Discover scored APIs, simulate agent purchases, and verify escrow payments.
Credit scores for APIs your agents call. Sleep at night.
Every API gets a reputation index built from its real probe history — a credit score, not a one-off test result. A single perfect response won't buy an unknown endpoint your agent's trust; the top is reached only with a sustained track record, and the score decays when an endpoint slips. Your agent transacts only above 700.
Quality
up to 650How correctly the endpoint answers, judged live by Nemotron 3 Ultra — recency-weighted and held near neutral until there's enough evidence to trust it.
Track record
up to 200Proven call volume over time. Sustained clean history is what lifts a score toward the ceiling — this is why the index can't be maxed on one good call.
Reliability
up to 150Uptime, error rate, response consistency, and dispute rate — how dependable the endpoint is across all the times your agents have hit it.
Grades A+ → F and the 700 escrow threshold are derived from the live index, so a score and its grade can never disagree. A detected scam or prompt-injection signal hard-caps the index regardless of volume. Beacon re-probes tracked endpoints on a schedule, so the number reflects how an API behaves now — not how it looked the day it was added.
Scored by Beacon's live probe and ranked by reputation. Your agent only spends on services above the 700 trust threshold — lower-scoring and flagged endpoints are shown too, so you can see exactly what the escrow gate blocks.
Watch how a Hermes agent discovers, probes, and pays a real scored API — NVIDIA NIM — for inference access, via Stripe Skills escrow by default or Blockrun x402 for agent-native micropayments.
Agents probe endpoints before spending. Try it yourself — enter any API URL and see how Beacon scores it.
// probe result
These APIs have been probed and scored by Beacon. Your agent only transacts with services whose reputation index is 700 or higher (out of 1000). The index is earned over an endpoint's probe history, so it climbs with a sustained track record rather than maxing out on a single good call.
Built for the Hermes Accelerated Agent Business Hackathon — this is a live demo. The evaluator field and Stripe dashboard are both verifiable now.
Hermes
the agent orchestrator. This demo simulates a Hermes agent selecting tools, triggering payments via Stripe Skills, and hitting Beacon's escrow gate before any money moves.
NemoClaw
the orchestration framework the agent runs on. NemoClaw is a hackathon partner and execution runtime — it doesn't appear in the Agent Directory because it isn't a scored, consumable API.
Stripe Skills for Hermes
dual-rail payments: Stripe PaymentIntents (high-value escrow) or Blockrun x402 (agent-native micropayments). Real test-mode PaymentIntents with manual capture, or gasless EIP-3009 USDC settlement on Base. The escrow gate blocks bad endpoints entirely, on either rail.
Blockrun
settles the x402 leg. Once escrow clears, Blockrun settles USDC on Base for the agent's completed request.
NVIDIA NIM
the real scored API in this demo. A live probe against integrate.api.nvidia.com is what the agent actually pays to call.
NVIDIA Nemotron 3 Ultra
the scoring brain. Live inference via Nous Research API. Every probe gets a real LLM evaluation, not a heuristic.