Upload a list. Get every row researched, with its sources.
Enrichment data is either stale or made by hand
Bought databases go out of date and rarely hold the one field you need. Doing it by hand takes a researcher, a browser and a week. Most AI tools answer one question at a time, and few say where an answer came from.
Four steps, then Run
- The table: upload a CSV or paste one. Its first line names the columns.
- The question: write it once, with columns as {{column}} — “What does {{Company}} charge for its cheapest plan?” — and pick the column holding each row’s website.
- The answer’s columns: the fields each row should gain, or free text.
- The budget: the most a row may spend, and how many rows run at once. Check the table, then Run.
Your table, with new columns
- Answers as columns: the table as you sent it, each answer value in a column of its own, filling in as rows finish.
- Sources you can check: the page a value was read on, wherever the agent read one, and a link to the row’s own run with every step it took.
- Ready for a spreadsheet: download as CSV or JSON lines. A cell that would start a spreadsheet formula is written as text.
- Nothing silently skipped: a row that cannot run is marked with the reason, and a row that could no longer be afforded says so — neither is charged.
It never spends more than you said
- A cap on every row: each row pays only for what it used, up to the cap; the table never spends more than rows times the cap.
- Each row on its own site: a row that starts at a company’s website reads that site, and no other, unless you allow more.
- A few at a time: three rows at once by default, up to ten, and never more than your plan runs at once.
- Stop at any time: Cancel stops the rows not yet finished; the answers so far are kept.
Who uses it
- Sales and RevOps: enrich a list of accounts with headquarters, leadership, pricing and what each company sells.
- Recruiting: open roles, teams and locations from each company’s careers pages.
- Procurement: supplier terms, certifications and contact pages, one supplier per row.
- Product and pricing: each competitor’s cheapest plan and what it includes, in one table.
Frequently asked questions
What is AI data enrichment?
Adding the facts a list is missing — a company’s headquarters, its pricing, its leadership — by having an AI agent research each row on the web. Here, each row is its own agent run, and its values come back linked to the pages they were read on.
How much does it cost?
Each row pays for the pages it read and the model’s tokens, up to the cap you set per row, so the most a table can cost is rows times that cap. Checking the table first is free and gives about what it will cost, from your last tables’ rows or, before you have any, the cap.
How many rows can a table have?
Up to 25 rows on the free plan, 200 on Starter, 500 on Growth and 1,000 on Scale.
What happens to a row the agent cannot answer?
A row that cannot run — an empty prompt, a cell that is not a website — is marked invalid with the reason and never charged. A row that reaches its cap stops there, and the table carries on with the next.
Can I run it from code?
Yes: POST /agent/batch with the table as CSV or rows, and POST /agent/batch/preview to check it and estimate the cost first. A webhook can tell you when the last row has finished.
Your list, researched by morning
25 rows to try it free. See the estimate before anything runs.