·How it works
Tell your AI who you're looking for; an LLM makes the match.
Once it hits, you see the details and contact — visible only to the two matched.
Every other marketplace lists you in public — a feed, profiles, a search box strangers scroll. Pairoa has none of that. The moment your AI publishes your need, it goes private — out of every feed and search, where no other user can browse or search their way to it. The only people it's ever shared with are the ones you genuinely match with — never a crowd, never a list. (To find that match, an LLM reads it privately — see our Privacy Policy.) Here's the whole journey, step by step.
A normal marketplace
You post in public → everyone browses → you sift through strangers
Pairoa
You tell your AI → it stays private — always → no other user sees it, only the people you match with do
you
Find me a technical cofounder.
your AI
On it — I'll post that to Pairoa, privately. Quick thing: it matches a lot better if I add what you bring. Want to include that?
you
Sure — product founder, 30 paid alpha users, building in AI infra. Still at my day job, so keep it quiet.
published privately
not in any list · contact sealed
Step 01
You tell your AI what you want.
Say it the way you’d tell a friend, right inside the AI client you already use — Claude, Cursor, Cline, Codex. Your AI passes it to Pairoa for you. If it’s too vague to match well, your AI just asks you a quick follow-up first.
1 / 4
▸ Curious how this actually works under the hood?
Step 01
Your AI calls
publish_needon Pairoa's MCP server with what you seek, what you offer, and a separately-held contact. A vector embedding shortlists candidates by meaning; an LLM judge then reads that shortlist to decide the match. Your raw text stays private, never in a public index.Step 02
No API ever returns a need that isn't matched to you. There is deliberately no feed, search, or browse endpoint — that's the technical red line, not a setting.
Step 03
A two-step funnel: a
pgvector pre-filter shortlists ~20 candidates by meaning, then an LLM judge reads only that shortlist and returns is_match · confidence · why. (We send bounded i_seek / i_offer snippets to a third-party LLM to judge — see our Privacy Policy for exactly what that means.)Step 04
On a match, both sides are notified by a short email carrying only 2–6 safe tags and a link — no original text, no contact, no name, no tracking token. Open the dashboard and both parties unlock each other's need, intro, and contact at once — neither side sees the other first. A single need can match several counterparts over its lifetime; close it and the raw text is dropped, only the matched record kept.