AI processors

AI processors run a model over every row. They are metered by usage, so run a few rows first to gauge the cost.

• Query AI sends a prompt to a model, built from your columns with /. It has no web access, so give it everything it needs in the prompt.
• Web Research asks a research question per row and returns a cited answer, the go-to when the answer lives on the internet.
• Run an Agent puts one of your agents to work on each row, arriving with its own instructions, scoped tools, and knowledge. Nobody is there to approve its actions, so any that need approval are skipped unless they’re set to Always allow.
• Summarize Text condenses a long text column, at the length you choose.

​

AI Data Extractor

Turns messy text into clean columns. Define the fields you want, each with a name and a type, and the extractor pulls them out of the text into a single object cell, one field per value. Use a field downstream by typing / and drilling into it, or flatten it out with Extract Field From Object. It can also extract every occurrence as a list, for example each person mentioned in a document. Like Query AI, it works only with the text you give it.

​

Scoring Agent

Scores any text against criteria you write yourself. Add each criterion with a weight, set the score range, and every row gets a score per criterion, the reasoning behind it, and a weighted total. For scoring companies against a rubric your organization has already built in QX, use Score Companies on the Integrations page instead.

​

Draft An Email

Writes a personalized email per row. You supply the fixed text, and anywhere you want AI-written content you add a curly-brace placeholder, such as {write one sentence on why this company is a fit}. The AI fills the placeholders using your columns and context, and can research the recipient on the web unless you switch that off. The result is text in a cell. To place it in a real mailbox as a draft, feed it into Gmail Draft or Outlook Draft, see Integrations.