Reputation & escrow for agentic infrastructure

A credit score
for AI services.

Every MCP server and API endpoint gets scored by NVIDIA Nemotron 3 Ultra before money moves. Good endpoints clear the bar and touch Stripe. Bad ones don't — and that's verifiable, not just a UI promise.

Powered by NVIDIA Nemotron 3 Ultra · Stripe PaymentIntents · EigenLayer AVS · Nous Research

beacon — zsh
$
GEThttps://api.beacon.hq/v1/score/{endpoint}
200 OK48ms
{
"endpoint":"api.weather-ai.com/v1/forecast",
"trust_score":94,
"grade":"A",
"accuracy":0.97,
"uptime_pct":99.91,
"latency_p99_ms":180,
"dispute_rate":0.001,
"scam_flag":false,
"evaluator":"nemotron-3-ultra",
"escrow_recommendation":"PASS",
"attestation":"0x7f3a9c21e8b4d610",
"attested_at":"2026-06-30T08:14:02.119Z",
}
nemotron-3-ultra · EigenLayer AVS · slashing-enabledverified ✓
100M+Agentic transactions on rails
14,000+MCP servers scored and climbing
$3–5TAgentic commerce projected by 2030
30%SaaS vendors shipping MCP by 2026
The missing layer

The rails are done.
Trust isn't.

Identity tells you who an agent is. Payments tell you how money moves. Neither tells you whetherthe service is any good — or won't rip you off.

Solved

Identity

AP2 / FIDO

Cryptographic agent identities. Every autonomous agent can prove who it is with non-spoofable credentials.

Solved

Payments

x402 / Stripe

Programmable money movement. Agents can pay APIs autonomously over HTTP, with on-chain settlement and receipts.

The gap

Reputation

Beacon

Is the endpoint actually good? Beacon scores every MCP server and API with NVIDIA Nemotron 3 Ultra, then enforces that score with a Stripe escrow gate. Bad actors never reach Stripe at all.

  • Shadow MCP servers
  • Prompt-injection payment attacks
  • Unvetted, anonymous endpoints
How it works

Score, hold, attest —
in that order.

01

Probe

When an agent wants to pay a new API, Beacon hits the endpoint live. NVIDIA Nemotron 3 Ultra — served via Nous Research's inference API — judges the response for accuracy, error rate, and hallucination. That feeds directly into a 0–100 trust score. If the live call fails, a deterministic mock evaluator falls back, but the evaluator field always tells you which brain scored it.

NVIDIA Nemotron 3 Ultra · Nous Research
02

Escrow

Endpoints scoring ≥ 70 pass through to a real Stripe PaymentIntent with capture_method: manual — funds held until the agent calls execute. Endpoints scoring below 70 are blocked: zero Stripe API calls, zero PaymentIntents, zero risk. The safety gate is auditable directly in the Stripe dashboard.

Stripe PaymentIntents · Manual capture
03

Attest

Verifier nodes restake ETH to sign scores via an EigenLayer AVS. Collusion or false reporting triggers slashing of the staked collateral. This is not a trust-us promise — it's a verify-or-punish guarantee. Every score ships with a cryptographic proof your agent can independently verify.

EigenLayer AVS · ETH slashing
04

Query before you pay

POST /v1/probe to score the endpoint, then POST /v1/escrow/validate to hold or block funds. PASS decisions create a Stripe PaymentIntent. FAIL decisions terminate cleanly — no Stripe calls, no exposure. Both halves are independently verifiable by any observer.

x402 · MCP · Stripe
What Beacon provides

Real scoring. Real escrow.
Both verifiable.

NVIDIA Nemotron 3 Ultra

Nemotron-scored trust API

Every probe is judged by NVIDIA Nemotron 3 Ultra via Nous Research's inference API. The evaluator field on every response tells you whether a live model or a deterministic fallback scored it — no hidden black box.

Live probing engine

Live endpoint probing

Beacon hits endpoints live, not from cached data. Nemotron evaluates the actual response for accuracy, hallucination, and error rate. Endpoints can't game scores by serving only real users well.

SLA monitoring

Uptime & latency SLAs

Real-time uptime percentage and p99 latency tracked across every probe. Know before you pay whether an endpoint is reliably fast or quietly degraded.

Dispute registry

Dispute & scam-flag history

Every unresolved dispute, refund request, and scam report is indexed and surfaced. A permanent, auditable record that bad actors cannot erase or reset.

Stripe PaymentIntents

Stripe escrow enforcement

Trust score ≥ 70 creates a real Stripe PaymentIntent with manual capture — held until the agent executes. Below 70 the gate blocks entirely: zero Stripe calls, zero risk. Verifiable in the Stripe test dashboard.

EigenLayer AVS · Slashing

EigenLayer AVS attestation

Verifier nodes restake ETH and face slashing for dishonest attestations. Every score ships with a cryptographic proof — your agent can verify independently, not just trust Beacon's word.

Why verifiable

Not a UI promise.
A verify-or-punish guarantee.

Nemotron scores are economically enforced by EigenLayer verifiers who stake ETH and face slashing for dishonest attestations. The escrow gate is Stripe itself — not a message saying "blocked." Both halves are independently auditable by any observer.

Flywheel

1
Agent probes endpoint via Beacon
2
Nemotron 3 Ultra scores live response
3
Score ≥ 70 → Stripe PaymentIntent held
4
EigenLayer AVS attests the outcome
5
Richer history → more agents trust Beacon

Evaluated by Nemotron, not heuristics

NVIDIA Nemotron 3 Ultra judges every live probe for accuracy, hallucination, and error rate via Nous Research's inference API. The evaluator field on every score tells you exactly which brain ran the assessment.

Enforced by Stripe, not a message

A trust score below 70 doesn't show a warning — it terminates the flow before any Stripe API call is made. PASS decisions create real PaymentIntents with manual capture. The enforcement is Stripe itself.

Guaranteed by slashing, not promises

EigenLayer verifier nodes stake real ETH to attest scores. Collusion or false data triggers on-chain slashing. This is an economic guarantee backed by skin in the game, not a terms-of-service clause.

Auditable by anyone, not just Beacon

Every score includes a cryptographic attestation proof. The Stripe test dashboard independently shows PaymentIntents only for PASS decisions. Any observer can verify both halves without trusting Beacon.

Competitive landscape

Why Beacon, not
the alternatives.

Existing tools benchmark models or rank within a single marketplace. Beacon is the only neutral, verifiable, live-endpoint reputation layer that travels across the entire agent economy.

Beacon
Model benchmarks
Artificial Analysis, LMArena
Marketplace rankings
OpenRouter, Circle
Ecosystem reputation
Recall, Sahara
Scores live endpoints, not static models
Static model evals only
Usage metrics, not accuracy
Neutral & cross-marketplace
Lab-owned, limited scope
Self-serving, siloed
Verifiable / auditable scoring
Staked on-chain attestation
Dispute & scam-flag history
Review flags only
Conditional escrow enforcement
Identity-anchored (no reputation reset)
AP2 / FIDO anchor
Full support
Partial support
Not supported
Pricing

Pay as your agents
scale their trust.

Start querying for free. Volume and verification needs unlock deeper infrastructure.

Free

For agents and developers getting started.

$0forever
  • 1,000 score reads / month
  • Public leaderboard access
  • Trust score API (read-only)
  • Community support
Start free
Most popular
Pro

For production agents with real volume.

$99/ month
  • 100,000 score reads / month
  • $0.001 / query beyond included
  • Batch routing API
  • Signed-receipt telemetry
  • Synthetic accuracy probing
  • Email support
Get API key
Enterprise

For platforms and high-stakes agentic workflows.

Customcontact us
  • Unlimited score reads
  • Conditional escrow settlement
  • SLA attestations
  • Staked verifier network access
  • On-chain audit exports
  • Dedicated support + SLA
Talk to us

Reads are free to start so your agents can query Beacon everywhere. You pay as volume and verification needs grow.

beacon

Route your agent's money
with confidence.

Create your account and get early API access. One GET request before every payment — Nemotron-scored, Stripe-enforced, EigenLayer-attested.

Already have an account? Sign in · No credit card required.