Self-hosting

anymd is MIT-licensed and runs as a single Cloudflare Worker. Everything the hosted service uses is in the repository at github.com/digitopvn/anymd.

What you need

• A Cloudflare account with Workers, D1, KV, R2, Vectorize and Workers AI available.
• Node.js and npm.
• A domain on Cloudflare, if you want a custom hostname.

1. Clone and install

git clone https://github.com/digitopvn/anymd.git
cd anymd
npm install
npx wrangler login

2. Create the Cloudflare resources

The repository's wrangler.jsonc declares two environments, staging and production. Each needs:

Binding	Type	Used for
DB	D1 database	Users, keys, library, usage, traces, billing, pages, posts
OAUTH_KV	KV namespace	MCP OAuth state
CACHE	KV namespace	Conversion cache, anonymous counters, published-page cache
MEDIA	R2 bucket	Media, social images, CLI tarball
VECTORS	Vectorize index	Semantic search embeddings
AI	Workers AI	Embeddings, query fan-out, document conversion
RL_ANON, RL_AUTH	Rate limiting	Per-minute limits for anonymous and signed-in callers
RL_DOMAIN	Rate limiting	Per-minute fetch budget per target site, shared by all callers
RL_MCP, RL_MCP_MUTATION	Rate limiting	Per-minute MCP limits per user and credential: every request, and the stricter budget for tools that change data. Optional: without them MCP falls back to RL_AUTH

Create them (production shown; repeat with -staging names for staging):

npx wrangler d1 create anymd-production
npx wrangler kv namespace create OAUTH_KV
npx wrangler kv namespace create CACHE
npx wrangler r2 bucket create anymd
npx wrangler vectorize create anymd-chunks-production --dimensions=1024 --metric=cosine

Vectorize metadata indexes

Semantic search filters vectors by owner, and deletes them by document. Create both metadata indexes before any document is embedded, because Vectorize only indexes metadata on vectors written after the index exists:

npx wrangler vectorize create-metadata-index anymd-chunks-production --propertyName=user_id --type=string
npx wrangler vectorize create-metadata-index anymd-chunks-production --propertyName=doc_id --type=string

The index dimension (1024) matches the bge-m3 embedding model used by the library.

3. Point wrangler.jsonc at your resources

wrangler.jsonc ships with Digitop's account and resource ids. Replace, per environment:

• account_id
• d1_databases[].database_id (and database_name if you renamed it)
• kv_namespaces[].id for OAUTH_KV and CACHE
• r2_buckets[].bucket_name and vectorize[].index_name if you renamed them
• routes with your own hostnames
• ratelimits[].namespace_id: any integer unique within your account
• vars: PUBLIC_URL, CDN_URL, GITHUB_REPO, BILLING_PROVIDER (creem or polar), CREEM_SERVER (test or production), POLAR_SERVER (sandbox or production)

Serve the R2 bucket from a custom domain (Cloudflare dashboard → R2 → your bucket → Custom domains) and put that origin in CDN_URL.

The SSRF guard in src/convert/index.ts refuses to convert anymd's own hostnames, and also the hosts of your PUBLIC_URL and CDN_URL, so your deployment can't be asked to fetch itself.

Private deployments

Two optional vars close a copy you run only for yourself or your team. Both default to off, which keeps the hosted behavior:

• SIGNUP_DISABLED: "true" lets only the emails in ADMIN_EMAILS create accounts, by password or by GitHub/Google sign-in. Existing accounts still sign in.
• ANONYMOUS_CONVERT_DISABLED: "true" refuses conversions without an account or API key (401 authentication_required) on the URL, REST and file endpoints, instead of allowing the anonymous daily limit.

Sign up with your ADMIN_EMAILS address right after the first deploy: sign-up does not verify email, so whoever registers that address first becomes the owner.

4. Apply database migrations

Migrations live in migrations/. Apply them to the remote database:

npm run db:migrate:staging
npm run db:migrate:production

These scripts use the database names anymd-staging and anymd-production; adjust package.json if yours differ. For local development, apply them to the local database instead:

npx wrangler d1 migrations apply anymd-staging --local --env staging

Every migration is additive, so the previous Worker version keeps working while you roll out. 0007_admin_control_plane.sql adds the audit, account status, credit-grant lifecycle, settings version and webhook outcome columns the admin control plane needs; apply it before deploying the Worker that uses them.

5. Set secrets

First, set ADMIN_EMAILS in wrangler.jsonc (a plain var, per environment). It holds the comma-separated emails that become owner when they sign up.

All secrets are optional. Features that need one stay off, or degrade gracefully, until it is set. Set them per environment, one by one or from your .env in one go:

npx wrangler secret put OPENROUTER_API_KEY --env production
npm run secrets:production   # pushes every known secret found in .env, without printing values

Secret	Enables
CREEM_API_KEY	Creem checkout and customer portal (with BILLING_PROVIDER=creem)
CREEM_WEBHOOK_SECRET	Verifying Creem webhooks
POLAR_ACCESS_TOKEN	Polar checkout, customer portal and usage-based overage (with BILLING_PROVIDER=polar)
POLAR_WEBHOOK_SECRET	Verifying Polar webhooks
GITHUB_TOKEN	Higher GitHub API rate limit for the changelog
RESEND_API_KEY	Transactional email (welcome, password reset)
RAPIDAPI_KEY, VIDCAP_API_KEY	YouTube transcripts (either works; RapidAPI is tried first) and the opt-in video download (either works; VidCap is tried first, with VIDEO_QUEUE). RAPIDAPI_KEY also enables Facebook, Instagram, Threads (API4) and LinkedIn adapters and their provider-backed comments.
YOUTUBE_FALLBACK_API_KEY	A self-hosted VidCap-compatible API (for example yt-dlp on your own server), tried last for YouTube transcripts and video downloads, after VidCap and RapidAPI. Set its base URL in the YOUTUBE_FALLBACK_URL var; it must serve GET /youtube/caption?url=&locale= and GET /youtube/media?url= in VidCap's response format and accept the key as X-API-Key. Media must return stream URLs on googlevideo.com or youtube.com. Works on its own, without the hosted providers
OPENROUTER_API_KEY	Query fan-out and Jev via OpenRouter (Workers AI is the fan-out fallback), article-image OCR and descriptions using qwen/qwen3.6-35b-a3b, and the opt-in video analysis using google/gemini-3.8-flash
TYPESAFE_API_KEY	Jev directly from TypeSafe, when OpenRouter isn't configured
GITHUB_CLIENT_ID, GITHUB_CLIENT_SECRET	"Continue with GitHub" sign-in
GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET	"Continue with Google" sign-in

Sign in with GitHub or Google

A provider's button appears only when both its client id and secret are set. Register this callback URL with the provider, using your own host:

https://anymd.cc/api/auth/oauth/github/callback
https://anymd.cc/api/auth/oauth/google/callback

The callback defaults to PUBLIC_URL plus that path; set GITHUB_CALLBACK_URL or GOOGLE_CALLBACK_URL only to override it. A GitHub OAuth App accepts one callback host, so staging needs its own app: put its values in .env as STAGING_GITHUB_CLIENT_ID and STAGING_GITHUB_CLIENT_SECRET, and npm run secrets:staging pushes them. A Google client can serve both hosts once the staging callback is added to its authorized redirect URIs. Request the user:email (GitHub) and openid email profile (Google) scopes; anymd asks for them itself.

Sign-in links to an existing account only through an email address the provider marks as verified. Because sign-up with a password does not verify email, linking removes any password on that account and signs out its sessions; the owner can add a password again with "Forgot password".

The authoritative list is the Env interface in src/env.ts. For local development, put values in .dev.vars (git-ignored). Never commit secrets.

Billing

BILLING_PROVIDER picks the one payment provider that takes checkouts: creem or polar. The other stays off even if its secrets are set, and without the chosen provider's key billing stays off while everything else works.

Creem

Use CREEM_SERVER=test for anything that isn't production; it talks to https://test-api.creem.io with a test-mode key. anymd finds your products by name: anymd Pro (monthly), anymd Pro (yearly), anymd Scale (monthly) and anymd Scale (yearly). Discount codes are plain Creem discounts (for example LAUNCH30). Creem has no usage meter, so it cannot bill overage: with Creem, plans that set overagePer1k keep going past their credits without an extra charge. Use Polar if you want overage billed.

scripts/creem-setup.mjs creates the four products, the discount codes and the webhook, and is safe to re-run. It saves the webhook signing secret to .env without printing it:

npm run creem:setup -- --webhook https://staging.example.com/api/webhooks/creem                  # test mode, CREEM_API_KEY
npm run creem:setup -- --production --webhook https://example.com/api/webhooks/creem             # live, CREEM_PRODUCTION_API_KEY

npm run secrets:production then pushes CREEM_PRODUCTION_API_KEY and CREEM_PRODUCTION_WEBHOOK_SECRET as the production Worker's CREEM_API_KEY and CREEM_WEBHOOK_SECRET. If Creem does not return the secret, copy it from Creem → Developers → Webhooks into .env.

Polar

Use POLAR_SERVER=sandbox for anything that isn't production. anymd finds your Polar products by metadata, so no product ids live in config: give each recurring product anymd_plan = pro or scale. Polar meters usage, so it can bill overage for plans that set overagePer1k in src/billing/plans.ts.

scripts/polar-setup.mjs creates the credits meter, the included-credit benefits, the four products, the discount codes and the webhook in one run, and is safe to re-run:

npm run polar:setup -- --webhook https://staging.example.com/api/webhooks/polar                  # sandbox, POLAR_ACCESS_TOKEN
npm run polar:setup -- --production --webhook https://example.com/api/webhooks/polar             # production, POLAR_PRODUCTION_ACCESS_TOKEN

npm run secrets:production then pushes POLAR_PRODUCTION_ACCESS_TOKEN and POLAR_PRODUCTION_WEBHOOK_SECRET as the production Worker's POLAR_ACCESS_TOKEN and POLAR_WEBHOOK_SECRET.

6. Build and deploy

npm run deploy:staging
npm run deploy:production

Each runs npm run build (Tailwind CSS and client scripts) and then wrangler deploy for that environment.

Local development runs against the staging configuration:

npm run dev

If Vectorize isn't reachable from local dev, hybrid search still returns keyword results.

7. Continuous deployment

The upstream repository deploys on push:

Branch	Environment	Hostname
dev	staging	staging.anymd.cc
main	production	anymd.cc, www.anymd.cc

The workflows live in .github/workflows/. For your fork, add two repository secrets, CLOUDFLARE_API_TOKEN (with Workers, D1, KV, R2 and Vectorize edit permissions) and CLOUDFLARE_ACCOUNT_ID, and change the hostnames. A deploy job boils down to:

npm ci
npm run typecheck
npm run db:migrate:production
npm run deploy:production

Apply migrations before deploying code that depends on them.

Operating it

• Logs: npx wrangler tail --env production. Observability is enabled in wrangler.jsonc, so logs also appear in the Cloudflare dashboard.
• Rollback: npx wrangler rollback --env production returns to the previous Worker version. It does not undo D1 migrations.
• Point clients at your copy: set ANYMD_BASE_URL for the CLI and use your hostname in MCP configs.

License

MIT. Copyright (c) 2026 Digitop.ai. Keep the license notice when you redistribute.