Contrie

The brief

An AI agent that pulls data off the web gets back a block of text and no way to tell whether it is right. Every vendor in the category sells extraction volume — pages fetched, requests served — and leaves the agent to trust the answer. Contrie was built on the opposite premise: the useful product is not the extracted value, it is the evidence that the value is correct.

The build

A hosted data service agents can call directly, plus a tool server they can connect to without writing an integration. Every extraction comes back with a quality score, per-field grounding that points at the exact source text each value came from, a trace of how the run reached it, and what it cost. Keys, rate limits, usage metering, and self-serve billing are part of the product rather than an afterthought — an agent can sign up, get a key, and start spending without a human in the loop.

Outcome

Live at contrie.com and open for early access while the first accounts are onboarded. The evidence payload — score, grounding, trace, cost — ships on every response, which means a caller can decide whether to trust a field instead of taking the whole answer on faith.