposed addition.
When can I use it?
The app is not built yet. We are taking early-access requests to decide which collectors we build first. Describe what you want from the internet and a person will reply with possible or not, and roughly what it would cost.
Early access
Tell us what you want from the internet.
The app is not built yet. Early-access requests decide which collectors we build first. Describe your ask and a person replies with two things: possible or not, and roughly what it would cost.