modelright

modelright

Live registry of AI model specs, context windows, and per-1M-token pricing — running at modelright.dev.

A Next.js + Postgres app that keeps a real, current catalog of LLMs: the OpenRouter catalog is synced hourly, Artificial Analysis enriches models with benchmark and speed data, and every price change lands in an append-only snapshot log that powers the changes feed and price-history charts.

What's in it

• Catalog — /models searchable/filterable by provider, status, price ceiling, context floor, modality, and free-tier; /providers directory
• Model detail — specs, pricing, price-history chart, cost calculator, Artificial Analysis benchmark card (/models/[provider]/[slug])
• Compare — up to 4 models side-by-side with an interactive picker and cheapest-highlighting (/compare); SEO-friendly pair pages at /compare/a-vs-b
• Task picks — /pick and /pick/{chat,coding,long-context,vision,cheap-bulk} rank models by an explicit rule written on the page
• Changes feed — /changes lists recent price/status changes and model additions/removals, with an RSS feed
• Public API — GET /api/v1/models (same filters as the UI), /api/v1/models/{provider}/{slug}, /api/v1/providers, /api/v1/changes; OpenAPI at /openapi.json
• MCP server — Streamable-HTTP endpoint at /api/mcp (also /.well-known/mcp) with six read tools (search_models, get_model, list_providers, compare_models, get_changes, get_ingest_status)
• Agent surfaces — /llms.txt, /agents.txt, A2A agent card (/.well-known/agent-card.json), Smithery server card (/.well-known/mcp/server-card.json), server.json MCP-registry listing
• Analytics — cookieless page-view beacon (daily-rotating salted visitor hash, no raw IPs), edge-middleware AI-crawler tracking (/api/bot-hit), and MCP call logging, all rendered on a token-gated /admin + /admin/analytics
• Ingest pipeline — POST /api/ingest (Bearer token, zod-validated, idempotent upsert) and POST /api/cron/sync (hourly OpenRouter + AA sync, marks removed models rather than deleting)

Stack

• Next.js 14 (App Router, edge middleware) + React 18 + TypeScript
• Drizzle ORM over PostgreSQL (postgres driver) — migrations in drizzle/
• mcp-handler for the MCP Streamable-HTTP transport
• Vitest + happy-dom — 200+ tests, ~90% line coverage on src/lib + src/app/api

Setup

npm install
cp .env.example .env   # set DATABASE_URL
npm run db:migrate
npm run db:seed        # optional fixture data so pages render pre-ingest
npm run dev            # http://localhost:3000

Scripts

Script	Purpose
npm run dev / build / start	Next.js lifecycle
npm run typecheck	tsc --noEmit
npm run test	Vitest with coverage
npm run db:generate / db:migrate / db:seed	Drizzle schema lifecycle + fixtures

Environment

See .env.example. Highlights: DATABASE_URL (required), INGEST_TOKEN + CRON_SECRET (write-path auth), AA_API_KEY (Artificial Analysis enrichment), ADMIN_TOKEN (gates /admin), ANALYTICS_SALT (daily visitor-hash salt — rotate to reset identities).

Built autonomously

This repo is maintained by a LiveGraph improvement loop: queued items in QUEUE.md are picked up, implemented, tested, and merged by agents. PLAN.md holds the product roadmap, docs/loop-reports/ is the per-item delivery log, and GROWTH.md/growth/assets/ hold the distribution plan and media kit (including a walkthrough video). If you're evaluating agent-driven development, the queue history and loop reports are the interesting part.

License

MIT — see LICENSE.