Ask AI — Cloudflare Worker backend

This Worker powers the Ask AI page on the site. The static site (GitHub Pages) can’t hold an API key, so all Claude calls go through here, where the key stays secret and abuse is gated (Turnstile + rate limits + size caps). Every interaction is logged to a Notion database.

browser /ask/  ──POST /ask──►  Worker  ──►  Anthropic API
                                  └──(async)──►  Notion history DB

One-time setup

0. Prerequisites

Node 18+ and a Cloudflare account (free tier is fine).
cd worker && npm install
Log in: npx wrangler login

1. Anthropic API key

Create a key at console.anthropic.com → API Keys and load a little credit (pay-as-you-go; this is separate from any Claude.ai subscription).
Keep it for step 6.

2. Cloudflare Turnstile (anti-bot)

Cloudflare dashboard → Turnstile → Add site. Add aniketwagde.com (and swagget.github.io if you test there).
Copy the Site key (public — goes in _config.yml) and the Secret key (goes in step 6).

3. Notion history database

In Notion, create a new database (full page) named e.g. “Ask AI History”.
Add these properties exactly (name and type matter): | Property | Type | |—|—| | Title | Title (default) | | Mode | Select (add options Question, Fit) | | Input | Text | | Reply | Text | | Country | Text | | Model | Text | | Tokens | Number | | Timestamp | Date |
Share the database with your Notion integration (the same token used by notion-sync, or make a new integration at notion.so/my-integrations and click Connections → your integration on the database).
Copy the database id (the 32-char hex in its URL) for step 6.

4. Create the KV namespace

npx wrangler kv namespace create RL
npx wrangler kv namespace create RL --preview

Paste the two printed ids into wrangler.toml (id and preview_id).

5. (Local testing, optional)

cp .dev.vars.example .dev.vars and fill in real values (DEV="true" skips Turnstile locally).

npm run dev, then:

curl -X POST http://localhost:8787/ask -H 'Content-Type: application/json' \
  -d '{"mode":"question","text":"What are your main research areas?"}'

6. Set secrets + deploy

npx wrangler secret put ANTHROPIC_API_KEY
npx wrangler secret put TURNSTILE_SECRET
npx wrangler secret put NOTION_TOKEN
npx wrangler secret put NOTION_HISTORY_DB_ID
npm run deploy

Wrangler prints the Worker URL, e.g. https://ask-aniket.<your-subdomain>.workers.dev.

7. Wire the frontend

Put the Worker URL and the Turnstile site key into the site’s _config.yml:

ask_ai:
  endpoint: "https://ask-aniket.<your-subdomain>.workers.dev/ask"
  turnstile_sitekey: "0x4AAAA..."

Rebuild the site. Done.

Tuning

Edit [vars] in wrangler.toml and re-deploy: model, MAX_TOKENS, IP_LIMIT/IP_WINDOW_SECONDS, DAILY_CAP, size caps. The corpus is sent with prompt caching, so repeat calls are cheap; DAILY_CAP is the hard spend bound.

Notes

The Worker fetches the knowledge base from KB_URL and edge-caches it for 1h, so content updates (via notion-sync) appear without redeploying the Worker.
Logging is best-effort and async — a Notion outage never blocks a reply.

Aniket Wagde