# AgentMetrix.ai
> AgentMetrix.ai is the FinOps and governance platform for the agentic web. It tracks AI agent traffic, measures agent readability scores, attributes LLM costs across agent-to-agent chains, detects semantic drift, auto-routes tasks to cheaper models, and generates AI-corrected OpenAPI specs and MCP manifests from failure logs.

## Core Capabilities

**Agent Readability Audit** — Score any API documentation URL from 0–100. Identifies missing operationIds, parameter descriptions, error schemas, and MCP definitions that cause AI agents to fail.

**Cost Attribution & FinOps** — Track exact LLM token costs per agent run, per model, per A2A chain. Versioned pricing table for 21 models across OpenAI, Anthropic, Google, Mistral, and AWS Bedrock. Stripe Billing Meters integration for B2B invoicing.

**Autonomous Loop Circuit Breakers** — Monitor real-time tool call velocity and spend per trace ID. Auto-pause or kill agent workflows when thresholds are exceeded (e.g., >15 tool calls in 30s or >$5 in one session). Quarantine workflows for human review.

**Semantic Drift Monitor** — Compare the semantic intent of the original root prompt against downstream agent prompts. Flag "Context Divergence Warnings" when Agent D's context has drifted too far from Agent A's original intent. Drift score 0.0–1.0.

**Dynamic Model Routing Engine** — Analyze historical performance data and auto-route agent tasks to the most cost-effective model meeting the success rate threshold. Identifies tasks where cheaper models achieve 90%+ success rates at 60-95% lower cost.

**AEO Documentation Upgrader** — Analyze agent failure patterns and automatically generate corrected OpenAPI specs, MCP server manifests, and llms.txt files. Drop-in replacements for broken documentation.

**A2A Relationship Ledger** — Track agent-to-agent call chains with root_trace_id propagation. See full cost breakdown across every hop in a multi-agent workflow.

**Spend Cap Rules Engine** — Create rules that alert, pause, or block agent calls when spending exceeds a threshold in a rolling time window. Slack webhook integration.

**Enterprise Analytics** — A2A Sankey flow diagrams, Token Burn Rate sparklines with Green/Yellow/Red warning zones, and MCP Friction Heatmaps showing which tools cause the most agent failures.

**Public API Leaderboard** — Pre-audited scores for 50+ developer APIs. Stripe leads at 91/100, AWS at 58/100.

**LLM Cost Calculator** — Public tool to calculate exact costs for any model, token count, and usage volume with monthly projections.

## Authentication
All API calls require OAuth 2.1 with PKCE. Base URL: https://api.agentmetrix.ai/v1
Required scopes: read:sessions, write:sessions, read:cost, write:cost, read:reports

## A2A Trace Headers
To enable agent-to-agent cost attribution and circuit breaker monitoring, add these headers:
- X-AgentMetrix-Agent-ID: your-agent-identifier
- X-AgentMetrix-Trace-ID: unique-trace-id-for-this-hop
- X-AgentMetrix-Root-Trace-ID: root-trace-id-for-the-chain
- X-AgentMetrix-Parent-Trace-ID: parent-hop-trace-id

## Key Endpoints
- POST /api/ingest — Track agent events (cost attribution, circuit breaker, A2A ledger)
- POST /api/trpc/audit.run — Run an Agent Readability Audit on any URL
- GET /api/trpc/cost.getSiteSpend — Get spend breakdown by day and model
- GET /api/trpc/cost.getA2AChains — Get A2A chain cost ledger with hop drilldown
- POST /api/trpc/circuitBreaker.createRule — Create a circuit breaker rule
- GET /api/trpc/circuitBreaker.getQuarantined — Get quarantined agent sessions
- GET /api/trpc/semanticDrift.getAlerts — Get semantic drift alerts
- GET /api/trpc/modelRouting.getSuggestions — Get model routing cost-saving recommendations
- POST /api/trpc/aeoUpgrader.analyzeAndGenerate — Generate corrected OpenAPI/MCP documentation
- GET /api/trpc/cost.getLlmPricing — Get current LLM pricing table (21 models)
- POST /api/trpc/cost.calculateCost — Calculate cost for given token counts
- POST /api/trpc/demo.startSession — Start a tracked demo session
- GET /api/trpc/demo.getFunnelData — Get demo conversion funnel analytics (admin)

## Tracking Snippet
```html
<script async src="https://agentmetrix.ai/agent.js" data-id="YOUR_SITE_ID"></script>
```

## MCP Server
Endpoint: https://mcp.agentmetrix.ai
Tools: get_schema_score, run_audit, get_cost_breakdown, get_circuit_breaker_status, get_drift_alerts, get_routing_suggestions, generate_aeo_upgrade, get_agent_sessions

## Pricing
- Developer: $0/mo — 10,000 agent calls/month
- Growth: $149/mo — 250K agent calls + Circuit Breakers + Semantic Drift + Slack Alerts + AEO Upgrader
- Scale: $499/mo — 1.5M agent calls + Dynamic Model Routing + Enterprise Analytics + SOC 2
- Enterprise: Custom — Unlimited + dedicated infrastructure + SLA

## Pages
- /demo — Interactive self-guided product demo (8 steps, ~5 minutes)
- /audit — Free Agent Readability Audit tool
- /cost-calculator — LLM Cost Calculator (public)
- /scores — Public API Leaderboard (50+ APIs)
- /guide — Agentic Web Guide
- /blog — Technical blog (7 AEO-optimized articles)
- /integrations — Integrations catalog (Slack, GitHub, Datadog, Sentry)
- /mcp — MCP Server documentation
- /aeo-upgrader — AEO Documentation Upgrader (authenticated)
- /enterprise — Enterprise Analytics (Sankey, Burn Rate, Friction Heatmap)
- /cost — Cost Attribution Dashboard

## Full Documentation
- OpenAPI Spec: https://agentmetrix.ai/openapi.json
- Agentic Web Guide: https://agentmetrix.ai/guide
- Blog: https://agentmetrix.ai/blog
- ai-plugin.json: https://agentmetrix.ai/.well-known/ai-plugin.json