The agent: what it does, what it doesn't

The agent is a civic-specialized assistant that answers questions about the data graph in plain English. You can find it on the homepage, on most catalog pages, and on most hub pages — and wherever it lives, it's the same agent, scoped to wherever you are.

What it does:

- Answers questions across the eight federal data sources with citations on every load-bearing claim - Reasons across cross-source connections (donor → candidate → bill → contractor → award) - Surfaces relevant context (committee proximity, donor industry, related trades) - Drafts factual delta alerts when a saved investigation's answer changes - Drafts custom briefings on your watched entities

What it doesn't do:

- Give legal, financial, or political advice. Ask "should I vote for X?" and you'll get a factual deflection. - Make claims about private individuals not in the data - Editorialize or interpret beyond what the data supports - Predict the future — past patterns are described; future inference is qualified - Surface Reporter content by default (only when a Reporter opts in, clearly attributed)

If you ever feel like the agent answered something it shouldn't have, email hello@receipts.report. The agent is a major product surface and we take that seriously.