docs: add banner image to README top, update model download flow (empty shell + on-device download)

README.md

@@ -1,5 +1,7 @@
 # PhoneClaw
 
+![banner](assets/banner.jpg)
+
 [English](README_EN.md) · [报告问题](https://github.com/kellyvv/phoneclaw/issues) · [功能建议](https://github.com/kellyvv/phoneclaw/issues)
 
 本地运行的私人 iPhone AI Agent，不联网，不上传，完全在设备端。
@@ -24,6 +26,7 @@ PhoneClaw 是一个运行在 iPhone 上的本地 AI Agent。它使用 Gemma 4
 - 支持图片输入（多模态）
 - 基于文件的 Skill 系统，新增能力不需要重写整个 Agent
 - 支持多轮工具调用
+- 默认可空壳安装 App，再在手机端下载 E2B / E4B
 - 已内置权限管理、System Prompt 编辑、模型切换
 - 针对 iPhone 内存上限做了缓存清理和历史裁剪
 
@@ -56,7 +59,16 @@ cd phoneclaw
 pod install
 ```
 
-### 3. 下载模型
+### 3. 可选：预下载模型到本地
+
+默认推荐方式已经变成：
+
+1. 先通过 Xcode 把空壳 App 安装到手机
+2. 打开 App
+3. 进入 `模型设置 / Model Settings`
+4. 在手机上直接下载 `Gemma 4 E2B` 或 `Gemma 4 E4B`
+
+只有你想把模型一起打包进 App 时，才需要在电脑上先下载到 `Models/` 目录。
 
 PhoneClaw 识别的目录名必须和下面保持一致。推荐先安装 Hugging Face CLI：
 
@@ -127,8 +139,9 @@ open PhoneClaw.xcworkspace
 
 - 右上角拼图按钮：Skill 管理
 - 右上角滑杆按钮：模型设置 / 系统提示词 / 权限
+- 如果当前是空壳安装，在模型设置页点击 `下载 / Download`
 
-建议先在权限页开启日历、提醒事项、通讯录，然后试试：
+建议先下载一个模型，再在权限页开启日历、提醒事项、通讯录，然后试试：
 
 ```
 这台手机的设备信息是什么
@@ -136,20 +149,30 @@ open PhoneClaw.xcworkspace
 帮我存一下王总的电话 13812345678
 ```
 
-## 如何只打包一个模型
+## 默认安装方式与模型打包
 
-这是最常见的发布方式，尤其是只想发 E2B 时。
+### 方案 A — 空壳安装，手机端下载模型
 
-### 方案 A — 只打包 E2B
+这是当前默认推荐方式。
+
+优点：
+
+1. Xcode 安装包更小
+2. 电脑端首次安装更快
+3. 可以在手机里自由切换下载 E2B 或 E4B
+
+默认工程已经不会把 `Models/` 里的模型自动打进 App 包。安装后直接去配置页下载即可。
+
+### 方案 B — 只打包 E2B
 
 1. 保留 `Models/gemma-4-e2b-it-4bit`，移除 `Models/gemma-4-e4b-it-4bit`
 2. 在 Xcode 的 Project Navigator 里删除不用的模型 folder reference，选择 Remove Reference
-3. 在 PhoneClaw > Build Phases > Copy Bundle Resources 里确认只剩要打包的模型
+3. 在 PhoneClaw > Build Phases > Copy Bundle Resources 里手动把要打包的模型加回去，并确认只剩要分发的模型
 4. 修改 `LLM/MLXLocalLLMService.swift` 里的 `availableModels`，只保留实际会随 App 分发的模型（否则配置页仍然会显示不存在的选项）
 
-### 方案 B — 同时打包 E2B + E4B
+### 方案 C — 同时打包 E2B + E4B
 
-保留两个目录和两个 Xcode 资源引用即可。用户可在 App 的模型设置页里切换。
+保留两个目录，并把两个模型 folder reference 都手动加回 `Copy Bundle Resources`。用户可在 App 的模型设置页里切换。
 
 ## 自定义 Skill
 
@@ -193,7 +216,7 @@ examples:
 通常是因为对应 Skill 还没有真正执行到系统 API。如果之前已经拒绝过一次，iOS 也不会反复弹框，需要到系统设置里手动开启。
 
 为什么切模型后加载失败？
-先确认：模型目录名和代码里的 `availableModels` 一致；该模型确实被打进了 App 包；设备内存足够。
+先确认：模型目录名和代码里的 `availableModels` 一致；如果你走的是空壳安装，模型已经在手机端下载完成；如果你走的是内置分发，该模型确实被打进了 App 包；设备内存足够。
 
 为什么提醒事项创建失败？
 最新代码会先尝试复用现有提醒列表；如果系统里没有可写列表，会再尝试自动创建一个 PhoneClaw 提醒列表。如果这一步仍失败，通常是系统提醒源本身不可写。

README_EN.md

@@ -1,5 +1,7 @@
 # PhoneClaw
 
+![banner](assets/banner.jpg)
+
 [中文](README.md) · [Report an Issue](https://github.com/kellyvv/phoneclaw/issues) · [Request a Feature](https://github.com/kellyvv/phoneclaw/issues)
 
 A local AI Agent that runs entirely on your iPhone. No cloud, no network, no data uploads.
@@ -24,6 +26,7 @@ PhoneClaw uses Gemma 4 to run inference entirely on-device. It does not depend o
 - Image input support (multimodal)
 - File-driven Skill system — add new capabilities without rewriting the Agent
 - Multi-round tool calling
+- Supports shell-only app install, then on-device E2B / E4B download
 - Built-in permission management, system prompt editor, and model switcher
 - Dynamic history trimming and GPU cache management for iPhone memory limits
 
@@ -56,7 +59,16 @@ cd phoneclaw
 pod install
 ```
 
-### 3. Download a model
+### 3. Optional: pre-download a model locally
+
+The default recommended flow is now:
+
+1. Install the app shell to the iPhone from Xcode
+2. Open the app
+3. Go to `Model Settings`
+4. Download `Gemma 4 E2B` or `Gemma 4 E4B` directly on the phone
+
+You only need the `Models/` directory on your Mac if you want to bundle a model inside the app itself.
 
 Directory names must match exactly as shown below. Install the Hugging Face CLI first:
 
@@ -127,8 +139,9 @@ After opening the app:
 
 - Top-right puzzle icon: Skill management
 - Top-right slider icon: Model settings / system prompt / permissions
+- If you installed a shell-only app, tap `Download` in the model settings page first
 
-Enable Calendar, Reminders, and Contacts in the permissions page first, then try:
+Download a model first, then enable Calendar, Reminders, and Contacts in the permissions page, then try:
 
 ```
 What is this device's information?
@@ -136,20 +149,30 @@ Remind me tonight at 8 to send the file
 Save Wang's phone number 13812345678
 ```
 
-## Bundling Only One Model
+## Default Install Flow and Model Bundling
 
-This is the most common distribution setup, especially when shipping E2B only.
+### Option A — Shell app + on-device model download
 
-### Option A — E2B only
+This is now the default recommended setup.
+
+Advantages:
+
+1. Much smaller install size from Xcode
+2. Faster first-time app installation from the Mac
+3. Users can choose E2B or E4B directly on the phone
+
+By default, the project no longer bundles anything from `Models/` into the app.
+
+### Option B — E2B only
 
 1. Keep `Models/gemma-4-e2b-it-4bit`, remove `Models/gemma-4-e4b-it-4bit`
 2. In Xcode's Project Navigator, delete the unused model folder reference and choose Remove Reference
-3. In PhoneClaw > Build Phases > Copy Bundle Resources, confirm only the desired model remains
+3. In PhoneClaw > Build Phases > Copy Bundle Resources, manually add back the model you want to ship and confirm only that one remains
 4. Edit `availableModels` in `LLM/MLXLocalLLMService.swift` to only include the models actually shipped (otherwise the settings page will show options that don't exist)
 
-### Option B — Both E2B and E4B
+### Option C — Both E2B and E4B
 
-Keep both directories and both Xcode resource references. Users can switch in the app's model settings page.
+Keep both directories, then manually add both folder references back into `Copy Bundle Resources`. Users can switch in the app's model settings page.
 
 ## Adding Custom Skills
 
@@ -193,7 +216,7 @@ Why are there no permission dialogs after install?
 The corresponding Skill has likely not reached the system API call yet. If you previously denied permission, iOS will not prompt again — go to system Settings to re-enable.
 
 Why does the model fail to load after switching?
-Verify that the model directory name matches `availableModels` in code, that the model was actually included in the app bundle, and that the device has enough memory.
+Verify that the model directory name matches `availableModels` in code, that the model has finished downloading on-device if you are using the shell-only install flow, or that it was actually included in the app bundle if you are shipping it built-in, and that the device has enough memory.
 
 Why does creating a reminder fail?
 The latest code first attempts to reuse an existing writable reminder list. If none is found, it tries to automatically create a PhoneClaw list. If that also fails, the system reminder source itself is likely read-only.