docs: update README for iOS app direction + sidecar architecture

This commit is contained in:
jay 2026-05-15 04:50:22 +02:00
parent b2b82c82ce
commit aa8aa42655
1 changed files with 74 additions and 26 deletions

100
README.md
View File

@ -1,53 +1,101 @@
# pi-remote-control
A `pi` extension that exposes running sessions over WebSocket — used today
as a browser-based remote control, and the foundation for a native iOS app
currently in development.
## Disclaimer
This project is for personal use and research only. It is provided as-is, and the author accepts no liability for any damage, loss, misuse, or operational consequences that result from installing or using it. The server has no built-in authentication beyond a session token and no HTTPS on the dynamic port — see [Security notes](#security-notes) for details. Do not use it for safety-critical, multi-user, or untrusted-network deployments.
Personal use and research only. Provided as-is, no liability for damage,
misuse, or operational consequences. See [Security notes](#security-notes).
## Install
---
## Current state: browser client
The extension ships a working HTML/WebSocket client that mirrors a pi
session in any browser — including iPhone Safari.
### Install
```bash
pi install https://github.com/goofansu/pi-remote-control
pi install git:git.vpsj.de/jay/pi-remote-control
```
## Usage
### Usage
Run `/remote-control` to open the menu:
Run `/remote-control` inside pi to open the menu:
- **Turn on / Turn off** — start or stop the server
- **Configure URL** — set the base URL exposed by your local tunnel or proxy, saved to `~/.pi/agent/remote-control.json`
- **Status** — show the QR code and connection URL (only when server is running)
- **Configure URL** — set the base URL for your tunnel or proxy
- **Status** — show the QR code and current session URL
> **Note:** On first use, you must configure the URL before the server can start.
To start the server automatically on launch:
To start the server automatically:
```bash
pi --remote-control
```
## Use case
The server binds to `127.0.0.1` and is reached through a local tunnel
(e.g. Surge Ponte, Tailscale). Open the QR URL in any browser.
The remote-control server binds to `127.0.0.1` on the host running `pi` and is reached through a local tunnel or proxy. This example uses [Surge Ponte](https://kb.nssurge.com/surge-knowledge-base/guidelines/ponte), which provides an end-to-end encrypted device-to-device tunnel without exposing the server to the LAN.
![pi remote control on iPhone](assets/screenshot-mobile.png)
The setup is:
---
1. Install this extension on the Mac that runs `pi`.
2. Enable Surge Ponte on that Mac and give it a device name such as `pi`.
3. On the same Mac, open `pi` and run the `/remote-control` command.
4. Choose `Configure URL` and set the base URL to your Surge Ponte hostname, for example `http://pi.sgponte`.
5. Choose `Turn on`.
6. Open `Status` to get the QR code and connection URL for the current session.
7. On another device on the same Surge Ponte network, open that URL in a browser.
## In development: native iOS app
In this setup, the browser URL is `http://pi.sgponte:<port>`, where the port is assigned when the server starts. Use `Status` to get the current URL or scan the QR code — it changes each time the server restarts.
A native iOS app is being built on top of this extension's WebSocket
infrastructure. Design goals:
Here's what it looks like on iPhone — this is an actual session asking `pi` about its hardware environment:
- **Byte-exact mirror** of the terminal session — what you see over SSH
is what you see on the phone, rendered via SwiftTerm.
- **Session persistence** — sessions run for days, the app reconnects
instantly after backgrounding (< 1s, via sequence-cursor delta replay).
- **Multi-session** — spawn, name, and switch between pi sessions from
the phone as easily as browser tabs.
- **Pi-aware augmentation** — modifier bar tuned for pi (Ctrl, Esc, Tab,
arrows, Shift+Enter), slash-command palette, status bar showing what pi
is currently doing, push notifications when pi is waiting for input.
- **QR pairing** — scan once, self-signed TLS + token pinned automatically.
<img src="assets/screenshot-mobile.png" width="300" alt="pi remote control on iPhone via pi.sgponte">
### Architecture (in progress)
```
pi (Ink TUI) ◄──► tmux session ◄──► SSH (Mac terminal)
tmux -C (control mode)
pi-remote sidecar (this extension, extended)
│ wss:// (binary ANSI + JSON side-channel)
iOS app (SwiftUI + SwiftTerm)
```
Streaming uses tmux control mode (`tmux -C`) rather than `pipe-pane`
verified reliable across alternate-screen transitions in a PoC spike.
### Implementation docs
All planning and coordination lives in [`docs/`](./docs/):
| File | Purpose |
|---|---|
| [`docs/NEXT-STEPS.md`](./docs/NEXT-STEPS.md) | Current state + what to do next |
| [`docs/PHASE-1-sidecar.md`](./docs/PHASE-1-sidecar.md) | Sidecar production-ready |
| [`docs/PHASE-2-ios-mvp.md`](./docs/PHASE-2-ios-mvp.md) | iOS app MVP |
| [`docs/PHASE-3-ios-augmentation.md`](./docs/PHASE-3-ios-augmentation.md) | iOS polish features |
| [`docs/SYNC.md`](./docs/SYNC.md) | Multi-agent coordination |
| [`docs/reference/SPEC-ios-app.md`](./docs/reference/SPEC-ios-app.md) | Full feature spec (v3) |
---
## Security notes
- The server only listens on localhost. Remote access depends on whatever local tunnel or proxy you configure.
- There is no multi-user authentication. Treat the connection URL as a secret for the lifetime of the session.
- If you use a reverse proxy instead of Surge Ponte, configure it to terminate TLS at a fixed `https://` endpoint and forward to the server's dynamic backend port. Do not expose the dynamic port directly over a public network, as the server does not support HTTPS and any token or session cookie would be transmitted in cleartext.
- The server only listens on localhost. Remote access depends on your tunnel.
- No multi-user authentication. The connection URL is a per-session secret.
- **iOS app** (in development) uses bearer tokens stored in the iOS Keychain,
self-signed TLS with fingerprint pinning via QR pairing, and optional
Face-ID gate — no CA or public PKI required.
- If using a reverse proxy, terminate TLS there and do not expose the
dynamic backend port directly.