product · AI research agent

An AI research agent that shows where its answers come from

Ask a question the way you would ask an analyst. The agent searches the web, reads the pages it needs and answers — as a paragraph or as the exact table you asked for — linking its values to the pages they were read on. You set the budget, and it never spends past it.

No card. 1,000 credits to start. Blocked pages, 404s and unchanged pages are never charged.

2,000
credits: a run’s cap unless you set another. It stops there, with what it found.
0
credits for a page a site refused.
+20%
on the model provider’s token price — or 1 credit per 1,000 tokens on your own model.
01the problem

Research is the job nobody has time for

Comparing five vendors’ plans, checking which countries changed a rule, finding who runs a company: each is an afternoon of tabs, copying and pasting, and a spreadsheet nobody can check afterwards. A chat assistant is quicker, but you cannot tell which number came from where.

02how it works

Ask, shape, watch, follow up

  • Ask: write what you want found. Name the sites to start from, or let it search the web.
  • Shape the answer: describe the columns you want and the answer comes back as data in that shape, ready for a spreadsheet or your code.
  • Watch it work: every search, page and model step appears as it happens, with what it cost.
  • Follow up: ask the next question in the same conversation. Pages it has already read are read again free.
03what comes back

Answers you can check

  • Sources on every answer: the pages it read, numbered — and for an answer in your format, the page each value was read on, wherever it read one.
  • Your format: a paragraph, or data matching the fields you described.
  • Pages other tools miss: when the data is behind “Load more”, a tab or a filter, it opens the page in a real browser and clicks. It never types, so it fills in no form and logs in nowhere.
  • What it found, even at the limit: a run that reaches its cap returns what it had, and Continue carries on with a new budget, keeping the pages already paid for.
04in your control

Spend and scope are yours to set

  • A budget per run: 2,000 credits unless you set another; the run stops there.
  • The sites it may read: start it from a website and the app keeps it on that site unless you allow more; through the API you name the sites it may read.
  • Cancel at any time: it stops before its next step, and what it used stays on the ledger.
  • Your own model if you like: run it on your workspace’s model connection (any OpenAI-compatible endpoint with tool calls) and pay 1 credit per 1,000 tokens.
05what teams ask it

Questions it answers well

  • Competitor plans and prices: every plan with its price per user and what it adds over the one below.
  • Account research: a company’s headquarters, founding year, founders and chief executive, each from a page that states it.
  • Rules across countries: a rate or a requirement for every country on a list, in one table.
  • Vendor comparisons: free plans, limits and what each one leaves out, with the pricing page cited for each.
  • Catalogues and datasets: products with their prices and ratings, or a ranked list built from a page’s table.
06and then

From one answer to a feed

One answer is useful. The same question every morning is a feed. Keep and re-run puts a run on a schedule and tells you only what changed. With proposals on, the agent can suggest what is worth watching — nothing is created or charged until someone on your team approves it.

questions

Frequently asked questions

What can the AI research agent do?

It searches the web, reads the site maps and pages it needs, and answers your question — as text, or as data in the fields you describe — with the pages it used as sources. It can open pages that need a browser and click through tabs, filters and “Load more” to reach the data.

How much does a run cost?

The pages it reads, at what each took to read (a refused page is free), and the model’s tokens at the provider’s price plus 20%. Every run has a cap — 2,000 credits unless you set another — and stops there with what it found.

How do I know the answer is right?

Check it. A structured answer links its values to the pages they were read on, and the run’s trace lists every search and page it read. The agent treats what pages say as data, never as instructions.

Does it log in to sites or fill in forms?

No. It reads what a visitor can see. It may click, scroll, select and wait to reveal content, but it never types, so it fills in no form and logs in nowhere.

Can I use it from code?

Yes, through the HTTP API: POST /agent, with follow-ups, continuing a run and a webhook for its progress. The Python and Node SDKs and the MCP server do not offer the agent yet.

Hand it the question you keep putting off

1,000 free credits, no card. Set a budget and watch it work.