magpie-gemini-web 0.9.0 → 0.10.0

package/README.en.md

@@ -21,7 +21,7 @@ The plugin provides two independent providers that can be used individually or t
 | **Image generation** | Nano Banana models, for Magpie's image generation (Magpie v0.1.2075+, see below) | Not supported |
 | **Sign-in method** | Sign in directly in an isolated popup browser window | Copy `cookie` from a private/incognito window |
 | **Usage cards** | Shows models currently on cooldown and recovery times | Syncs **5-hour** & **weekly** usage and reset times from the web |
-| **Local backend** | AIStudio2API (`v0.2.19`) | gemini-web2api-go (`4.27.0-magpie.1` optimized build) |
+| **Local backend** | AIStudio2API (`v0.3.0`) | gemini-web2api-go (`4.27.0-magpie.2` optimized build) |
 
 ---
 
@@ -152,6 +152,7 @@ When a plugin error asks you to run a command, it includes the full ready-to-cop
 | :--- | :--- |
 | `npm run studio:status` | Show local service and account status |
 | `npm run studio:start` / `studio:stop` | Start / stop the local service |
+| `npm run studio:clean` | Stop the service and remove downloaded binaries & Camoufox (keeps accounts & config) |
 | `npm run studio:login` | Add a Google account |
 | `npm run studio:relogin -- [<account>]` | Re-authenticate an expired account (argument optional if only one) |
 | `npm run studio:install -- --archive <file>` | Offline install from a pre-downloaded archive |
@@ -169,6 +170,9 @@ When a plugin error asks you to run a command, it includes the full ready-to-cop
 - Both backends listen on `127.0.0.1` only. Cookies and sessions remain on your machine; Magpie stores only a locally generated API key.
 - **Gemini Web data directory:** `~/.local/share/magpie-gemini-web` (override with `GEMINI_WEB_DATA_DIR`), logs at `backend.log`. Supports macOS, Linux, and Windows (x64 / arm64).
 - **AI Studio data directory:** `~/.local/share/magpie-ai-studio` (override with `AI_STUDIO_DATA_DIR`), logs at `backend.log` (rotated on next start after exceeding 10 MB). Uses port 2048 by default (or picks a free port and saves it in `config.json`). Supports macOS, Linux (x64 / arm64), and Windows x64.
+- **Uninstalling & cleaning downloaded assets:** Removing the plugin in Magpie (or running `npm run studio:clean` / `npm run backend:clean`) stops the local services and deletes the downloaded `AIStudio2API`, `Camoufox`, and `gemini-web2api-go` binaries and archives.
+  - `auth/`, `gemini.db`, `config.json`, `.env`, and `secrets.json` are kept so reinstalling the plugin or sending a new request restores the backend on demand without signing in or importing cookies again; to prevent automatic reinstallation permanently, remove the sign-in in Magpie or delete `config.json` / `secrets.json`.
+  - Default data directories are shared per OS user; set separate `AI_STUDIO_DATA_DIR` and `GEMINI_WEB_DATA_DIR` when running multiple isolated Magpie config directories so removing the plugin in one does not clean another's shared binaries.
 - Gemini Web takes at most 2 requests at once per sign-in (AI Studio 1), adjustable in Magpie's provider settings. Google account rate-limiting or flagging cannot be completely ruled out—use at your own risk.
 
 **AI Studio upgrades and model selection:**
@@ -181,10 +185,11 @@ When a plugin error asks you to run a command, it includes the full ready-to-cop
 - When stopped, a replaced executable is restored from the verified pinned archive; failed verification never launches the replacement. Existing account and runtime data files are kept, but **an older backend may not read data migrated by a newer version**. Before manually downgrading, stop the service and back up the entire data directory, including `auth/` and `runtime/requests.db`. If data migration is known to have occurred, update the plugin to match the newer backend.
 - The backend's base request deadline and the plugin's default response/stream-idle timeout are 30 minutes (override via `studio.requestTimeoutMs`). This allows long non-streaming replies to finish collecting on the server. A stalled backend can also take 30 minutes to time out, and its single worker can hold up later requests. Cancel a request or choose a shorter timeout when needed; shorter limits can interrupt valid long replies. Upstream controls deadline scaling for automatic continuation.
 - Source review of the locally installed Magpie `v0.1.1285` (`d783df46`) found no shorter fixed total timeout on this plugin's generation path: `Provider.Do → plugin.Fetch → loader.fetch` bypasses the ordinary HTTP transport's 10-minute response-header timeout. The gateway's 5-minute `IdleTimeout` applies between requests. Clients or reverse proxies can impose shorter limits; this was not a 30-minute end-to-end waiting test.
+- Deep Research models (`deep-research-preview-04-2026`, `deep-research-max-preview-04-2026`) are supported: listed when signed in with a Google AI Pro / Ultra account; the backend adapts asynchronous Interactions tasks onto `/v1/chat/completions` streaming (including reasoning and built-in server-side search), without client `tool_call` (Function Calling) or `temperature`.
 - Models with `-build` or `assistant_route` are filtered out: they inject the Build coding assistant's own system prompt and are not offered as ordinary Agent models. Refresh with `magpie provider models ai-studio` after upgrading.
 
 **Gemini Web optimized backend upgrades & rollback:**
-- One-click setup uses [4.27.0-magpie.1](https://github.com/BUKOWSKIREAL/magpie-gemini-web/releases/tag/backend-v4.27.0-magpie.1) by default (built from pinned upstream v4.27.0 with XSRF caching, request cancellation propagation, early tool-call completion, and Argon model verification). Chat requests are timed by silence: while Google queues a request it sends a heartbeat every 60 s, and the backend waits through it for the first words. A try is given up, and retried, only after Google has sent nothing for `request_timeout_sec` (180 s by default), so a long reply that keeps coming is never cut off either.
+- One-click setup uses [4.27.0-magpie.2](https://github.com/BUKOWSKIREAL/magpie-gemini-web/releases/tag/backend-v4.27.0-magpie.2) by default (built from pinned upstream v4.27.0 with XSRF caching, request cancellation propagation, early tool-call completion, and Argon model verification). Chat requests are timed by silence: while Google queues a request it sends a heartbeat every 60 s, and the backend waits through it for the first words. A try is given up, and retried, only after Google has sent nothing for `request_timeout_sec` (180 s by default), so a long reply that keeps coming is never cut off either.
 - **Upgrading:** Running backends are not interrupted automatically. Stop the existing process (PID in `backend.pid` inside the data directory); the next Magpie launch or request will upgrade the binary while preserving config and cookies.
 - **Rollback to upstream / restore optimized build** (run in the plugin directory after stopping the backend):
   ```bash

package/README.md

@@ -21,7 +21,7 @@
 | **图片生成** | Nano Banana 系列模型，供 Magpie 的图片生成使用（需要 Magpie v0.1.2075+，见下文） | 不支持 |
 | **登录方式** | 弹出独立浏览器窗口直接登录 Google | 用无痕窗口登录并复制 `cookie` |
 | **额度卡片** | 显示冷却中的模型与恢复时间 | 同步网页版 **5 小时** 与 **每周** 用量及重置时间 |
-| **本地后端** | AIStudio2API (`v0.2.19`) | gemini-web2api-go (`4.27.0-magpie.1` 优化版) |
+| **本地后端** | AIStudio2API (`v0.3.0`) | gemini-web2api-go (`4.27.0-magpie.2` 优化版) |
 
 ---
 
@@ -152,6 +152,7 @@ magpie plugin options magpie-gemini-web '{"studio":{"proxy":"http://127.0.0.1:78
 | :--- | :--- |
 | `npm run studio:status` | 查看本地服务与账号状态 |
 | `npm run studio:start` / `studio:stop` | 启动 / 停止本地服务 |
+| `npm run studio:clean` | 停止服务并清理已下载的后端与 Camoufox 浏览器（保留账号与配置） |
 | `npm run studio:login` | 添加 Google 账号 |
 | `npm run studio:relogin -- [<账号>]` | 重新登录过期账号（仅一个账号时可省略参数） |
 | `npm run studio:install -- --archive <文件>` | 使用已下载的安装包离线安装 |
@@ -169,6 +170,9 @@ magpie plugin options magpie-gemini-web '{"studio":{"proxy":"http://127.0.0.1:78
 - 两个后端均只监听 `127.0.0.1`，Cookie 与登录态仅保存在本机，Magpie 仅保存本地生成的随机 API Key。
 - **Gemini Web 数据目录**：`~/.local/share/magpie-gemini-web`（可通过 `GEMINI_WEB_DATA_DIR` 修改），日志位于 `backend.log`。支持 macOS、Linux、Windows（x64 / arm64）。
 - **AI Studio 数据目录**：`~/.local/share/magpie-ai-studio`（可通过 `AI_STUDIO_DATA_DIR` 修改），日志位于 `backend.log`（超 10 MB 下次启动自动轮转）。默认使用 2048 端口，被占用时自动顺延并记录在 `config.json`。支持 macOS、Linux（x64 / arm64）与 Windows x64。
+- **卸载与清理下载资产**：在 Magpie 中删除插件（或运行 `npm run studio:clean`、`npm run backend:clean`）会停止本地服务并删除 `AIStudio2API`、`Camoufox` 与 `gemini-web2api-go` 二进制及安装包。
+  - 清理会保留 `auth/`、`gemini.db`、`config.json`、`.env` 与 `secrets.json`，以便重装插件或下次发起请求时自动按需重装后端、免去重新登录或导 Cookie；若要彻底停用自动重装，请在 Magpie 中删除对应登录卡片或一并删除数据目录中的配置文件。
+  - 默认数据目录按当前系统用户共享；若在同一账号下隔离运行多个 Magpie 配置目录，请分别设置 `AI_STUDIO_DATA_DIR` 与 `GEMINI_WEB_DATA_DIR`，避免卸载其一时清理共享资产。
 - Gemini Web 每个登录同时最多 2 个请求（AI Studio 为 1），可在 Magpie 的 provider 设置里调整；仍无法完全排除 Google 账号风控风险，请自行评估使用。
 
 **AI Studio 升级与模型范围：**
@@ -181,10 +185,11 @@ magpie plugin options magpie-gemini-web '{"studio":{"proxy":"http://127.0.0.1:78
 - 已停止但文件被替换的后端会从校验通过的固定版本安装包恢复；校验失败不会启动替换文件。安装会保留账号和运行数据文件，但**不保证旧版本能读取新版迁移过的数据**。手动降级前先停止服务并备份整个数据目录（包括 `auth/`、`runtime/requests.db` 等）；已知发生数据迁移时应升级插件以匹配新版后端。
 - AI Studio 后端基础请求期限为 30 分钟；插件默认单次响应等待／流空闲超时也是 30 分钟（仍可用 `studio.requestTimeoutMs` 调整）。这允许非流式长回复在服务端完整收集；后端卡死时也可能等满 30 分钟，且单 worker 会拖住后续请求。可主动取消请求或设置更短的等待时间，但较短期限可能中断正常长回复。上游自动续写的期限按其规则计算。
 - 已核对本机 Magpie `v0.1.1285`（`d783df46`）源码：插件请求走 `Provider.Do → plugin.Fetch → loader.fetch`，绕过普通 HTTP transport 的 10 分钟响应头限制；此插件生成路径没有发现更短的固定总超时。网关 5 分钟 `IdleTimeout` 仅针对请求之间的空闲连接。客户端或反向代理仍可设置更短期限；这不是一次实等 30 分钟的端到端验证。
+- 支持 Deep Research 模型（`deep-research-preview-04-2026`、`deep-research-max-preview-04-2026`）：需要登录 Google AI Pro / Ultra 账号才会在列表中出现；由后端将异步 Interactions 任务适配到 `/v1/chat/completions` 流式返回（含思考过程与服务端内置联网搜索），不支持客户端 `tool_call`（Function Calling）或 `temperature`。
 - 模型列表过滤 `-build` / `assistant_route` 模型：这类模型附带 Build 写代码助手的系统提示词，不作为普通 Agent 模型提供。升级后可执行 `magpie provider models ai-studio` 刷新列表。
 
 **Gemini Web 优化后端升级与回退：**
-- 插件默认使用本仓库发布的 [4.27.0-magpie.1](https://github.com/BUKOWSKIREAL/magpie-gemini-web/releases/tag/backend-v4.27.0-magpie.1)（基于上游 v4.27.0 集成 XSRF 缓存、请求取消传播、工具调用提前收尾与 Argon 校验补丁）。对话请求按上游沉默计时：Google 让请求排队时每 60 秒发一次心跳，后端会一直等到出字；连续 `request_timeout_sec`（默认 180 秒）没有任何数据才放弃这一次并重试，还在出字的长回复也不会被截断。
+- 插件默认使用本仓库发布的 [4.27.0-magpie.2](https://github.com/BUKOWSKIREAL/magpie-gemini-web/releases/tag/backend-v4.27.0-magpie.2)（基于上游 v4.27.0 集成 XSRF 缓存、请求取消传播、工具调用提前收尾与 Argon 校验补丁）。对话请求按上游沉默计时：Google 让请求排队时每 60 秒发一次心跳，后端会一直等到出字；连续 `request_timeout_sec`（默认 180 秒）没有任何数据才放弃这一次并重试，还在出字的长回复也不会被截断。
 - **旧版升级**：插件不会打断运行中的后端。先停止旧进程（PID 见数据目录 `backend.pid`），下次启动 Magpie 或发请求时会自动更新二进制并保留原有配置与账号。
 - **回退官方版 / 恢复优化版**（停止后端后在插件目录执行）：
   ```bash

package/docs/studio-release.md

@@ -1,6 +1,13 @@
 # AI Studio 托管后端构建
 
-插件使用 `0.2.19-magpie.1`，基于 AIStudio2API `v0.2.19` 的固定 commit。上游版本、commit、Go 版本、C/C++ 工具链与 SpiderMonkey 静态库（`toolchains`）、补丁列表、安装包及二进制 SHA-256 均记录在 `lib/studio-release.json`。`upstream.camoufox` 是这个 commit 要求的 Camoufox 版本（上游 `internal/camoufoxnative/download.go` 里的 `camoufoxRelease`），更新上游 commit 时要一起核对，构建脚本发现不一致会报错。MIT 许可见 `LICENSE-AISTUDIO`。
+插件使用 `0.3.0-magpie.1`，基于 AIStudio2API `v0.3.0` 及其后续修复 commit `48f75fa722b25d22a493e75f6d4e5f9d5035d7e3`。上游版本、commit、Go 版本、C/C++ 工具链与 SpiderMonkey 静态库（`toolchains`）、补丁列表、安装包及二进制 SHA-256 均记录在 `lib/studio-release.json`。`upstream.camoufox` 是这个 commit 要求的 Camoufox 版本（上游 `internal/camoufoxnative/download.go` 里的 `camoufoxRelease`），更新上游 commit 时要一起核对，构建脚本发现不一致会报错。MIT 许可见 `LICENSE-AISTUDIO`。
+
+## 0.3.0 升级兼容
+
+- 基线为上游 `v0.3.0` 加 PR #109（`48f75fa722b25d22a493e75f6d4e5f9d5035d7e3`）。包含账号异步任务流、本机 Chrome / Cookie / CDP 浏览器账号导入、超过 63 字符的工具函数名自动压缩还原、`STRIP_THOUGHT_SIGNATURES`、Responses / Interactions / 音频状态持久化（`runtime/states.db`）、`gemini-live-2.5-flash` 与 `gemini-2.5-pro-tts` 等模型更新，以及上游默认 `WAA_BACKEND=spidermonkey`。Camoufox、SpiderMonkey 静态库和固定工具链沿用上一版。
+- 后台账户任务适配：上游 `POST /api/accounts` 与 `POST /api/accounts/{id}/login` 改为启动不随 HTTP 请求取消的后台任务（返回 `202 Accepted` 与 `{ task }`）。插件与 CLI 通过 `studioLogin` 轮询 `GET /api/accounts/tasks` 等待完成，再从 `GET /api/accounts` 读取账号，并在结束或超时退出时调用 `DELETE /api/accounts/tasks/{id}` 清理或取消后台任务（超时会自动关闭弹出的 Camoufox 登录窗口）。
+- 配置项变更：上游移除了废弃的 `TEMPORARY_CHAT`，新增 `STRIP_THOUGHT_SIGNATURES`（默认 `false`）。插件移除 `TEMPORARY_CHAT` 默认项，保留管理页保存的 `STRIP_THOUGHT_SIGNATURES`，并隔离宿主环境的同名变量。
+- `patches/studio-0.2.19-managed-config.patch` 同步适配上游 `v0.3.0` 新增的 `workerDefaultKeys`（`WARM_WORKER_LIMIT`、`MAX_ACTIVE_WORKERS`）分支：托管构建（`internal/config.managedBy=magpie-gemini-web`）继续完整写出全部非环境变量控制项，避免插件默认并发设置覆盖管理页保存的上游默认值。
 
 ## 0.2.19 升级兼容
 
@@ -29,7 +36,7 @@
 
 上游从 v0.2.14 起原生支持分发方管理更新。构建时通过 `-ldflags` 写入以下值：
 
-- `internal/app.version=v0.2.19-magpie.1`：明确的版本标识。同基线或更旧的官方版、以及更旧的托管版需要升级；更新的官方版、更新的托管版或不认识的定制版仍提示冲突。
+- `internal/app.version=v0.3.0-magpie.1`：明确的版本标识。同基线或更旧的官方版、以及更旧的托管版需要升级；更新的官方版、更新的托管版或不认识的定制版仍提示冲突。
 - `internal/app.managedBy=magpie-gemini-web`：`GET /api/update`（包括 `?refresh=1`）只返回本地版本、`mode: managed` 和 `managed_by`，不访问 GitHub；`POST /api/update` 返回 HTTP 409，不下载、不替换文件、不重启。管理页显示“更新由 magpie-gemini-web 管理”。
 
 `internal/config.managedBy=magpie-gemini-web` 同时启用上面的配置兼容补丁。
@@ -106,7 +113,12 @@ AI_STUDIO_TEST_ARCHIVE=/absolute/path/to/archive.tar.gz npm test
 
 真实安装验证检查旧版请求正文迁移及重启后可读、管理页保存的备用请求设置、启动、版本、GET 更新状态、POST 更新返回 409、二进制及 PID 不变、管理页面显示托管提示、停止与状态。`.github/workflows/studio.yml` 在五种系统/架构的原生 runner 上重新构建、对比 SHA-256 并运行该验证；本机交叉编译不能代替 Linux/Windows 真机验证。
 
-## 0.2.19 升级验证（2026-11-02）
+## 0.3.0 升级验证（2026-11-05）
+
+- 五个平台的候选构建（[run 41976798432](https://github.com/BUKOWSKIREAL/magpie-gemini-web/actions/runs/41976798432)）一次性通过上游测试、真实安装冒烟验证（含 `strip_thought_signatures=false`、`/api/accounts/tasks` 接口与管理页配置持久化）及插件测试；安装包和二进制 pin 来自该次构建。
+- 随后的严格验证构建（[run 41977975873](https://github.com/BUKOWSKIREAL/magpie-gemini-web/actions/runs/41977975873)）在五个平台的全新 runner 上重新构建，哈希与 pin 完全一致，真实安装冒烟验证与插件测试全部通过。
+
+## 0.2.19 升级验证（2026-11-02，历史记录）
 
 - 五个平台的候选构建（[run 41378024737](https://github.com/BUKOWSKIREAL/magpie-gemini-web/actions/runs/41378024737)）通过上游测试、真实安装冒烟验证及插件测试；安装包和二进制 pin 来自该次构建。Windows 首次在安装前的 PowerShell 进程查询超时，失败作业重跑通过，未修改测试或放宽超时。提交 pin 后仍需由严格验证工作流在新 runner 上重建并逐一比对哈希，结果见升级 PR。
 - 本地使用 CI 的 darwin-arm64 安装包运行完整插件测试：212 项通过、无跳过；Python bridge 测试 41 项通过。打包检查包含配置补丁，不包含本地构建和账号数据。
@@ -137,14 +149,14 @@ AI_STUDIO_TEST_ARCHIVE=/absolute/path/to/archive.tar.gz npm test
 - 新版 Google 登录/生成端到端验证尚未完成。
 - 此分支未发布后端 release 或 npm 包。本机日常后端仍使用 `0.2.16-magpie.1`，新版仅在临时数据目录运行。
 
-发布前必须审查各平台 CI 与构建记录，再将安装包及许可证上传到本仓库 `studio-v0.2.19-magpie.1` release。该 release 的安装包可公开下载后，才能发布 npm `0.9.0`，否则新安装会遇到 404。合并 PR 不代表已经满足发布条件；构建脚本及验证 CI 不自动发布 release 或 npm 包。
+发布前必须审查各平台 CI 与构建记录，再将安装包及许可证上传到本仓库 `studio-v0.3.0-magpie.1` release。该 release 的安装包可公开下载后，才能发布对应 npm 版本，否则新安装会遇到 404。合并 PR 不代表已经满足发布条件；构建脚本及验证 CI 不自动发布 release 或 npm 包。
 
 本地试用建议使用独立数据目录：
 
 ```bash
-node scripts/studio.mjs install --data-dir /path/to/studio-019-trial --archive /absolute/path/to/archive
-node scripts/studio.mjs start --data-dir /path/to/studio-019-trial
-node scripts/studio.mjs stop --data-dir /path/to/studio-019-trial
+node scripts/studio.mjs install --data-dir /path/to/studio-030-trial --archive /absolute/path/to/archive
+node scripts/studio.mjs start --data-dir /path/to/studio-030-trial
+node scripts/studio.mjs stop --data-dir /path/to/studio-030-trial
 ```
 
 使用现有账号数据测试前，先停止旧服务并备份；不要让两个进程同时写同一个数据目录。

package/index.mjs

@@ -1,5 +1,6 @@
 import { normalizeBaseURL, settings, credentials, listModels, forward, UNREACHABLE_HEADER, PLUGIN_ICON } from './lib/backend.mjs';
-import { ensureBackend, localBackend } from './lib/native.mjs';
+import { ensureBackend, localBackend, localSetup } from './lib/native.mjs';
+import { registerMagpieCleanup, isPluginRemovedInActiveMagpie } from './lib/cleanup.mjs';
 import { parseCookieInput, importCookie, ensureProxy, validateProxyURL, requireLoginBackend } from './lib/setup.mjs';
 import { usage, poolChoices, poolAccounts, accountEmail } from './lib/usage.mjs';
 import { taskSupported, taskMethod, taskLoader, taskModels, isTaskAuth } from './lib/task.mjs';
@@ -31,14 +32,19 @@ async function signInFor(local, account) {
   return {
     type: 'success',
     key: local.apiKey,
-    metadata: { baseURL: local.baseURL, account: account.id, ...(email ? { email } : {}) }
+    metadata: {
+      baseURL: local.baseURL, account: account.id, ...(email ? { email } : {}),
+      // The account index (/u/N) the email is for; usage takes the email only while it holds.
+      ...(email && account.authuser > 0 ? { authuser: account.authuser } : {})
+    }
   };
 }
 
 // managedBackend is the local backend this plugin set up, when baseURL points at it.
-function managedBackend(baseURL) {
-  const local = localBackend();
-  return local && local.baseURL === baseURL ? local : null;
+function managedBackend(baseURL, key) {
+  if (isPluginRemovedInActiveMagpie()) return null;
+  const local = localBackend() || (key ? localSetup() : null);
+  return local && local.baseURL === baseURL && (!key || local.apiKey === key) ? local : null;
 }
 
 // forwardManaged is forward(), but when the backend this plugin manages can't be reached it
@@ -53,7 +59,7 @@ async function forwardManaged(input, init, getAuth, opts) {
   } catch {
     return res;
   }
-  if (!managedBackend(c.baseURL)) return res;
+  if (!managedBackend(c.baseURL, c.key)) return res;
   try {
     await ensureBackend({ proxy: opts.proxy });
   } catch {
@@ -68,6 +74,7 @@ export const GeminiWebPlugin = async (_input, options = {}) => {
   }
   const { studio: _studio, ...webOptions } = options;
   const opts = settings(webOptions);
+  registerMagpieCleanup(_input);
   const proxyPrompt = {
     type: 'text',
     key: 'proxy',
@@ -222,11 +229,11 @@ export const GeminiWebPlugin = async (_input, options = {}) => {
         } catch (err) {
           return { error: err.message, windows: [], signIn: 'kept' };
         }
-        if (!managedBackend(c.baseURL)) {
+        if (!managedBackend(c.baseURL, c.key)) {
           return { error: '只能读取插件自己管理的本地后端的额度（连接已有后端时读不到）', windows: [], signIn: 'kept' };
         }
-        const { account, email } = auth.metadata || {};
-        return { ...(await usage(Number.isInteger(account) ? { account, email } : {})), signIn: 'kept' };
+        const { account, email, authuser } = auth.metadata || {};
+        return { ...(await usage(Number.isInteger(account) ? { account, email, authuser } : {})), signIn: 'kept' };
       }
     },
     provider: {
@@ -238,7 +245,7 @@ export const GeminiWebPlugin = async (_input, options = {}) => {
         try {
           return await listModels(c, opts);
         } catch (err) {
-          if (!err.unreachable || !managedBackend(c.baseURL)) throw err;
+          if (!err.unreachable || !managedBackend(c.baseURL, c.key)) throw err;
           await ensureBackend({ proxy: opts.proxy });
           return listModels(c, opts);
         }

package/lib/backend-release.json

@@ -1,7 +1,7 @@
 {
-  "version": "4.27.0-magpie.1",
+  "version": "4.27.0-magpie.2",
   "repository": "BUKOWSKIREAL/magpie-gemini-web",
-  "tag": "backend-v4.27.0-magpie.1",
+  "tag": "backend-v4.27.0-magpie.2",
   "upstream": {
     "repository": "https://github.com/zexadev/gemini-web2api-go.git",
     "commit": "e97cfccc7e68e356ac504ba220e0b004382029a8"
@@ -9,28 +9,28 @@
   "goVersion": "1.27.1",
   "binaries": {
     "darwin-arm64": {
-      "asset": "gemini-web2api-go_v4.27.0-magpie.1_darwin_arm64",
-      "sha256": "ffc3c4c23fcf1c5000bae4dc75e6677a9ace067914f67099d361d32ec759d70b"
+      "asset": "gemini-web2api-go_v4.27.0-magpie.2_darwin_arm64",
+      "sha256": "45c01d9a99142957eadf30b75ee17da13ea468ca9b4b0d342b9e7b2b440a9072"
     },
     "darwin-x64": {
-      "asset": "gemini-web2api-go_v4.27.0-magpie.1_darwin_amd64",
-      "sha256": "9e6c907a37d890cab5d89797f32072ff19e9b473c607770205c9f8628f18a177"
+      "asset": "gemini-web2api-go_v4.27.0-magpie.2_darwin_amd64",
+      "sha256": "4739f96c703f08ac8c21a7b699564bb26f80298a0ea8c30b126de501fc61d451"
     },
     "linux-x64": {
-      "asset": "gemini-web2api-go_v4.27.0-magpie.1_linux_amd64",
-      "sha256": "043e4de4672ed3e614cea44268b7c9309ffa87a368b99ac40413e7cd5e833839"
+      "asset": "gemini-web2api-go_v4.27.0-magpie.2_linux_amd64",
+      "sha256": "837d5416507daf628984dd7e32aba69380f54d3385e8fd0b478af9b5a0c3dc1b"
     },
     "linux-arm64": {
-      "asset": "gemini-web2api-go_v4.27.0-magpie.1_linux_arm64",
-      "sha256": "0c243071655d8b54714da7e30a6c30a847a4813b0be674cc9a9fec6a1464bb86"
+      "asset": "gemini-web2api-go_v4.27.0-magpie.2_linux_arm64",
+      "sha256": "8c070298efb4db79839b214c113cc486a4693cc3d144fc17c30daa4b96d9a99b"
     },
     "win32-x64": {
-      "asset": "gemini-web2api-go_v4.27.0-magpie.1_windows_amd64.exe",
-      "sha256": "c6a093ca0bb71246df3f21d7b061ef73064ace274d612acdfbffe704d9ce60b2"
+      "asset": "gemini-web2api-go_v4.27.0-magpie.2_windows_amd64.exe",
+      "sha256": "ec9550617a73aa9e17847dd511edc9b64425fa980a5024ea385d6fd28ccb7d2d"
     },
     "win32-arm64": {
-      "asset": "gemini-web2api-go_v4.27.0-magpie.1_windows_arm64.exe",
-      "sha256": "a4f969c0a4f3e6f21f409f597e3d29344807cf417bcfba645cb646af6e9e3fd3"
+      "asset": "gemini-web2api-go_v4.27.0-magpie.2_windows_arm64.exe",
+      "sha256": "7cb9d2a47fd2106da4db8e5e4983547b962d3e03dd3ed5f0ac529b2eab31f97a"
     }
   }
 }

package/lib/cleanup-detached.cjs

@@ -0,0 +1,61 @@
+'use strict';
+
+process.title = 'magpie-gemini-web-cleanup';
+
+const fs = require('node:fs');
+
+const cfg = JSON.parse(process.env.MAGPIE_CLEANUP_CFG || '{}');
+const {
+  magpieDir,
+  pluginRoot,
+  studioDir,
+  webDir,
+  pidFile,
+  mode = 'watch',
+  pollMs = 2000
+} = cfg;
+
+const sleep = ms => new Promise(resolve => setTimeout(resolve, ms));
+
+function removePidFile() {
+  if (!pidFile) return;
+  try {
+    const recorded = Number(fs.readFileSync(pidFile, 'utf8').trim());
+    if (recorded === process.pid) {
+      fs.rmSync(pidFile, { force: true });
+    }
+  } catch {}
+}
+
+(async () => {
+  try {
+    // Load the cleanup module into memory immediately on startup so the watcher still has all
+    // code in memory even after `bun remove` deletes node_modules/magpie-gemini-web from disk.
+    const {
+      isPluginInMagpie,
+      hasDownloadedAssets,
+      cleanDownloadedAssets
+    } = await import('./cleanup.mjs');
+
+    if (mode === 'immediate') {
+      await cleanDownloadedAssets({ studioDir, webDir });
+      return;
+    }
+
+    for (;;) {
+      const state = isPluginInMagpie(magpieDir, { pluginRoot });
+      if (state === false) {
+        await cleanDownloadedAssets({ studioDir, webDir });
+        return;
+      }
+      if (!hasDownloadedAssets({ studioDir, webDir })) {
+        return;
+      }
+      await sleep(pollMs);
+    }
+  } catch {
+    // Best-effort background cleanup
+  } finally {
+    removePidFile();
+  }
+})();