# AgentRank — Agent Changelog

> Date-stamped changes to the AgentRank API, pricing and agent manifests.
> Plain markdown for easy parsing. Latest OpenAPI spec: https://jannick.io/openapi.json
> Homepage: https://jannick.io · API docs: https://jannick.io/api

## 2026-10-15 — 1.4.0
- Agency plan live at €129 per month (excl. VAT): everything in Pro, 100 client scans per billing period (max 50/day), monitoring up to 50 sites, white-label reports/PDFs/share links under the agency's own brand, and a client overview. Switching between Pro and Agency converts the subscription in place with prorated billing via Stripe.
- Scoring now covers six categories: Discoverability, Clarity, Pricing Transparency, Actionability, Trust and Local Visibility. Older reports keep their original five categories.
- Scan quotas count every call, saved or not (save:false included). Added Agency quotas (100 per billing period, max 50/day) and the 402 agency_quota error code.
- The AI visibility demo now requires a free account (max 3 per day) and is explicitly labelled a simulation — not live answers from ChatGPT, Perplexity or Gemini.
- Weekly monitor limits: Pro up to 10 sites, Agency up to 50 sites.
- White-label share links from active Agency subscriptions open without an AgentRank login.
- The AI traffic snippet requires a per-site secret key; logAgentVisit stores only the bot name, user agent, path, host, a truncated IP hash and the site key id.
- Manifests and OpenAPI aligned at version 1.4.0; review date 2026-10-15.

## 2026-09-20 — 1.3.0
- Free plan changed from 3 scans to 1 scan. A free account still gets the complete report for that scan (score, grade, five-signal breakdown, prioritized fixes, generated llms.txt and JSON-LD). runAudit returns 402 limit_reached after the first saved scan.
- Pro yearly billing is live: €290 per year (two months free) alongside €29 per month, both excluding VAT.
- All prices are shown and charged excluding VAT. Dutch and EU consumers are charged 21% VAT at checkout; EU businesses with a valid VAT number are reverse-charged. VAT number NL004399760B58 appears on every invoice.
- Scores now blend the model's judgement with machine-verified facts (robots.txt, sitemap.xml, llms.txt, JSON-LD types, agent manifests) and fixed category weights. Audits carry score_model "v2" and a protocol fact block.
- PDF export moved to Pro. AI visibility simulation (Pro) added to the published feature list.

## 2026-09-15 — 1.2.0
- runAudit's Pro fair-use 429 responses now include Retry-After in positive integer seconds, calculated from saved scan timestamps in the relevant rolling window (50 per 24 hours; 200 per 30 days). The other window may still apply. Existing 401 and 402 behavior is unchanged.
- OpenAPI now documents Retry-After, scope, follow-up scan inputs and the actual shared-report response. Interactive scans are live; force is a Pro/admin request marker, not a cache bypass requirement.
- Reviewed all three HTTP operations against their implementations. Aligned OpenAPI, the MCP-style catalog and agent capability card at version 1.2.0; retained contact, capabilities and skills.
- Removed an unresolved MCP discovery $schema URL. The MCP-style manifest is an application-specific HTTP capability catalog, not a standardized discovery document or a hosted MCP server. The agent card does not imply an implemented A2A task transport.
- Manifest review date: 2026-09-15. Next manual review due: 2026-12-15. Review protocol references, endpoint inputs, authentication, output shapes, limits and capabilities; update these files when behavior changes. This date does not create an automated review or certify client interoperability.
- Connected the existing four JSON-LD blocks with additional @id references without removing their content; added route-specific breadcrumb metadata. WebSite uses about for the Organization relationship (an Organization is not a CreativeWork parent).
- Kept the existing llms.txt and homepage offers unchanged; did not replace them with the PDF's simplified generated assets.

## 2026-09-12
- Added the `getBenchmark` endpoint: `POST .../getBenchmark` with `{ "urls": ["a.com","b.com","c.com"] }` (2–3 URLs) returns a side-by-side comparison from existing audit records — no new scans. Response: `{ rows: [...], insights: [...] }`. Documented in /openapi.json.
- Added the `benchmark_sites` tool and a `scan_trends` resource to the MCP manifest (/.well-known/mcp.json).
- Documented both in /llms.txt ("For AI agents") and added a competitor-benchmarking FAQ item to the homepage JSON-LD.
- Added this changelog file at /changelog.txt, linked from /llms.txt and /api.
- Pricing unchanged: Free €0 (3 scans, no card), Pro €29/month (fair use 50/day, 200/month).

## Earlier
- `runAudit` — scan a site (POST, signed-in session). Deep link: https://jannick.io/?url={target}.
- `getPublicReport` — read a report its owner shared publicly (POST, signed-in session).
- Agent manifests: /openapi.json, /.well-known/agent-card.json, /.well-known/mcp.json, /agents.json, /llms.txt, /robots.txt.