# AXRI (Agentic eXperience Readiness Index)

> AXRI is a web application that evaluates a website's readiness for AI agents by running a set of machine-readiness checks and producing a scorecard with evidence and recommendations.

Last-Updated: 2026-02-17
Primary-Language: en

## Product Summary

AXRI scans a target website and returns:
- A readiness score (0-100)
- A letter grade (A-D)
- Check-by-check status with evidence snapshots
- Actionable recommendations

The product is designed for teams improving site interoperability for AI agents and automated clients.

## Primary Pages

- Home: `/`
- About checks and scoring: `/about`
- This file: `/llms.txt`

## API Surface

- Analyze URL (runs or reuses result): `POST /api/analyze`
- Recent evaluations (latest per domain): `GET /api/evaluations?limit=20`
- Evaluation details by id: `GET /api/evaluations/{id}`

### Analyze Request

`POST /api/analyze`
```json
{
  "url": "https://example.com"
}
```

### Analyze Response (shape)

```json
{
  "id": "123",
  "targetUrl": "https://example.com",
  "normalizedOrigin": "https://example.com",
  "analyzedAt": "2026-02-17T10:00:00.000Z",
  "checkCount": 7,
  "cached": false,
  "score": 82,
  "grade": "B",
  "checks": [],
  "recommendations": []
}
```

## Checks Evaluated

Core/optional checks currently include:
- `llms.txt` availability
- JSON-LD structured data
- API / machine endpoint hints
- Machine-friendly content type negotiation (markdown/json/plain-text fallback)
- Sitemap discoverability
- `ai-plugin.json` (optional)
- `ai.txt` (optional)
- `security.txt` (optional)
- `robots.txt` (optional)

Shopping-related checks are detected in backend logic (Agent Card / commerce protocol signals), and currently treated as informational in scoring.

## Scoring Model (Current)

- If `llms.txt` is found: full readiness weight is satisfied.
- If `llms.txt` is not found, fallback weights are used:
  - JSON-LD: 25
  - API hints: 35
  - Machine-friendly content: 30 (partial credit possible for plain text)
  - Sitemap discoverability: 10

Optional and shopping checks are surfaced for context but currently do not add points.

## Caching and Freshness Rules

- If a domain has been evaluated in the last 24 hours, AXRI returns the previously stored evaluation (`cached: true`) instead of running a new full scan.
- A per-domain check counter is still incremented for every analyze request.
- If the last evaluation is older than 24 hours, AXRI performs a fresh scan, stores new results, and increments the counter.

## Data Storage

Backend uses PostgreSQL (raw SQL, no ORM).

Primary tables:
- `evaluations`
- `domain_check_counters`

Relevant SQL:
- Base schema: `db/schema.sql`
- Counter migration/backfill: `db/migrations/20260217_domain_check_counters.sql`

## Operational Notes

- AXRI needs `DATABASE_URL` configured to persist and serve stored evaluations.
- If DB is not configured, API routes that rely on persistence may return empty lists or errors.
- Network fetches to target URLs use timeouts and content-type probing.

## Project Stack

- Next.js (App Router)
- React
- TypeScript
- PostgreSQL via `pg`

## Intended Consumers

- Engineering teams building AI-agent-compatible web properties
- Developer tools and automation agents that need machine-readiness diagnostics
- Internal platform teams tracking readiness improvements over time

## Limitations

- Standards in this ecosystem are evolving; checks are directional indicators, not a final definition of readiness.
- Site behavior can vary by geography, auth, bot protection, and transient network conditions.
- A single homepage-first scan does not represent every path of a complex site.

## Contact / Ownership

AXRI is built and maintained by Tria Labs.