chore: release v0.54.0

- feat(providers): add DeepSeek preset (Anthropic-compatible /anthropic endpoint, V4 Pro + V4 Flash, haiku mapping to Flash for cheaper auxiliary calls)
- feat(providers): add GPT-5.5 to OpenAI OAuth (Codex API) model list and bump fallback default
- refactor(providers): upgrade Xiaomi MiMo (pay-as-you-go + Token Plan) from mimo-v2-pro to mimo-v2.5-pro
- fix(providers): add CLAUDE_CODE_DISABLE_NONSTREAMING_FALLBACK + CLAUDE_CODE_EFFORT_LEVEL to MANAGED_ENV_KEYS and shadow-home strip list to prevent DeepSeek-only env from leaking across providers
- docs(providers): add DeepSeek row to en/zh providers table; sync MiMo-V2.5-Pro + GPT-5.5 references

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: op7418

RELEASE_NOTES.md

@@ -1,30 +1,28 @@
-## CodePilot v0.53.0
+## CodePilot v0.54.0
 
-> 设计 Agent 新增 OpenAI GPT Image 2 图像生成能力，同时给 Nano Banana 和 GPT Image 都加上"第三方兼容 API"选项——有中转代理的用户不用再改代码。配套补齐模型选择、实时模型标识、2K/4K 尺寸支持、第三方连接测试等一圈可见性改进。
+> 本版本补齐服务商生态：新增 DeepSeek 独立预设，OpenAI OAuth 加入 GPT-5.5，小米 MiMo 两个套餐升级到 V2.5-Pro，同时修掉切换服务商时的环境变量残留问题。
 
 ### 新增功能
 
-- **OpenAI GPT Image 2 图像生成** — 设计 Agent 的服务商列表里新增 OpenAI (Image) 一档，支持 GPT Image 2 / 1.5 / 1 / 1-mini 四个模型。GPT Image 2 尺寸按官方规格算真实的像素比：1:1/16:9/9:16/3:2/2:3/4:3/3:4/4:5/5:4/21:9 十个比例 × 1K/2K/4K 三档分辨率，每一档都会尊重你选的比例而不是套用老三档（1024×1024 / 1536×1024 / 1024×1536）。支持参考图（垫图）走 `/v1/images/edits` 端点，连续编辑不再被丢弃
-- **媒体服务商第三方兼容** — Nano Banana 和 GPT Image 都多了"Third-party"预设，只需要填 Base URL 和 Key，其他配置完全照搬官方。现在可以把图像生成走自己架设的 OpenAI 兼容 / Gemini 兼容代理
-- **当前图像模型实时显示** — 聊天里的"图片生成"卡片右上角会显示当前要用的模型和服务商（比如"GPT Image 2 · OpenAI (Image)"）。当默认服务商缺 Key 或被改过类型时会显示琥珀色提示，点一下跳到设置页
-- **"图片生成默认"服务商标记** — 同时配了多个图像服务商的用户，现在可以在设置里一键把某一行设为"用于图片生成"。点某个服务商下的模型胶囊按钮会自动把那一行标为默认
+- **DeepSeek 服务商** — 在服务商列表里新增 DeepSeek 独立预设，走官方 Anthropic 兼容端点 `api.deepseek.com/anthropic`，只需填 Key 即用。默认主模型 DeepSeek V4 Pro，Haiku 档位映射到更便宜的 DeepSeek V4 Flash，压缩/总结这类辅助调用能自动走便宜档
+- **OpenAI OAuth 支持 GPT-5.5** — ChatGPT Plus/Pro 授权登录后，模型下拉里新增 GPT-5.5（排在 GPT-5.4 之上），新会话未指定模型时默认用 GPT-5.5
+
+### 修复问题
+
+- **切换服务商时环境变量残留** — 之前如果用户在系统环境里设过 DeepSeek 文档里的 `CLAUDE_CODE_DISABLE_NONSTREAMING_FALLBACK` / `CLAUDE_CODE_EFFORT_LEVEL`，切到其它服务商后这两个变量仍会带到子进程里，影响其它服务商的请求行为。现在切换服务商时会连同这两个 key 一起清掉，避免跨服务商污染
 
 ### 优化改进
 
-- **设计 Agent 选型更确定** — 原先同时配置多个图像服务商时由数据库行顺序决定用哪个，行为不可控。现在走"显式 providerId → 模型名家族前缀 → 用户设定的默认 → 优先 Gemini 兼容"四步优先级，每一步都可预测
-- **第三方媒体服务商连接测试正确路由** — 之前 OpenAI/Gemini 图像服务商的"测试连接"按钮跑的是 Anthropic `/v1/messages` 探针，对这两个服务永远失败。现在 OpenAI Image 走 `GET /v1/models` + Bearer，Gemini Image 走 `GET /v1beta/models?key=...`，第三方配置也能真实验证
-- **服务商 Base URL 精确匹配加上协议隔离** — 加入 GPT Image 后 `https://api.openai.com/v1` 同时属于多个预设。之前聊天用的 openai-compatible 服务商在这个 URL 下会错误继承 GPT Image 模型目录，现在精确匹配会检查协议一致性
-- **第三方媒体服务商强制要求 Base URL** — 创建和编辑时如果 OpenAI/Gemini Image 的 Base URL 为空会被拦下并提示。之前留空会静默回落到官方端点，让"第三方配置"形同虚设
-- **失效默认图像服务商有清理入口** — 默认图像服务商的行被删除、类型被改成非媒体类型、或 Key 被清空时，服务商设置页会显示一条琥珀色提示 + 清除按钮。后端也会在删除/改类型时自动清理设置，两边兜底
+- **小米 MiMo 升级到 V2.5-Pro** — 按量付费和 Token Plan 两个预设里的默认模型从 `mimo-v2-pro` 全部切到 `mimo-v2.5-pro`，界面上显示名也同步更新为 MiMo-V2.5-Pro
 
 ## 下载地址
 
 ### macOS
-- [Apple Silicon (M1/M2/M3/M4)](https://github.com/op7418/CodePilot/releases/download/v0.53.0/CodePilot-0.53.0-arm64.dmg)
-- [Intel](https://github.com/op7418/CodePilot/releases/download/v0.53.0/CodePilot-0.53.0-x64.dmg)
+- [Apple Silicon (M1/M2/M3/M4)](https://github.com/op7418/CodePilot/releases/download/v0.54.0/CodePilot-0.54.0-arm64.dmg)
+- [Intel](https://github.com/op7418/CodePilot/releases/download/v0.54.0/CodePilot-0.54.0-x64.dmg)
 
 ### Windows
-- [Windows 安装包](https://github.com/op7418/CodePilot/releases/download/v0.53.0/CodePilot.Setup.0.53.0.exe)
+- [Windows 安装包](https://github.com/op7418/CodePilot/releases/download/v0.54.0/CodePilot.Setup.0.54.0.exe)
 
 ## 安装说明
 

apps/site/content/docs/en/providers.mdx

@@ -66,8 +66,9 @@ CodePilot includes built-in configuration presets for major Chinese providers. A
 | **Kimi Coding Plan** | Moonshot Kimi coding edition | Pay-as-you-go |
 | **Moonshot** | Moonshot API | Pay-as-you-go |
 | **MiniMax (Domestic/International)** | MiniMax M2.7 | Token Plan |
+| **DeepSeek** | DeepSeek V4 Pro / V4 Flash (Anthropic-compatible endpoint) | Pay-as-you-go |
 | **火山引擎 Ark** | ByteDance Volcengine (Doubao, GLM, DeepSeek, Kimi) | Coding Plan |
-| **小米 MiMo** | Xiaomi MiMo-V2-Pro (pay-as-you-go or Token Plan) | Pay-as-you-go / Token Plan |
+| **小米 MiMo** | Xiaomi MiMo-V2.5-Pro (pay-as-you-go or Token Plan) | Pay-as-you-go / Token Plan |
 | **阿里云百炼 Coding Plan** | Alibaba Cloud (Qwen, GLM, Kimi, MiniMax) | Coding Plan |
 
 When adding a Chinese provider in CodePilot, the system automatically handles the authentication method — you just need to enter the key provided by the respective platform. Each provider card shows a direct link to obtain your API key.

apps/site/content/docs/zh/providers.mdx

@@ -66,8 +66,9 @@ CodePilot 内置了国内主流服务商的配置预设，选择后自动填充
 | **Kimi Coding Plan** | 月之暗面 Kimi 编程版 | 按量付费 |
 | **Moonshot** | 月之暗面 Moonshot API | 按量付费 |
 | **MiniMax（国内/国际）** | MiniMax M2.7 | Token Plan |
+| **DeepSeek** | DeepSeek V4 Pro / V4 Flash（Anthropic 兼容端点） | 按量付费 |
 | **火山引擎 Ark** | 字节跳动火山引擎（豆包、GLM、DeepSeek、Kimi） | Coding Plan |
-| **小米 MiMo** | 小米 MiMo-V2-Pro（按量付费或 Token Plan） | 按量 / Token Plan |
+| **小米 MiMo** | 小米 MiMo-V2.5-Pro（按量付费或 Token Plan） | 按量 / Token Plan |
 | **阿里云百炼 Coding Plan** | 阿里云（通义、GLM、Kimi、MiniMax） | Coding Plan |
 
 在 CodePilot 中添加国内服务商时，系统会自动处理认证方式，你只需填写对应平台提供的密钥。每个服务商卡片上都有直接获取 API Key 的链接。

package-lock.json

@@ -1,6 +1,6 @@
 {
   "name": "codepilot",
-  "version": "0.53.0",
+  "version": "0.54.0",
   "private": true,
   "license": "BUSL-1.1",
   "workspaces": [

package.json

@@ -9,6 +9,7 @@ import { getOAuthStatus } from '@/lib/openai-oauth-manager';
 // OpenAI models available through ChatGPT Plus/Pro OAuth (Codex API)
 // Reasoning effort defaults to 'medium' server-side (not user-configurable)
 const OPENAI_OAUTH_MODELS = [
+  { value: 'gpt-5.5', label: 'GPT-5.5' },
   { value: 'gpt-5.4', label: 'GPT-5.4' },
   { value: 'gpt-5.4-mini', label: 'GPT-5.4-Mini' },
   { value: 'gpt-5.3-codex', label: 'GPT-5.3-Codex' },
@@ -222,7 +223,7 @@ export async function GET() {
 
       // Legacy: inject ANTHROPIC_MODEL from env overrides if not already present
       // Also check upstreamModelId to avoid duplicates (e.g. catalog has modelId='sonnet'
-      // with upstreamModelId='mimo-v2-pro', and env has ANTHROPIC_MODEL='mimo-v2-pro')
+      // with upstreamModelId='mimo-v2.5-pro', and env has ANTHROPIC_MODEL='mimo-v2.5-pro')
       try {
         const envOverrides = provider.env_overrides_json || provider.extra_env || '{}';
         const envObj = JSON.parse(envOverrides);

src/app/api/providers/models/route.ts

@@ -85,7 +85,7 @@ export function FeatureAnnouncementDialog() {
                 <p><span className="text-foreground font-medium">AI SDK 引擎</span> — 无需安装 CLI，支持所有已配置的服务商</p>
                 <p><span className="text-foreground font-medium">Claude Code 引擎</span> — 通过 CLI 驱动，获得完整的命令行能力</p>
               </div>
-              <p>同时支持 <span className="font-medium">OpenAI 授权登录</span>，ChatGPT Plus/Pro 用户可在服务商设置中登录后直接使用 GPT-5.4 等模型。</p>
+              <p>同时支持 <span className="font-medium">OpenAI 授权登录</span>，ChatGPT Plus/Pro 用户可在服务商设置中登录后直接使用 GPT-5.5 等模型。</p>
             </>
           ) : (
             <>
@@ -94,7 +94,7 @@ export function FeatureAnnouncementDialog() {
                 <p><span className="text-foreground font-medium">AI SDK engine</span> — no CLI needed, works with all configured providers</p>
                 <p><span className="text-foreground font-medium">Claude Code engine</span> — driven by CLI for full command-line capabilities</p>
               </div>
-              <p>Also supports <span className="font-medium">OpenAI OAuth login</span> — ChatGPT Plus/Pro users can sign in under Providers to use GPT-5.4 and more.</p>
+              <p>Also supports <span className="font-medium">OpenAI OAuth login</span> — ChatGPT Plus/Pro users can sign in under Providers to use GPT-5.5 and more.</p>
             </>
           )}
         </div>

src/components/layout/FeatureAnnouncementDialog.tsx

@@ -19,6 +19,7 @@ import Bailian from "@lobehub/icons/es/Bailian";
 import XiaomiMiMo from "@lobehub/icons/es/XiaomiMiMo";
 import Ollama from "@lobehub/icons/es/Ollama";
 import OpenAI from "@lobehub/icons/es/OpenAI";
+import DeepSeek from "@lobehub/icons/es/DeepSeek";
 
 // ---------------------------------------------------------------------------
 // Brand icon resolver
@@ -43,6 +44,7 @@ export function getProviderIcon(name: string, baseUrl: string): ReactNode {
     return <XiaomiMiMo size={18} />;
   if (url.includes("11434") || lower.includes("ollama")) return <Ollama size={18} />;
   if (url.includes("api.openai.com") || lower.includes("openai") || lower.includes("gpt image")) return <OpenAI size={18} />;
+  if (url.includes("deepseek") || lower.includes("deepseek")) return <DeepSeek size={18} />;
   if (lower.includes("bedrock")) return <Bedrock size={18} />;
   if (lower.includes("vertex") || lower.includes("google")) return <Google size={18} />;
   if (lower.includes("aws")) return <Aws size={18} />;
@@ -89,6 +91,7 @@ function resolveIcon(iconKey: string): ReactNode {
     'xiaomi-mimo': <XiaomiMiMo size={18} />,
     ollama: <Ollama size={18} />,
     openai: <OpenAI size={18} />,
+    deepseek: <DeepSeek size={18} />,
     server: <HardDrives size={18} className="text-muted-foreground" />,
   };
   return ICON_MAP[iconKey] || <HardDrives size={18} className="text-muted-foreground" />;

src/components/settings/provider-presets.tsx

@@ -82,6 +82,8 @@ const AUTH_KEYS_TO_STRIP = new Set([
   'ANTHROPIC_DEFAULT_OPUS_MODEL',
   'ANTHROPIC_SMALL_FAST_MODEL',
   'CLAUDE_CODE_SUBAGENT_MODEL',
+  'CLAUDE_CODE_DISABLE_NONSTREAMING_FALLBACK',
+  'CLAUDE_CODE_EFFORT_LEVEL',
   'ANTHROPIC_FOUNDRY_API_KEY',
   // Bedrock / Vertex routing — they switch the provider entirely
   'CLAUDE_CODE_USE_BEDROCK',

src/lib/claude-home-shadow.ts

@@ -530,6 +530,41 @@ export const VENDOR_PRESETS: VendorPreset[] = [
     },
   },
 
+  // ── DeepSeek ──
+  {
+    key: 'deepseek',
+    name: 'DeepSeek',
+    description: 'DeepSeek Anthropic-compatible API — V4 Pro / V4 Flash',
+    descriptionZh: 'DeepSeek Anthropic 兼容 API — V4 Pro / V4 Flash',
+    protocol: 'anthropic',
+    authStyle: 'auth_token',
+    baseUrl: 'https://api.deepseek.com/anthropic',
+    defaultEnvOverrides: {
+      CLAUDE_CODE_SUBAGENT_MODEL: 'deepseek-v4-pro',
+      CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC: '1',
+      CLAUDE_CODE_DISABLE_NONSTREAMING_FALLBACK: '1',
+      CLAUDE_CODE_EFFORT_LEVEL: 'max',
+    },
+    defaultModels: [
+      { modelId: 'sonnet', upstreamModelId: 'deepseek-v4-pro', displayName: 'DeepSeek V4 Pro', role: 'default' },
+      { modelId: 'opus', upstreamModelId: 'deepseek-v4-pro', displayName: 'DeepSeek V4 Pro', role: 'opus' },
+      { modelId: 'haiku', upstreamModelId: 'deepseek-v4-flash', displayName: 'DeepSeek V4 Flash', role: 'haiku' },
+    ],
+    defaultRoleModels: {
+      default: 'deepseek-v4-pro',
+      sonnet: 'deepseek-v4-pro',
+      opus: 'deepseek-v4-pro',
+      haiku: 'deepseek-v4-flash',
+    },
+    fields: ['api_key'],
+    iconKey: 'deepseek',
+    meta: {
+      apiKeyUrl: 'https://platform.deepseek.com/api_keys',
+      docsUrl: 'https://platform.deepseek.com/docs',
+      billingModel: 'pay_as_you_go',
+    },
+  },
+
   // ── Volcengine Ark ──
   {
     key: 'volcengine',
@@ -556,20 +591,20 @@ export const VENDOR_PRESETS: VendorPreset[] = [
   {
     key: 'xiaomi-mimo',
     name: 'Xiaomi MiMo',
-    description: 'Xiaomi MiMo Pay-as-you-go API — MiMo-V2-Pro',
-    descriptionZh: '小米 MiMo 按量付费 — MiMo-V2-Pro',
+    description: 'Xiaomi MiMo Pay-as-you-go API — MiMo-V2.5-Pro',
+    descriptionZh: '小米 MiMo 按量付费 — MiMo-V2.5-Pro',
     protocol: 'anthropic',
     authStyle: 'auth_token',
     baseUrl: 'https://api.xiaomimimo.com/anthropic',
     defaultEnvOverrides: {},
     defaultModels: [
-      { modelId: 'sonnet', upstreamModelId: 'mimo-v2-pro', displayName: 'MiMo-V2-Pro', role: 'default' },
+      { modelId: 'sonnet', upstreamModelId: 'mimo-v2.5-pro', displayName: 'MiMo-V2.5-Pro', role: 'default' },
     ],
     defaultRoleModels: {
-      default: 'mimo-v2-pro',
-      sonnet: 'mimo-v2-pro',
-      opus: 'mimo-v2-pro',
-      haiku: 'mimo-v2-pro',
+      default: 'mimo-v2.5-pro',
+      sonnet: 'mimo-v2.5-pro',
+      opus: 'mimo-v2.5-pro',
+      haiku: 'mimo-v2.5-pro',
     },
     fields: ['api_key'],
     iconKey: 'xiaomi-mimo',
@@ -586,20 +621,20 @@ export const VENDOR_PRESETS: VendorPreset[] = [
   {
     key: 'xiaomi-mimo-token-plan',
     name: 'Xiaomi MiMo Token Plan',
-    description: 'Xiaomi MiMo Token Plan subscription — MiMo-V2-Pro',
-    descriptionZh: '小米 MiMo Token Plan 订阅套餐 — MiMo-V2-Pro',
+    description: 'Xiaomi MiMo Token Plan subscription — MiMo-V2.5-Pro',
+    descriptionZh: '小米 MiMo Token Plan 订阅套餐 — MiMo-V2.5-Pro',
     protocol: 'anthropic',
     authStyle: 'auth_token',
     baseUrl: 'https://token-plan-cn.xiaomimimo.com/anthropic',
     defaultEnvOverrides: {},
     defaultModels: [
-      { modelId: 'sonnet', upstreamModelId: 'mimo-v2-pro', displayName: 'MiMo-V2-Pro', role: 'default' },
+      { modelId: 'sonnet', upstreamModelId: 'mimo-v2.5-pro', displayName: 'MiMo-V2.5-Pro', role: 'default' },
     ],
     defaultRoleModels: {
-      default: 'mimo-v2-pro',
-      sonnet: 'mimo-v2-pro',
-      opus: 'mimo-v2-pro',
-      haiku: 'mimo-v2-pro',
+      default: 'mimo-v2.5-pro',
+      sonnet: 'mimo-v2.5-pro',
+      opus: 'mimo-v2.5-pro',
+      haiku: 'mimo-v2.5-pro',
     },
     fields: ['api_key'],
     iconKey: 'xiaomi-mimo',