@ailuntz/minilink 0.9.0 → 0.9.3

package/README.md

@@ -1,46 +1,58 @@
-# minilink
+# MiniLink
 
-在 ChatGPT / Codex Desktop 的原生模型菜单中使用已有 profiles。
+在 ChatGPT / Codex Desktop 的原生模型菜单中使用自定义模型。
 
-读取 `~/.codex/*.config.toml`：有 `model_catalog_json` 就使用该目录，否则继承原生模型列表。支持 API key 的 Responses 连接还会追加 `-Chat`，用于不带工具和工作区上下文的纯文字聊天。
+## 定位与支持周期
+
+官方 [Profiles](https://learn.chatgpt.com/docs/config-file/config-advanced?site_locale=en#profiles) 已支持独立配置文件：`codex --profile deep-review` 会在 `~/.codex/config.toml` 上叠加 `~/.codex/deep-review.config.toml`。配置使用顶层字段，无需嵌套在 `[profiles.deep-review]` 中。
+
+**截至 2026-09-24，仅添加 Profile 不会将额外供应商的模型加入 Desktop 菜单。MiniLink 读取这些配置，补充桌面端的模型选择能力。**
+
+MiniLink 定位为过渡方案：在官方提供等效的原生菜单支持前维护这一能力；官方支持完善后，优先迁移至原生方案，结束对应兼容功能的维护。兼容性以已验证的 Desktop 版本为准。
 
 ## 安装
 
-macOS，Node.js ≥22.16。
+需 Node.js ≥22.16，并已安装 Desktop。
 
-```sh
-npm install -g @ailuntz/minilink
-```
+| 平台 | 接入状态 |
+| --- | --- |
+| macOS | 支持 npm 自动安装 |
+| Windows / Linux | 已完成手动接入验证，npm 自动安装尚未支持 |
 
-完全退出并重开 Desktop，然后开启：
+macOS：
 
 ```sh
+npm install -g @ailuntz/minilink
 minilink on
 ```
 
+完全退出并重开 Desktop。
+
+## 命令
+
 | 命令 | 用途 |
 | --- | --- |
-| `on` | 开启并选择模型和推理强度 |
-| `off` | 关闭追加选项和模型接管 |
-| `doctor` | 检查配置、版本和运行状态 |
+| `minilink on` | 开启 |
+| `minilink link` | 选择模型与推理强度 |
+| `minilink off` | 关闭 |
+| `minilink doctor` | 检查配置与运行状态 |
+
+通过 `minilink link` 保存选择后，完全退出并重开 Desktop。
+
+## 配置
 
-`on` 直接在终端选择模型和推理强度，用于下一次发送消息的任务；无需网页或选择任务。额度提示挡住菜单时同样可用。选择持续到执行 `off` 或退出 Desktop。
+读取 `~/.codex/*.config.toml`。通过 `model_catalog_json` 指定模型目录；未指定时继承原生目录。密钥使用提供商配置的环境变量，也可保存在 `~/.codex/.env`。
 
-界面跟随终端语言（中文 / English）。支持 SSH 连接到运行 Desktop 的电脑；请使用 `ssh -t`。不带终端时只开启路由，不弹出选择。没有运行 Desktop 的服务器无法控制桌面模型菜单。
+支持 Responses API。符合条件的连接提供 `-Chat` 选项，用于纯文字聊天。
 
-启动时后台检查更新，每天最多一次。有新版便静默下载安装，断网或失败则继续使用旧版。新版及 profiles、连接、密钥的修改在重启 Desktop 后生效。`MINILINK_NO_UPDATE_CHECK=1` 可关闭自动更新。
+供应商示例：[DeepSeek / 豆包 / Sub2API](examples/providers/)。修改配置后重启 Desktop。
 
 ## 卸载
 
+macOS：
+
 ```sh
 npm uninstall -g @ailuntz/minilink
 ```
 
-约两分钟后自动移除接入，重启 Desktop 生效。保留用户配置和运行中任务所需的文件。
-
-## 注意
-
-- 跨连接切换需要任务已保存且空闲；临时任务、尚无历史或含子代理的任务请新建。
-- Chat 仅支持纯文字历史，不支持附件、中途追加和上下文压缩。
-- 切换模型保留当前任务权限。
-- 密钥使用提供商声明的环境变量，也可放在 `~/.codex/.env`。支持 `CODEX_HOME`。
+卸载后完全退出并重开 Desktop。

package/bin/minilink.mjs

@@ -9,7 +9,7 @@ import { selectModel, text, language } from '../src/terminal.mjs';
 import { installation, environmentStatus } from '../src/install.mjs';
 
 const self = fileURLToPath(import.meta.url), args = process.argv.slice(2);
-const commands = ['on', 'off', 'doctor'];
+const commands = ['on', 'link', 'off', 'doctor'];
 const output = value => console.log(typeof value === 'string' ? value : JSON.stringify(value, null, 2));
 
 async function main() {
@@ -21,7 +21,11 @@ async function main() {
   if (command === 'on' || command === 'off') {
     if (command === 'on') discoverProfiles();
     atomicJSON(configPath(), { version: 1, enabled: command === 'on' });
-    return output(command === 'off' ? text('off') : await selectModel(pickerEndpoints(codexHome())));
+    return output(text(command));
+  }
+  if (command === 'link') {
+    if (!readConfig().enabled) throw new Error(text('disabled'));
+    return output(await selectModel(pickerEndpoints(codexHome())));
   }
   if (command === 'doctor') {
     const profiles = discoverProfiles(), env = childEnvironment(profiles);

package/examples/providers/.env.example

@@ -0,0 +1,3 @@
+DEEPSEEK_API_KEY=
+ARK_API_KEY=
+SUB2API_API_KEY=

package/examples/providers/README.md

@@ -0,0 +1,13 @@
+# 供应商配置
+
+将所需配置放入 `~/.codex/`，按 `.env.example` 将密钥补充到 `~/.codex/.env`，然后重启 Desktop。DeepSeek 和豆包使用独立模型目录；Sub2API 继承 Desktop 的原生模型目录，只需一个配置文件。
+
+| 供应商 | 模型 |
+| --- | --- |
+| DeepSeek | V4.1 Flash、V4 Pro |
+| 豆包 | Seed 2.1 Pro、Seed 2.1 Lite（260915） |
+| Sub2API | 继承原生模型目录 |
+
+MiniLink 的 Chat 模式仅支持文字。
+
+模型核对：2026-09-24。[DeepSeek](https://api-docs.deepseek.com/quick_start/pricing/) · [豆包](https://docs.volcengine.com/docs/ark/model-list?lang=zh)

package/examples/providers/deepseek-models.json

@@ -0,0 +1,83 @@
+{
+  "models": [
+    {
+      "slug": "deepseek-flash",
+      "display_name": "DeepSeek V4.1 Flash",
+      "description": "DeepSeek V4.1 Flash",
+      "default_reasoning_level": "low",
+      "supported_reasoning_levels": [
+        {
+          "effort": "low",
+          "description": "low"
+        },
+        {
+          "effort": "high",
+          "description": "high"
+        },
+        {
+          "effort": "max",
+          "description": "max"
+        }
+      ],
+      "shell_type": "unified_exec",
+      "visibility": "list",
+      "supported_in_api": true,
+      "priority": 0,
+      "upgrade": null,
+      "base_instructions": "You are a helpful coding assistant. Follow the user instructions, inspect the workspace when needed, and explain results clearly.",
+      "support_verbosity": false,
+      "supports_reasoning_summaries": false,
+      "default_reasoning_summary": "none",
+      "supports_parallel_tool_calls": false,
+      "truncation_policy": {
+        "mode": "bytes",
+        "limit": 10000
+      },
+      "experimental_supported_tools": [],
+      "context_window": 1000000,
+      "input_modalities": [
+        "text",
+        "image"
+      ]
+    },
+    {
+      "slug": "deepseek-v4-pro",
+      "display_name": "DeepSeek V4 Pro",
+      "description": "DeepSeek V4 Pro",
+      "default_reasoning_level": "low",
+      "supported_reasoning_levels": [
+        {
+          "effort": "low",
+          "description": "low"
+        },
+        {
+          "effort": "high",
+          "description": "high"
+        },
+        {
+          "effort": "max",
+          "description": "max"
+        }
+      ],
+      "shell_type": "unified_exec",
+      "visibility": "list",
+      "supported_in_api": true,
+      "priority": 0,
+      "upgrade": null,
+      "base_instructions": "You are a helpful coding assistant. Follow the user instructions, inspect the workspace when needed, and explain results clearly.",
+      "support_verbosity": false,
+      "supports_reasoning_summaries": false,
+      "default_reasoning_summary": "none",
+      "supports_parallel_tool_calls": false,
+      "truncation_policy": {
+        "mode": "bytes",
+        "limit": 10000
+      },
+      "experimental_supported_tools": [],
+      "context_window": 1000000,
+      "input_modalities": [
+        "text"
+      ]
+    }
+  ]
+}

package/examples/providers/deepseek.config.toml

@@ -0,0 +1,9 @@
+model_provider = "deepseek"
+model_catalog_json = "deepseek-models.json"
+
+[model_providers.deepseek]
+name = "DeepSeek"
+base_url = "https://api.deepseek.com"
+env_key = "DEEPSEEK_API_KEY"
+wire_api = "responses"
+requires_openai_auth = false

package/examples/providers/doubao-models.json

@@ -0,0 +1,68 @@
+{
+  "models": [
+    {
+      "slug": "doubao-seed-2-1-pro-260915",
+      "display_name": "Doubao Seed 2.1 Pro",
+      "description": "Doubao Seed 2.1 Pro",
+      "default_reasoning_level": "low",
+      "supported_reasoning_levels": [
+        {
+          "effort": "low",
+          "description": "low"
+        }
+      ],
+      "shell_type": "unified_exec",
+      "visibility": "list",
+      "supported_in_api": true,
+      "priority": 0,
+      "upgrade": null,
+      "base_instructions": "You are a helpful coding assistant. Follow the user instructions, inspect the workspace when needed, and explain results clearly.",
+      "support_verbosity": false,
+      "supports_reasoning_summaries": false,
+      "default_reasoning_summary": "none",
+      "supports_parallel_tool_calls": false,
+      "truncation_policy": {
+        "mode": "bytes",
+        "limit": 10000
+      },
+      "experimental_supported_tools": [],
+      "context_window": 1048576,
+      "input_modalities": [
+        "text",
+        "image"
+      ]
+    },
+    {
+      "slug": "doubao-seed-2-1-lite-260915",
+      "display_name": "Doubao Seed 2.1 Lite",
+      "description": "Doubao Seed 2.1 Lite",
+      "default_reasoning_level": "low",
+      "supported_reasoning_levels": [
+        {
+          "effort": "low",
+          "description": "low"
+        }
+      ],
+      "shell_type": "unified_exec",
+      "visibility": "list",
+      "supported_in_api": true,
+      "priority": 0,
+      "upgrade": null,
+      "base_instructions": "You are a helpful coding assistant. Follow the user instructions, inspect the workspace when needed, and explain results clearly.",
+      "support_verbosity": false,
+      "supports_reasoning_summaries": false,
+      "default_reasoning_summary": "none",
+      "supports_parallel_tool_calls": false,
+      "truncation_policy": {
+        "mode": "bytes",
+        "limit": 10000
+      },
+      "experimental_supported_tools": [],
+      "context_window": 1048576,
+      "input_modalities": [
+        "text",
+        "image"
+      ]
+    }
+  ]
+}

package/examples/providers/doubao.config.toml

@@ -0,0 +1,9 @@
+model_provider = "doubao"
+model_catalog_json = "doubao-models.json"
+
+[model_providers.doubao]
+name = "Doubao"
+base_url = "https://ark.cn-beijing.volces.com/api/v3"
+env_key = "ARK_API_KEY"
+wire_api = "responses"
+requires_openai_auth = false

package/examples/providers/sub2api.config.toml

@@ -0,0 +1,8 @@
+model_provider = "sub2api"
+
+[model_providers.sub2api]
+name = "Sub2API"
+base_url = "https://www.966gravity.com"
+env_key = "SUB2API_API_KEY"
+wire_api = "responses"
+requires_openai_auth = false

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ailuntz/minilink",
-  "version": "0.9.0",
+  "version": "0.9.3",
   "description": "Use your Codex profiles in the Desktop model menu.",
   "type": "module",
   "license": "MIT AND Apache-2.0",
@@ -13,6 +13,7 @@
   "files": [
     "bin/",
     "src/",
+    "examples/",
     "README.md",
     "LICENSE"
   ],

package/src/picker.mjs

@@ -10,8 +10,8 @@ import { additionalModels } from './model-menu.mjs';
 // Keep the selected model when Desktop substitutes its reserve model at quota.
 // A pending choice belongs to the next user task that starts or sends a message.
 export class Picker {
-  constructor({ bridge, getConfig, serialize, chats = [] } = {}) {
-    Object.assign(this, { bridge, getConfig, serialize, chats });
+  constructor({ bridge, getConfig, serialize, configRpc, chats = [] } = {}) {
+    Object.assign(this, { bridge, getConfig, serialize, configRpc, chats });
     this.pins = new Map(); this.pending = new Map(); this.next = null;
   }
   request(message) {
@@ -52,7 +52,8 @@ export class Picker {
   async state() {
     this.request(null);
     const models = await this.catalog();
-    return { next: this.next, models: models.map(m => ({ model: m.model, name: m.displayName || m.model,
+    const { config } = await this.configRpc('config/read', { includeLayers: false });
+    return { next: this.next, selected: { model: config.model, effort: config.model_reasoning_effort }, models: models.map(m => ({ model: m.model, name: m.displayName || m.model,
       effort: m.defaultReasoningEffort, efforts: m.supportedReasoningEfforts || [] })) };
   }
   async choose({ model, effort }) {
@@ -62,6 +63,13 @@ export class Picker {
     const supported = entry.supportedReasoningEfforts || [];
     effort = supported.length ? (effort ?? entry.defaultReasoningEffort) : null;
     if (supported.length && !supported.some(e => e.reasoningEffort === effort)) throw new Error(text('invalidEffort'));
+    await this.configRpc('config/batchWrite', {
+      edits: [
+        { keyPath: 'model', value: model, mergeStrategy: 'upsert' },
+        { keyPath: 'model_reasoning_effort', value: effort, mergeStrategy: 'upsert' }
+      ],
+      reloadUserConfig: true
+    });
     this.next = { model, effort };
     return { ok: true };
   }

package/src/proxy.mjs

@@ -1,5 +1,6 @@
 import { setting } from './legacy.mjs';
 import { runtimeDirectory } from './runtime.mjs';
+import { randomUUID } from 'node:crypto';
 import { spawn } from 'node:child_process';
 import { join } from 'node:path';
 import { rmSync, existsSync, readFileSync } from 'node:fs';
@@ -102,7 +103,19 @@ export async function runProxy(args, selfPath) {
         catch { route.error = `Profile ${route.profile} 的模型目录无法加载，请检查 model_catalog_json。`; }
       }
     };
-    Object.assign(picker, { bridge: switches, getConfig, serialize, refreshCatalogs });
+    const configRpc = async (method, params) => {
+      // Use the same alias translation and saved provider identity as Desktop.
+      const id = `minilink-picker-${randomUUID()}`;
+      const request = menu.request({ id, method, params });
+      try {
+        const result = await switches.rpc(method, request.params);
+        return menu.response({ id, result }).result;
+      } catch (error) {
+        menu.response({ id, error: { message: error.message } });
+        throw error;
+      }
+    };
+    Object.assign(picker, { bridge: switches, getConfig, serialize, refreshCatalogs, configRpc });
     for (const chat of chats) Object.assign(chat, { bridge: switches, menu });
     present = async message => {
       picker.response(message);

package/src/terminal.mjs

@@ -2,18 +2,19 @@ import { createInterface } from 'node:readline/promises';
 
 export const language = (env = process.env) => /^zh(?:[_-]|$)/i.test(env.LC_ALL || env.LC_MESSAGES || env.LANG || '') ? 'zh' : 'en';
 const messages = {
-  help: ['minilink\n\n  on       开启并选择模型\n  off      关闭额外选项和模型接管\n  doctor   检查配置和运行状态\n\n读取现有 profiles。其他 Codex 参数原样透传。', 'minilink\n\n  on       Enable and select a model\n  off      Disable extra models and overrides\n  doctor   Check configuration and status\n\nUses existing profiles. Other arguments are forwarded to Codex.'],
+  help: ['minilink\n\n  on       开启 MiniLink\n  link     选择模型和推理强度\n  off      关闭额外选项和模型接管\n  doctor   检查配置和运行状态\n\n读取现有 profiles。其他 Codex 参数原样透传。', 'minilink\n\n  on       Enable MiniLink\n  link     Select a model and reasoning effort\n  off      Disable extra models and overrides\n  doctor   Check configuration and status\n\nUses existing profiles. Other arguments are forwarded to Codex.'],
+  on: ['minilink on 已开启，请完全退出并重开 Desktop。', 'MiniLink is enabled. Fully quit and reopen Desktop.'],
   off: ['已关闭。已有任务保留当前提供商。', 'Disabled. Existing tasks keep their current provider.'],
-  unavailable: ['已开启。请在运行 Desktop 的电脑上使用；完全退出并重开 Desktop 后，再运行 minilink on。', 'Enabled. Run this on the computer hosting Desktop. Quit and reopen Desktop, then run minilink on again.'],
-  interactive: ['已开启。请在交互终端运行 minilink on；SSH 请使用 ssh -t。', 'Enabled. Run minilink on in an interactive terminal; use ssh -t over SSH.'],
+  unavailable: ['未连接到 Desktop。请在运行 Desktop 的电脑上完全退出并重开 Desktop，然后运行 minilink link。', 'No Desktop connection. On the computer hosting Desktop, fully quit and reopen Desktop, then run minilink link.'],
+  interactive: ['请在交互终端运行 minilink link；SSH 请使用 ssh -t。', 'Run minilink link in an interactive terminal; use ssh -t over SSH.'],
   multiple: ['检测到多个 Desktop 连接。请只保留一个连接后重试。', 'Multiple Desktop connections are running. Keep one connection open and try again.'],
   model: ['模型', 'Model'], effort: ['推理强度', 'Reasoning effort'],
   prompt: ['输入编号', 'Enter a number'], invalid: ['请输入列表中的编号。', 'Enter a number from the list.'],
   cancelled: ['已取消，未更改模型。', 'Cancelled. Model unchanged.'],
-  done: ['已选择。回到 Desktop 发送下一条消息即可生效。', 'Selected. Return to Desktop and send your next message to apply.'],
+  done: ['已保存到 config.toml。请完全退出并重开 Desktop，让模型菜单加载已保存的选择。', 'Saved to config.toml. Fully quit and reopen Desktop to load the saved selection in the model menu.'],
   empty: ['暂无可用模型，请检查连接后重试。', 'No models available. Check your connection and try again.'],
-  disconnected: ['连接已结束，请重新运行 minilink on。', 'Connection closed. Run minilink on again.'],
-  expired: ['连接已失效，请重新运行 minilink on。', 'Connection expired. Run minilink on again.'],
+  disconnected: ['连接已结束，请重新运行 minilink link。', 'Connection closed. Run minilink link again.'],
+  expired: ['连接已失效，请重新运行 minilink link。', 'Connection expired. Run minilink link again.'],
   catalogLarge: ['模型目录过大，请检查配置。', 'Model catalog is too large. Check your configuration.'],
   disabled: ['请先运行 minilink on。', 'Run minilink on first.'],
   missingModel: ['模型已不在目录中，请重新选择。', 'Model is no longer available. Select again.'],
@@ -44,6 +45,7 @@ export async function selectModel(endpoints, { input = process.stdin, output = p
   };
   const state = await api('/state');
   if (!state.models.length) throw new Error(t('empty'));
+  const preferred = state.next ?? state.selected;
   const rl = createInterface({ input, output });
   const abort = new AbortController();
   rl.on('SIGINT', () => abort.abort());
@@ -62,10 +64,10 @@ export async function selectModel(endpoints, { input = process.stdin, output = p
     throw new DOMException('Cancelled', 'AbortError');
   };
   try {
-    const model = await select(t('model'), state.models.map(m => ({ value: m.model, label: m.name })), state.next?.model);
+    const model = await select(t('model'), state.models.map(m => ({ value: m.model, label: m.name })), preferred?.model);
     const entry = state.models.find(m => m.model === model);
     const efforts = entry.efforts || [];
-    const effort = efforts.length > 1 ? await select(t('effort'), efforts.map(e => ({ value: e.reasoningEffort, label: t(e.reasoningEffort) })), state.next?.model === model ? state.next.effort : entry.effort) : efforts[0]?.reasoningEffort ?? null;
+    const effort = efforts.length > 1 ? await select(t('effort'), efforts.map(e => ({ value: e.reasoningEffort, label: t(e.reasoningEffort) })), preferred?.model === model ? preferred.effort : entry.effort) : efforts[0]?.reasoningEffort ?? null;
     if (abort.signal.aborted) return t('cancelled');
     await api('/choose', { model, effort });
     return t('done');