How search works
One sentence in, a ranked shortlist out. Between those two points I plan the search, read public sources live, cross-reference what they say, and attach the reasoning to every result. You watch each step as it runs.
1. You ask
Plain English is the whole query language. “Staff backend engineers in NYC who shipped Rust in production” is a complete query - no boolean operators, no filter panels. Follow-ups refine the same search: narrow by industry, broaden the geography, ask for similar profiles.
2. I plan
I break the ask into concrete sub-searches and decide which sources can answer each one: GitHub for shipped code, public LinkedIn and Crunchbase records for roles and companies, the open web for everything else.
3. I read sources live
Sources are read at search time, not from a stale database export. That means a person’s current role, a company’s latest round, a repository’s recent activity - as of now. The steps stream into the conversation while the search runs, so you always see what I’m reading and why.
4. I rank and explain
Candidates from different sources are cross-referenced and deduplicated, then ranked against what you actually asked for. Every result ships with the reasoning behind it - the public evidence that made it a match. No black-box scores.
5. You act
From the shortlist you can ask me to draft outreach in each person’s language, keep refining, or publish the search so others can see the full result set.
What a search costs
Each search costs credits in proportion to the work it takes: quick lookups cost a credit or two, deep multi-step research with web browsing costs more. Credits are reserved up front, so a search never overdrafts your balance. Full mechanics in Credits and pricing.