When Mini Apps Meet Agents: From UI to Conversation

---

Keychat for Human has mini apps—web apps running in a webview that can directly use Keychat's public key identity for login and Lightning Network for payments. Users don't need to register accounts or link bank cards. Open and go. This is the core advantage of mini apps: leveraging the host's identity and wallet to eliminate all friction.

So does Keychat for Agent need its own version of mini apps?

The first instinct is: have the agent open mini apps in a browser, inject a JS provider that bridges to Keychat for Agent's identity and wallet, and use them just like a human would. Technically feasible—Keychat for Agent, as a plugin, could auto-inject `window.nostr` and `window.webln` in the browser environment without extensions or changes to the OpenClaw core.

But this is the wrong direction.

Mini apps are designed for humans. Their essence is three layers: identity, payments, and business logic. The UI is just a shell for human eyes. Agents don't need to see UI—making them manipulate browsers and click buttons is fitting the foot to the shoe.

The right approach: expose the mini app's services as public agent services.

Specifically, the service provider runs a public agent service on OpenClaw with Keychat for Agent installed. The user's agent talks to it directly:

- Login: Public keys are exchanged during the Signal Protocol handshake—identity is inherently verified, no additional login flow needed
- Payments: Send a Lightning invoice in the chat, the agent pays it directly
- Business logic: Complete operations via message commands, which the service parses and responds to

What humans accomplish through a mini app's GUI, an agent accomplishes in a conversation. Same service, same identity and payment infrastructure, different interfaces—humans look at screens, agents talk.

This also means: every mini app naturally corresponds to a potential agent service. The service provider just needs to expose a Keychat messaging interface alongside the GUI to serve both humans and agents. Not two systems—two entry points to the same system.