Inference infrastructure · Solana

Can your app ask a question and get an answer before the frame renders?

Inferqon is a single inference endpoint. Hand it a token mint; it returns a computed intelligence payload — a risk score and the signals behind it — fused from audit, holder and liquidity sources. The feature isn't the data. It's the latency, and we measure every call in front of you.

Last inference latency

—ms end-to-end

p50 — · p95 — · across 0 calls this session

p50 latency—ms

p95 latency—ms

throughput · 1 conn—req/s

Try:

—score / 100

awaiting call

Submit a mint to compute an intelligence verdict.

inference.payload compute — ms node: edge-use1

// POST /v1/infer — payload renders here

Latency SLAs, not data SLAs

Everyone has the data. We sell the time between the request and the answer. Pick the budget your render loop can afford.

BATCH async

<800 ms p95

Bulk mint scoring
Webhook delivery
Fair-use throughput

EDGE default

<120 ms p95

Single-call inference
Regional POP routing
Warm model cache

REALTIME colocated

<35 ms p95

In-region socket
Pre-resolved sources
Dedicated inference lane

What the endpoint actually computes

The signals

Audit and holder structure come from Jupiter's assets/search index — mint & freeze authority state, top-holder concentration, bundler supply and an organic-flow score. Depth comes from DexScreener's deepest pair.

The score

A weighted composite anchored on organic score, then penalised for holder concentration and bundler supply, rewarded for revoked authorities, verification and real liquidity. One number in [0,100] with a verdict band.

The clock

Every source call is wrapped in performance.now(). The HUD shows real end-to-end latency, a rolling p50/p95 over your session, and single-connection throughput — no synthetic numbers.