Not one model guessing. Specialised agents, one pool of truth.
Each dimension of a match has its own agent — a specialist that reads one kind of signal and contributes it to a shared intelligence pool. They communicate, weigh each other, and converge on who closes. Tap any agent to see what it feeds.
Preference agent
Learns what this client accepts vs rejects — beyond the job description.
feeds: revealed preference
Rate agent
Knows the band this client actually closes at. Gates anyone above the ceiling.
feeds: rate fit · hard gate
Fit agent
Checks both sides — does the candidate accept the client, not just the reverse.
feeds: two-sided fit · hard gate
Timing agent
Weighs candidate urgency against client speed, so nobody cools in the wait.
feeds: timing
Skill agent
Reads raw technical strength from the CV — the one signal an LLM sees alone.
feeds: technical strength
Confidence agent
Separates what's read from what's inferred. Flags what to confirm before submitting.
feeds: signal confidence · meta
Two kinds of dimension
Hard gates kill a match outright — a rate above the client's ceiling, availability that doesn't exist, a domain they always reject — no matter how strong the rest. Weighted factors then rank everyone who clears the gates. An LLM has neither: it averages keywords. Huntchy gates first, then weighs.
The intelligence pool
No single agent decides. Each contributes its signal to the shared pool, where they weigh and check each other — the rate agent can veto the skill agent's favourite; the fit agent can flag a candidate the preference agent loved. The pool converges on the submission that actually closes, and every closed match flows back to make each agent sharper for that client.
/ 01 — read
Each agent reads its own slice of the role, the candidate, and the client history.
/ 02 — converge
Signals meet in the pool. Gates apply, weights — learned per client — rank the rest.
/ 03 — learn
The outcome feeds back. Each agent re-tunes its weight for this client. The next match is sharper.
Honest note: the agents are specialised reasoning roles, not magic — and the weights they carry are learned from your outcomes, not invented. The advance isn't a secret formula; it's a multi-agent model that knows which signal to trust for which client, fed by data an LLM never sees. That's what a better prompt can't copy.
An LLM is one voice reading a CV.
Huntchy is six specialists arguing toward the truth, tuned by every outcome. The pool is the product. The agents are how it thinks.