A 0.0089-Dollar Lead and 6,400 GitHub Stars: Jev Now Screens People for Your Recruiting Agent, 85% Cheaper Than Clay

TypeSafe AI, the company behind the Jev decision model, has found a new job for its classifier: deciding who your agent should call. Together with Treg, an agent tooling platform, it open-sourced a people-search pipeline that has pulled in 6,400 GitHub stars since launch. The pitch is blunt - a verified work-email lookup starts at $0.0089, and if the tool comes back empty, you do not pay.

The division of labor is the interesting part. An agent such as Claude Code or Codex takes a natural-language brief - say, marketing leads at Series B AI startups in San Francisco - and decomposes it into region, funding stage and role. Treg then calls the underlying data tools, including customers' own Hunter or LinkedIn-sourcing accounts, while Jev sits in the middle as a gate, judging each retrieved candidate against the brief and discarding the near-misses before the agent pays for enrichment.

Treg's own benchmark, People Search Bench, spans 119 real tasks across recruiting, B2B prospecting, deterministic lookups and creator searches. On the B2B category, Claude Code alone scored 43% and jumped to 78.2% with Treg attached; the other three categories came in at 80%, 76.3% and 62.9%. These are company-reported numbers on a company-built benchmark, so treat them as a marketing artifact until third parties replicate them.

The cost comparison is more concrete, if still self-run. Against the same 292-person list, Treg says it found emails for 289 people with 264 exact matches to publicly known addresses, for a total of $1.49. Clay found 280 with 262 exact matches for $10.34. Per correctly matched email that is $0.0056 versus $0.0395 - the 85% saving from the headline, earned mostly by refusing to pay for enrichment on candidates Jev rejects.

Early users have noted the seams. One complained that searches for their own name and colleagues surfaced little of value while ChatGPT and Gemini found plenty at a much higher price; Treg publicly asked the user to send the test set. That is the honest trade-off of a small classifier plus cheap data calls: it is dramatically cheaper on bulk lists and noticeably weaker on edge cases.

For TypeSafe, the release is another distribution channel for a model that has moved fast since September - a trillion tokens in three days, a claimed one-third of the Fortune 500 using it, and roughly $870 million raised at a $7.5 billion valuation 24 days after launch. Slotting Jev into recruiting, sales and research workflows is a bid to make the decision model a default dependency, one $0.0089 lookup at a time.