Cloudflare Workers — Janitor docs

These concepts apply to your Go application. Matching and private risk evaluation run in your application-owned TypeScript or Elixir engine; this client handles the HTTP connection. See your setup guide. Code labeled TypeScript describes the engine configuration.

This example runs Janitor in a Cloudflare Worker with D1, Cloudflare’s SQL database. It serves a small page and a /api/visitor endpoint. Local development uses a database on your machine and starts with AI disabled.

Run the example

You need Node 22.12+, pnpm 9.12.0 and a checkout of the Janitor repository. From its root:

pnpm install
pnpm build
cd examples/cloudflare-worker
pnpm exec wrangler d1 migrations apply VISITORS --local
pnpm dev

Open http://localhost:8787 and select Identify twice. The first call creates a browser ID; the second should return the same ID using the cookie. The page’s JavaScript is bundled locally, with no CDN dependency.

Add it to an existing Worker

Apply the D1 migrations to your database, bind it as VISITORS, and mount the handler at your chosen route:

import { createCloudflareVisitor } from "@janitor/adapters/cloudflare";

const visitor = createCloudflareVisitor({ db: env.VISITORS });
return visitor.handle(request);

Use that response for /api/visitor; continue serving your other routes normally. Keep a reusable adapter per binding configuration, as the example Worker does, so per-instance limits can apply across requests. Add the browser client to your app and keep the endpoint on the same origin.

Enable Jev when you need risk scores

Jev is TypeSafe’s AI model for structured judgments. On Cloudflare, it runs through the Workers AI AI binding without a separate TypeSafe API key:

createCloudflareVisitor({ db: env.VISITORS, ai: env.AI });

The example’s JEV_ENABLED variable defaults to false. To try real Workers AI during local development, run pnpm dev:ai. This uses the remote provider and can incur charges. Plain pnpm dev uses local bindings and makes no AI calls. Read the scores and failure behavior before acting on a result.

Deploy and maintain

Create your D1 database, replace the placeholder database_id in the Wrangler config, and apply the migrations remotely. Set JEV_ENABLED=true only if you want AI evaluation, then deploy through your normal Worker release process. pnpm build bundles and performs a dry run; it does not deploy.

Use HTTPS for cookies. Call visitor.cleanup() from existing maintenance to remove expired history. Use visitor.assess(request) if server code needs the private scores; handle() returns only the public visitor result. See storage and private assessments.