◆ Concept — figures are illustrative projections of the architecture, not live measurements
Find the 6 that matter in 5,000,000.
Deploy a fleet of agent-finders across a read-only index of millions of builders. A token-disciplined cascade narrows the haystack; we surface only the mutual, intent-matched few — and contact only the ones who opted in.
847 agents searching · 4.2s elapsed
● Live index — real & growing
—builders fingerprinted from contributors to the top AI repos · ranked by impact · read-only public code. This number is real and grows with every index run — the cascade below is the projected architecture at full scale.
The cascade · a real query → the few
loading a real query…
—
Index · real fingerprinted builders
▼ cheap L1 filter · 0 tokens
—
L1 filter · cheap, deterministic
▼ taste-weighted mesh rank
—
Matches · ranked by mesh
▼ mutual "seeking" fit · projected
needs directory
Bilateral intent · projected — needs the opt-in directory
▼ opt-in + provable authorization · projected
projected
Opt-in contact · projected — via AgentMail/Resend
Agent fleet
0deployed · 847 active · MCP + CLI
4.2s
Speed · time to matches
91%
Accuracy · precision
1,180
Efficiency · tokens/match
0.9M
Tokens · of 5M budget
Reliability over all · the judge can't collapse
94%
Groundingcited evidence verified in the code (anti-hallucination)
0.97
Reliability · swap-test consistency
7%
Abstain rate · escalates rather than guess
🔒 The bright line: the swarm searches a read-only index of millions; it contacts only opted-in, intent-matched builders, with provable principal authorization. Quality matching at scale — never cold outreach. Technical signals only; one-click opt-out.