AI research agents

Hire a researcher for every row

Ask a question in plain language. An agent searches the web, reads the sources, and writes a cited answer into the cell — once per row, across your whole list.

Bring your own keys · Works with 300+ models · No credit card

Questions a database can’t answer

Static data vendors sell you yesterday’s snapshot of standard fields. But the questions that actually qualify a deal — What’s their pricing model? Are they hiring SDRs? Did they just change CTOs? Do they mention SOC 2 on their site? — aren’t in any database. Someone has to go look.

In Liquid Rows, that someone is an agent. You write the question once as a column; the agent researches it for every row, live, and shows its work.

Built to survive real batches

Test small, run big

  1. Write the question as an agent column — reference other columns with variables, like “What pricing model does {{Company}} use?”
  2. Test it on one row straight from the editor and read the trace to tune your prompt.
  3. Pilot on the first ten rows to sanity-check quality and cost.
  4. Run the sheet with a spend limit set — and push results to your CRM as rows finish.

Any model, your keys

Agents run on the model you pick per column: OpenAI, Anthropic, Grok, or Groq keys, one OpenRouter key for 300+ models, or local models you host yourself. Tool-capable models are labeled in the picker, and you pay providers directly at cost — no token markup.

Frequently asked questions

What is an AI research agent?

An autonomous worker that answers a question by searching the web, opening results, reading pages, and composing an answer. In Liquid Rows, an agent column runs one agent per row, so each account gets researched individually.

How do I trust the answers?

Every answer records the source URLs it used, visible as citations on the cell, with the full step-by-step trace one click away. When an agent can’t find an answer, the cell says “not found” instead of guessing.

What happens when a search engine fails mid-run?

Searches fall through an engine chain automatically — if one engine rejects a query or runs out of credits, the next takes over and the batch keeps running.

Which AI models power the agents?

Your choice: OpenAI, Anthropic, Grok, or Groq keys; OpenRouter for 300+ models with one key; or local models you host. Tool-capable models are labeled in the picker.

How do I control what an agent run costs?

Pilot runs execute on just the first N rows, per-sheet spend limits stop bulk runs at your ceiling, and a per-cell cost ledger shows where the budget went.

Related

Stop researching accounts one tab at a time

Write the question once. Let agents do the other 9,999.

Run your first agent →