Agent Health Monitor

The diagnostic layer for autonomous agents: trust, health, and performance.

What is AHM

AHM is the diagnostic layer for autonomous agents — covering trust, health, and performance. It provides an objective, on-chain Agent Health Score (AHS) that answers the question every protocol, marketplace, and enterprise needs answered before an agent touches money, data, or decisions: is this agent solvent, reliable, and operational?

AHM scores agents on a 0–100 scale from verifiable on-chain signals, published on-chain and queryable via 14 REST endpoints or the AHM Shield SDK.

---

The Problem

The agent economy is scaling fast. Autonomous agents are registered in growing numbers across protocols like Virtuals ACP, Olas, and ERC-8004 on Base — delegating tasks, routing payments, and entering contracts on behalf of users and other agents. Trust infrastructure hasn’t kept up.

Most agents exhibit thin transaction histories, erratic behavioural patterns, or insufficient solvency for the tasks they claim to perform.

The stakes are rising. Agents are moving beyond simple chatbot interactions into payment authorisation, enterprise workflow automation, insurance adjudication, and multi-agent supply chains. A single under-capitalised or behaviourally erratic agent in a delegation chain can cascade failures across an entire workflow. Without a standardised health layer, every integration is a trust-me handshake.

---

The Solution: Agent Health Score (AHS)

The Agent Health Score is a composite 0–100 diagnostic built entirely from on-chain signals — unfalsifiable, permissionless, and universally queryable. By default AHS is a weighted composite of two dimensions, with a third available as an opt-in overlay:

D1 — Wallet Hygiene (30% weight · live)

Token portfolio quality, gas efficiency, transaction success rates, dust/spam token ratios, and wallet risk flags. Answers: can this agent pay for what it promises?

D2 — Behavioural Patterns (70% weight · live)

Timing regularity, counterparty diversity, adaptation patterns, and failure-recovery behaviour over weeks and months. Answers: does this agent behave predictably and reliably?

D3 — Infrastructure Health (live · opt-in overlay via agent URL)

Endpoint availability, response latency, and error rates for agents that expose a service URL. Answers: is this agent actually online and performing?

On-chain signals are the foundation because they are unfalsifiable — you cannot fake a transaction history without spending real money — and temporally rich — patterns emerge over weeks and months of activity, making scores resistant to Sybil manipulation.

AHS Grades

Every score maps to one of six grades. These are the thresholds the scoring engine applies — there is no other grade table:

Agents with no meaningful transaction history are returned as INSUFFICIENT confidence rather than being penalised with a low grade — unrated is not the same as degraded.

---

Tiered Trust Routing

Routing is a payment-gating decision derived from the grade letter — it is not a second grading scale. The /ahs/route/{address} endpoint returns one of three actions:

These are the defaults. Integrators can override them per-policy via PUT /ahs/route/policy — setting their own instant and escrow grade sets, allowlisting addresses to bypass routing entirely, disabling escrow so it falls through to reject, or applying confidence-based overrides. A Grade C agent is not permanently an escrow agent; it is one under the default policy.

This turns AHM into a trust gate that sits before payment authorisation. Protocols and marketplaces call /ahs/route to get a one-word routing decision and enforce it programmatically. No manual review. No trust-me handshakes.

---

Ecosystem Intelligence — Key Stats

Live ecosystem stats — including agent counts, average AHS, grade distribution, and zombie rate — are updated nightly. See the live dashboard for current figures.

AHM runs nightly scans across every major agent registry, maintaining the largest longitudinal dataset of agent health metrics in the ecosystem.

---

Integration

AHM is designed for programmatic consumption — by agents, protocols, and enterprise systems.

x402-native. All 14 endpoints support x402 micropayments (USDC on Base), meaning agents can pay per call with no API key, no subscription, and no human in the loop.

REST API with API key auth. Human developers purchase an API key via Stripe and authenticate with an X-API-Key header. Same endpoints, same data.

AHM Shield SDK. Drop-in middleware for agent frameworks. Install and enforce trust routing in three lines:

pip install ahm-shield

from ahm_shield import Shield

shield = Shield(api_key="ahm_sk_...")
decision = shield.route("0xABC1234567890abcdef1234567890abcdef12345")

if decision.action == "reject":
    raise Exception(f"Agent blocked — AHS {decision.score}, Grade {decision.grade}")

Endpoints (14 live)

---

Pricing

For Humans (Stripe · API Key)

All tiers include X-API-Key authentication and access to all 14 endpoints.

For Agents (x402 · Pay-Per-Call)

No API key required. Agents pay per call in USDC on Base via the x402 protocol. Prices range from $0.01 to $25.00 depending on endpoint complexity.

AHM Verify verify.agenthealthmonitor.xyz

---

Getting Started

Agent Health Monitor (AHM) is the diagnostic layer for autonomous agents — covering trust, health and performance. This guide is for developers integrating AHM into their products and for business or product leads evaluating AHM as a design partner.

What is AHM?

AHM provides real-time trust scoring for autonomous agents operating on-chain. In a world where AI agents transact, hire, and make decisions autonomously, there is no reliable way to know whether an agent is solvent, behaving consistently, operationally stable, or producing quality outputs. AHM solves this.

Every agent scored by AHM receives an Agent Health Score (AHS) — a composite score from 0 to 100 — along with a grade (A through F) and a set of dimensional scores that explain the result. The output is machine-readable and designed to be consumed directly by orchestrators, payment rails, and escrow contracts.

Live at: agenthealthmonitor.xyz

Why It Matters

Autonomous agents present a new category of counterparty risk. Unlike human contractors, agents can:

• Drain wallets silently through dust accumulation and spam transactions
• Exhibit inconsistent behaviour across sessions
• Fail operationally without warning
• Produce outputs that are superficially plausible but factually wrong

AHM provides the diagnostic layer that allows clients, orchestrators, and payment systems to make trust-gated decisions based on evidence rather than assumption.

How AHM Works

AHM scores each agent wallet from on-chain signals and maps the result to a grade, then to a routing decision. Those three things are defined once, above, rather than restated here:

• Dimensions and weights — what goes into the composite, and how D3 changes it.
• AHS grades — the six bands and their score ranges.
• Tiered trust routing — how a grade becomes an instant_settle, escrow or reject decision, and how to override the defaults.

Sanctioned wallets (OFAC, Chainalysis) score near zero on D1. Agents with no meaningful transaction history are returned with INSUFFICIENT confidence rather than a punitive grade — absence of data is not treated as evidence of risk.

API Quick Start

Authentication

Include your API key as a header:

x-api-key: your_api_key_here

API keys are available via the design partner programme or through agenthealthmonitor.xyz/app.

Score an Agent

GET /ahs/{wallet_address}

Example request:

curl https://agenthealthmonitor.xyz/ahs/0xYourAgentWalletAddress \
  -H "x-api-key: your_api_key_here"

Example response:

{
  "address": "0xYourAgentWalletAddress",
  "ahs_score": 74,
  "grade": "C",
  "grade_label": "Degraded",
  "d1_score": 81,
  "d2_score": 68,
  "sanctions_flags": [],
  "trust_routing": "escrow",
  "confidence": 0.82,
  "timestamp": "2026-04-21T12:00:00Z"
}

Batch Scoring

Score up to 25 agents in a single request:

POST /ahs/batch

{
  "addresses": ["0xAddress1", "0xAddress2", "0xAddress3"]
}

Output Quality Verification (AHM Verify)

POST /v1/outputs

Submit agent output for adversarial evaluation. Returns an ALLOW, HOLD or REJECT verdict with a confidence score and findings. $0.50 per verdict.

Full API documentation is available at docs.agenthealthmonitor.xyz.

AHM Shield

AHM Shield is a white-label trust scoring layer for partners who want to embed AHM’s capabilities into their own product. Available via PyPI:

pip install ahm-shield

AHM Shield includes partner key management, Stripe subscription billing, and configurable trust thresholds. Designed for orchestrators and payment rails that want to present trust scoring under their own brand.

AHM Verify

AHM Verify is AHM’s output quality evaluation service, and is separate from the AHS composite — it scores what an agent delivered, after the fact, rather than how its wallet behaves. It runs agent outputs through a six-role adversarial pipeline: four independent generators, an adversarial critic, and a synthesiser.

The pipeline is built entirely on Claude — the generators run on Claude Sonnet, the critic and synthesiser on Claude Opus. Each generator is prompted to adopt a different analytical style so the panel reasons from genuinely different angles, but all six roles are Anthropic models. There is no multi-vendor panel.

Live at: verify.agenthealthmonitor.xyz — $0.50 per verdict.

Verdicts include:

• verdict: ALLOW, HOLD or REJECT
• confidence: 0.0 to 1.0
• findings: specific observations from the evaluation pipeline
• methodology: pipeline configuration used

HOLD is the designed default when the panel disagrees or the evidence is thin — it means “not enough to adjudicate”, not “rejected”. For on-chain consumers the verdict maps to complete, hold_pending_review and reject respectively.

AHM Verify is aligned with the structured verification output schema being developed under ERC-8210 (Agent Assurance).

Ecosystem & Standards

AHM is an active participant in the agent economy standards ecosystem:

• ERC-8183 (Agentic Commerce) — AHM operates as a live evaluator, using on-chain behavioural signals to score agent outputs
• ERC-8210 (Agent Assurance) — AHM’s scoring architecture is aligned with the IRiskHook interface and IIndependenceSignal patterns being developed in this standard
• x402 Protocol — AHM’s pay-per-call endpoints use x402 for machine-native micropayments
• Base mainnet — AHM’s primary scanning and evaluation network

AHM’s trust registry covers agents across ACP, Olas, Celo, ERC-8004, and Arc. See the live dashboard for current count.

---

Design Partner Programme

AHM is offering select design partners three months of free enterprise access — unlimited API calls, priority support, and direct input into the product roadmap.

We are looking for protocols, marketplaces, and enterprises that are building agent-to-agent workflows and need trust infrastructure today.

Contact:

• Pablo — Founder
• Email: [email protected]
• Twitter/X: @agenttrust

---

Links

AHM — Trust layer for the agent economy.