feat(i18n): drop "(optional)" from Exa/AnySearch key labels; add DeepSeek V4.1 Flash preview

apps/desktop/src/i18n/messages.ts

@@ -651,8 +651,8 @@ const APP_MESSAGES = {
         brave: "Brave Search API key",
         firecrawl: "Firecrawl API key",
         tavily: "Tavily API key",
-        exa: "Exa API key (optional)",
-        anysearch: "AnySearch API key (optional)",
+        exa: "Exa API key",
+        anysearch: "AnySearch API key",
         zhihu: "Zhihu Access Secret",
       },
       envPrefix: "Values starting with ",
@@ -1262,8 +1262,8 @@ const APP_MESSAGES = {
         brave: "Brave Search API Key",
         firecrawl: "Firecrawl API Key",
         tavily: "Tavily API Key",
-        exa: "Exa API Key（可选）",
-        anysearch: "AnySearch API Key（可选）",
+        exa: "Exa API Key",
+        anysearch: "AnySearch API Key",
         zhihu: "知乎 Access Secret",
       },
       envPrefix: "以 ",

packages/runtime/src/models/providers/deepseek.ts

@@ -10,24 +10,68 @@ import { DEEPSEEK_MODELS } from "@earendil-works/pi-ai/providers/deepseek.models
 
 type DeepSeekApi = "openai-completions" | "openai-responses";
 
+/**
+ * DeepSeek V4.1 Flash intermediate-version preview (opened 2026-09-08,
+ * expires 2026-09-10 — the id suffix is the deadline). New architecture with
+ * native multimodal (text + image) input; pricing matches deepseek-v4-flash
+ * and each account is rate-limited to 20 concurrent requests. Announced after
+ * the upstream pi-ai catalog shipped, so it is declared here on the
+ * OpenAI-compatible completions endpoint (the preview has not announced
+ * Responses-API support).
+ */
+const DEEPSEEK_V41_FLASH_PREVIEW: Model<"openai-completions"> = {
+  id: "deepseek-v4.1-flash-expires-on-0910",
+  name: "DeepSeek V4.1 Flash Preview",
+  api: "openai-completions",
+  baseUrl: "https://api.deepseek.com",
+  provider: "deepseek",
+  reasoning: true,
+  input: ["text", "image"],
+  cost: {
+    input: 0.14,
+    output: 0.28,
+    cacheRead: 0.0028,
+    cacheWrite: 0,
+  },
+  contextWindow: 1_000_000,
+  maxTokens: 384_000,
+  compat: {
+    supportsStore: false,
+    supportsDeveloperRole: false,
+    maxTokensField: "max_tokens",
+    requiresReasoningContentOnAssistantMessages: true,
+    thinkingFormat: "deepseek",
+  },
+  thinkingLevelMap: {
+    minimal: null,
+    low: "low",
+    medium: null,
+    high: "high",
+    max: "max",
+  },
+};
+
 export function deepseekProvider(): Provider<DeepSeekApi> {
-  const models = Object.values(DEEPSEEK_MODELS).map((model) => {
-    if (
-      model.id !== "deepseek-v4-flash" &&
-      model.id !== "deepseek-v4-pro"
-    ) {
-      return model;
-    }
-    return {
-      ...model,
-      api: "openai-responses",
-      compat: {
-        supportsDeveloperRole: false,
-        supportsLongCacheRetention: false,
-        sessionAffinityFormat: "openai-nosession",
-      },
-    } satisfies Model<"openai-responses">;
-  });
+  const models = [
+    ...Object.values(DEEPSEEK_MODELS).map((model) => {
+      if (
+        model.id !== "deepseek-v4-flash" &&
+        model.id !== "deepseek-v4-pro"
+      ) {
+        return model;
+      }
+      return {
+        ...model,
+        api: "openai-responses",
+        compat: {
+          supportsDeveloperRole: false,
+          supportsLongCacheRetention: false,
+          sessionAffinityFormat: "openai-nosession",
+        },
+      } satisfies Model<"openai-responses">;
+    }),
+    DEEPSEEK_V41_FLASH_PREVIEW,
+  ];
 
   return createProvider<DeepSeekApi>({
     id: "deepseek",

packages/runtime/tests/models/providers/deepseek/deepseek.test.ts

@@ -0,0 +1,55 @@
+import { describe, expect, test } from "bun:test";
+
+import { deepseekProvider } from "../../../../src/models/providers/deepseek";
+
+describe("deepseekProvider", () => {
+  test("exposes the DeepSeek endpoint with the upstream models plus the V4.1 Flash preview", () => {
+    const provider = deepseekProvider();
+    const models = provider.getModels();
+
+    expect(provider.id).toBe("deepseek");
+    expect(provider.baseUrl).toBe("https://api.deepseek.com");
+
+    const ids = models.map((model) => model.id);
+    expect(ids).toContain("deepseek-v4-flash");
+    expect(ids).toContain("deepseek-v4-pro");
+    expect(ids).toContain("deepseek-v4.1-flash-expires-on-0910");
+  });
+
+  test("the V4.1 Flash preview is multimodal, 1M-context, and priced like V4 Flash", () => {
+    const provider = deepseekProvider();
+    const preview = provider
+      .getModels()
+      .find((model) => model.id === "deepseek-v4.1-flash-expires-on-0910");
+    expect(preview).toBeDefined();
+    expect(preview?.api).toBe("openai-completions");
+    expect(preview?.baseUrl).toBe("https://api.deepseek.com");
+    expect(preview?.provider).toBe("deepseek");
+    // Native multimodal input (new V4.1 architecture).
+    expect(preview?.input.map(String)).toEqual(["text", "image"]);
+    expect(preview?.reasoning).toBe(true);
+    expect(preview?.contextWindow).toBe(1_000_000);
+    // Same billing as deepseek-v4-flash.
+    expect(preview?.cost).toEqual({
+      input: 0.14,
+      output: 0.28,
+      cacheRead: 0.0028,
+      cacheWrite: 0,
+    });
+    expect(preview?.compat).toMatchObject({
+      maxTokensField: "max_tokens",
+      requiresReasoningContentOnAssistantMessages: true,
+      thinkingFormat: "deepseek",
+    });
+  });
+
+  test("keeps the Responses-API override for V4 Flash/Pro", () => {
+    const provider = deepseekProvider();
+    const models = provider.getModels();
+    const flash = models.find((model) => model.id === "deepseek-v4-flash");
+    const pro = models.find((model) => model.id === "deepseek-v4-pro");
+    expect(flash?.api).toBe("openai-responses");
+    expect(pro?.api).toBe("openai-responses");
+    expect(flash?.contextWindow).toBe(1_000_000);
+  });
+});

specs/pr-plans/0005-remove-optional-suffix.md

@@ -0,0 +1,66 @@
+# PR 计划 0005：删掉 Exa / AnySearch API key 标签里的 "optional"
+> 状态：**已实现（2026-09-08 核对）**：`messages.ts` 中 `Exa API key (optional)` / `AnySearch API key (optional)` 的 `(optional)` 已删（8 棵语言树同步）。**与 DeepSeek V4.1 Flash 内测模型（见下）合并为一个 PR 交付。** 改动在工作树，未提交。
+>
+> **追加：DeepSeek V4.1 Flash 内测模型（并入本 PR）**：`packages/runtime/src/models/providers/deepseek.ts` 新增 `deepseek-v4.1-flash-expires-on-0910`（id 后缀即截止日：2026-09-08 开、09-10 止）。原生多模态（text+image）、走 openai-completions（内测未宣布 Responses 支持）、上下文 1M / maxTokens 384k、定价同 v4-flash（in 0.14 / out 0.28 / cacheRead 0.0028）、每账号 20 并发、`thinkingFormat: "deepseek"` + thinkingLevelMap。测试 `deepseek.test.ts` 通过。
+
+> 状态：计划（未写代码）
+> 体量：极小（1 个文件，4 行文案）。适合作为独立小 PR 快速合入。
+
+## 1. 背景与目标
+
+搜索设置页里，Exa 与 AnySearch 的 API key 输入框标签写作「Exa API key (optional)」/「AnySearch API key（可选）」。两个 key 确实可选（无 key 也能匿名调用），但其他 provider（Brave / Firecrawl / Tavily / Zhihu）其实也都是"填了才能用"，标签风格却不统一——带括号后缀既不美观，也容易让人误以为只有这两个是可选的。
+
+**目标**：去掉后缀，标签统一为「Exa API key」/「AnySearch API key」。
+
+## 2. 现状（代码事实）
+
+"optional" **不是组件 prop**，而是硬编码在 i18n 消息树 `search.keys` 里：
+
+- 英文 `apps/desktop/src/i18n/messages.ts`
+  - 654：`exa: "Exa API key (optional)"`
+  - 655：`anysearch: "AnySearch API key (optional)"`
+- 中文 `apps/desktop/src/i18n/messages.ts`
+  - 1265：`exa: "Exa API Key（可选）"`
+  - 1266：`anysearch: "AnySearch API Key（可选）"`
+
+`apps/desktop/src/components/settings/search-page.tsx` 的 `ApiKeyField` 没有 `optional` 之类的 prop（label 取自 `t.search.keys[provider]`，126），所以**改文案即可，组件不用动**。
+
+## 3. 改动
+
+```diff
+  // en
+- exa: "Exa API key (optional)",
+- anysearch: "AnySearch API key (optional)",
++ exa: "Exa API key",
++ anysearch: "AnySearch API key",
+
+  // zh
+- exa: "Exa API Key（可选）",
+- anysearch: "AnySearch API Key（可选）",
++ exa: "Exa API Key",
++ anysearch: "AnySearch API Key",
+```
+
+**en / zh 必须同时改**，否则中英文案不一致（现有测试会强制两棵树结构一致，但不会检查文案是否同步，所以这一步靠手工核对）。
+
+## 4. 影响面
+
+- **无功能影响**：`ApiKeyField` 没有必填校验，搜索 key 全部可选；页面下方的 `keyNotes`（en 662 / zh 1273）已经说明了 Exa / AnySearch 可以无 key 使用，信息不会丢失。
+- **无结构变化**：消息树 leaf 数量不变，`messages.test.ts` 的结构一致性测试自动通过。
+
+## 5. 涉及文件
+
+- `apps/desktop/src/i18n/messages.ts`（4 行）
+
+## 6. 验证
+
+- `bun test apps/desktop/src/i18n/messages.test.ts`
+- 手动打开 Settings → Search，确认两个标签显示正常、说明文案仍在。
+
+## 7. 工作量
+
+极小（< 30 分钟，含验证）。
+
+## 8. 备注
+
+这是 PR #163（Exa / AnySearch / Zhihu provider）的收尾清理。可以在 #163 的 review 里顺带提一句，也可以单独开 PR——**单独开更干净**（reviewer 一眼能看完，合入快）。