From zero to a working tag in five minutes.

1. Sign up

Go to /sign-in. Email + password, Google, or GitHub. No card, no plan picker. You'll land on an empty dashboard with one CTA: Add your first site.

2. Add a site (the wizard does the work)

Click Add your first site →. You only need a domain — the wizard:

• fetches your homepage,
• scans for forms (contact, booking, search, order, newsletter),
• probes common /openapi.json paths,
• suggests actions you can expose with one click.

Each suggested action becomes editable: pick free / paid, point at your webhook URL, or use a prebuilt executor (Stripe Checkout, Reservio).

3. Define your custom actions (skip this if the wizard guessed right)

In the Actions tab, click + Add action. Three things matter most:

• Action id — snake_case identifier agents will call (e.g. cancel_subscription).
• Description for AI — plain language about when the agent should call this. This is the single most important field for LLM tool use.
• Executor — usually webhook pointing at an endpoint you already own. We POST signed envelopes there.

Got an OpenAPI spec? Click Import OpenAPI / Swagger in the action picker. Paste the URL, pick endpoints, batch-create.

4. Test before you save

The action editor has a built-in ▶ Run test request button. It hits your real executor with synthetic inputs and shows you the response inline. From the response, click "Use N as output schema" to auto-declare the response shape. AI agents read this and stop hallucinating fields.

5. Install the tag

The wizard sends you to /sites/[id]/setup with the snippet to paste. Two formats:

• Option A · script tag. One-liner, JS injects the link element.
• Option B · meta link. Pure HTML, no JS.

We publish the manifest at five channels for redundancy: <link>,<meta>, JSON-LD, HTTP Link: header, and the canonical /.well-known/agent-actions.json. Click Verify installation; we fetch your site and confirm.

6. Try the playground

Type the way a real user would speak to an agent. The playground picks a matching action (using your description as ground truth), fills synthetic params, runs the same executor an external agent would hit, and shows the full debug trace. With ANTHROPIC_API_KEY or OPENROUTER_API_KEY set, the playground uses a real LLM with tool-use; without, it runs a deterministic simulator.

7. Watch the audit log

Every consent, payment, action call, and 4xx/5xx is recorded. Filter, export to CSV/JSONL, revoke any agent's token. Owner email gets a one-time alert on the first ever invocation so you don't miss it.

8. Discover any other site

Public, no login: /discover. Paste any domain, see exactly what an agent would discover. Useful for debugging your own deployment from outside.

Going to production

For real-world use, set these env vars on your platform deployment:

• RESEND_API_KEY — owner notifications and signed transactional email.
• STRIPE_SECRET_KEY — Stripe Connect integration tile.
• RESERVIO_CLIENT_ID + RESERVIO_CLIENT_SECRET — Reservio OAuth.
• ANTHROPIC_API_KEY or OPENROUTER_API_KEY — real LLM in the playground.
• AAM_RPC_URL_BASE + AAM_RPC_URL_BASE_SEPOLIA — own RPC endpoints for x402 verification (faster than public RPC).
• AAM_RELAYER_PRIVATE_KEY — wallet that submits TransferWithAuthorization on-chain so your tenant doesn't have to.

Programmatic access

The platform is itself an AAM-aware site. Try the dogfood:

# 1. Fetch our manifest
curl -s https://aam-platform-gamma.vercel.app/.well-known/agent-actions.json | jq

# 2. Call a public action (no auth)
curl -X POST https://aam-platform-gamma.vercel.app/api/aam/platform/get_pricing \
  -H "Content-Type: application/json" -d '{}'

# 3. Or use the MCP bridge from any MCP-aware agent
npx @aam/mcp-bridge

Questions, bug reports, partnerships

Email tadeas@raska.eu or call cal.com/tadeas-raska-fuhanv/30min.