magpie-gemini-web 0.1.2 → 0.1.3

package/README.md

@@ -140,6 +140,20 @@ http://127.0.0.1:8083/admin
 | `requestTimeoutMs`| integer | `240000` | 转发单次请求的超时时间（毫秒，默认 4 分钟） |
 | `modelTimeoutMs` | integer | `10000` | 获取模型列表的超时时间（毫秒，默认 10 秒） |
 
+### 思考档位（effort）
+
+后端没有 effort 参数，扩展思考是单独的 `-thinking` 模型。插件把有 `-thinking` 版的模型声明为 `low` / `high` 两档，可以在 Magpie 里直接切换：
+
+| 档位 | 实际调用 | 说明 |
+| :--- | :--- | :--- |
+| `low`（或不指定） | `gemini-x` | 网页端默认档：Argon 为「低」档，Flash 不输出思考链，3.1 Pro 仍会少量思考 |
+| `high` | `gemini-x-thinking` | 网页端扩展思考（上游显示为 High），带 `reasoning_content` |
+
+- 插件读取请求里的 `reasoning_effort`：`none` / `minimal` / `low` 走普通版，其他值（如 `medium`、`xhigh`）走 `-thinking` 版。经 Magpie 调用时，Magpie 会先把 Responses API 的 `reasoning.effort`、Anthropic Messages 的 `thinking.budget_tokens` 换算成最接近的已声明档位；Anthropic 的 `output_config.effort` 不会被转发。
+- 直接选 `-thinking` 模型照常可用，档位设置对它不生效。
+- 网页端 Argon 的「中」档上游尚未支持（见 [#5](https://github.com/BUKOWSKIREAL/magpie-gemini-web/issues/5)）。
+- 升级插件后，执行 `magpie provider models gemini-web` 重新拉取模型列表，档位才会出现。
+
 ---
 
 ## 5. 本地开发与实验性 Argon 支持 (Local Development & Argon Support)

package/lib/backend.mjs

@@ -4,6 +4,14 @@ const MEDIA_SPECIAL = /(?:^|[-_])(image|music|canvas|video|research)(?:[-_]|$)/i
 const CONTROL_CHARS = /[\x00-\x1f\x7f]/;
 const NEWLINE_CHARS = /[\r\n]/;
 
+// The backend has no effort parameter: extended thinking is a separate "<id>-thinking" model.
+// A base model with such a twin offers low (itself) and high (the twin). Magpie maps any other
+// level to the nearest declared one before it reaches us; direct callers get the same rule.
+const THINKING_SUFFIX = '-thinking';
+const LOW_EFFORTS = new Set(['none', 'minimal', 'low']);
+const MODEL_CACHE_MS = 10 * 60 * 1000;
+const knownModels = new Map();
+
 const KNOWN_OPTIONS = new Set([
   'baseURL',
   'allowTools',
@@ -253,8 +261,12 @@ export async function listModels(c, opts) {
     throw new Error('No usable Gemini models found in backend response');
   }
 
+  const ids = new Set(validIds);
+  knownModels.set(modelCacheKey(c), { ids, at: Date.now() });
+
   const result = Object.create(null);
   for (const id of validIds) {
+    const hasTwin = !id.endsWith(THINKING_SUFFIX) && ids.has(id + THINKING_SUFFIX);
     result[id] = {
       id,
       providerID: 'gemini-web',
@@ -291,7 +303,7 @@ export async function listModels(c, opts) {
       },
       options: {},
       headers: {},
-      variants: {},
+      variants: hasTwin ? { low: {}, high: {} } : {},
       status: 'active'
     };
   }
@@ -299,6 +311,32 @@ export async function listModels(c, opts) {
   return result;
 }
 
+function modelCacheKey(c) {
+  return `${c.baseURL}\n${c.key}`;
+}
+
+// thinkingTwin returns the model a chat request should run on for its reasoning_effort, or null
+// to leave it alone. The twin is only used when the backend lists it; if the list can't be
+// fetched the request goes out unchanged.
+async function thinkingTwin(body, c, opts) {
+  const effort = body.reasoning_effort;
+  const model = body.model;
+  if (typeof effort !== 'string' || LOW_EFFORTS.has(effort)) return null;
+  if (typeof model !== 'string' || model.endsWith(THINKING_SUFFIX)) return null;
+
+  let cached = knownModels.get(modelCacheKey(c));
+  if (!cached || Date.now() - cached.at > MODEL_CACHE_MS) {
+    try {
+      await listModels(c, opts);
+    } catch {
+      return null;
+    }
+    cached = knownModels.get(modelCacheKey(c));
+  }
+  const twin = model + THINKING_SUFFIX;
+  return cached?.ids.has(twin) ? twin : null;
+}
+
 export async function forward(input, init, getAuth, opts) {
   const req = new Request(input, init);
 
@@ -417,6 +455,12 @@ export async function forward(input, init, getAuth, opts) {
         }
       }
     }
+
+    const twin = await thinkingTwin(body, c, opts);
+    if (twin) {
+      body.model = twin;
+      rawBody = JSON.stringify(body);
+    }
   }
 
   const timeoutSignal = AbortSignal.timeout(opts.requestTimeoutMs);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "magpie-gemini-web",
-  "version": "0.1.2",
+  "version": "0.1.3",
   "description": "Gemini Web provider for Magpie, backed by gemini-web2api-go",
   "type": "module",
   "main": "./index.mjs",