No. Your conversations, memory, and files live on your own server — the AgentKip project never has access to them. The website itself only stores what it needs to run: your sign-in info (via Clerk), waitlist emails, and any message you send through the contact form.

The app and the beta are free. You pay your own model-provider account directly for whatever you use — Kip doesn't mark that up or route it through a shared backend. Casual use tends to be cheap; heavier coding or research runs cost more, same as using that provider anywhere else.

Your server talks straight to the model provider you choose, using a key only you hold. That's what keeps your prompts, your data, and your bill entirely yours — nothing passes through a Kip-run relay.

A Mac, a Linux box, a Windows machine with Docker, or a small VPS — basically anything that can stay on and be reachable from your phone (directly on your network, or over Tailscale). It's mostly waiting on API calls, so it doesn't need to be powerful.

It's being set up now. Until it's live, the beta is invite-only — ask Brandon for a spot, or join the waitlist and you'll get a link as soon as one opens up.

Kip is built for iOS 27, with a compatibility build targeting iOS 26 so more devices can join the beta in the meantime.

The server descends from the open Hermes agent framework. Exactly how the app itself will be distributed and licensed is still being worked out — more details once that's settled.

Whatever your server exposes. Go through OpenRouter with a single key for access to a lot of models at once, connect directly to a provider like OpenAI or Anthropic, or run local models through Ollama.

Yes — there's a dedicated Code model lane, and durable runs mean a long task keeps going even if you back out of the app or lose connection for a moment. Delegating out to Claude Code or Codex from your phone is on the roadmap, but it isn't shipped yet.

Head to the Contact page — every message goes straight to Brandon, no ticket queue in between.