Vale construir sua versão de Rankwise?

Build me a local AI answer engine citation tracker for one website. Requirements:

- Node 22, TypeScript, SQLite via better-sqlite3, a CLI plus a server-rendered dashboard
  on localhost. No accounts, no telemetry, everything stays on my machine except the
  API calls.
- site.json holds my domain, brand aliases and competitor domains. questions.json holds
  up to 40 buyer questions I want to be the answer to.
- `rank run` sends every question through OpenAI, Anthropic, Gemini and Perplexity with
  each provider's web search or grounding tool on. Keys live in .env.
- Write one immutable row per run, question and provider: raw answer, cited URLs, model
  id and any error. Never overwrite a previous run, history is the whole point.
- Cap concurrency at 3 per provider, retry twice on 429 and 5xx with backoff, and show
  failed cells in the report rather than dropping them silently.
- Normalize every citation to hostname plus canonical path, strip tracking params, then
  compute my citation share and a top 25 cited domains table per engine.
- `rank serve` renders citation rate per engine over time, the questions where a
  competitor is cited and I am not, and a week-over-week diff.
- `rank audit <url>` fetches one page and reports the mechanical gaps only: missing or
  invalid JSON-LD, missing canonical, robots and llms.txt rules blocking AI crawlers,
  and thin heading structure. Print the JSON-LD block it would add.
- `rank export` writes runs and citations to CSV.
- Out of scope: writing articles, publishing to a CMS, and scraping the consumer chat
  UIs. Use the APIs only.
- README: setup, per-run cost estimate, a cron line for the weekly run, and a plain note
  that API answers only approximate what users are actually shown.