dsh-neartoken 0.16.1 → 0.17.0

package/README.md

@@ -5,7 +5,7 @@
   <a href="https://dshfind.com/zh/plugins/huanlinoto/dsh-neartoken"><img src="https://dshfind.com/api/card/huanlinoto/dsh-neartoken?lang=zh" alt="dsh-neartoken card"></a>
 </p>
 
-把 New API 站点（NearToken，`https://llm.pko.svcfusion.com`）变成 DSH 的模型供应商：**在设置页发起设备授权 → 用户在自己浏览器里登录站点并确认 → 插件自动接入 `neartoken` provider**。插件全程不接触密码。模型选择器再把这批模型按 New API 的渠道分组折成 **模型 → 分组 → 推理等级** 三级。
+把 New API 站点（NearToken，`https://neartoken.cc`）变成 DSH 的模型供应商：**在设置页发起设备授权 → 用户在自己浏览器里登录站点并确认 → 插件自动接入 `neartoken` provider**。插件全程不接触密码。模型选择器再把这批模型按 New API 的渠道分组折成 **模型 → 分组 → 推理等级** 三级。
 
 ## 架构
 
@@ -62,7 +62,7 @@
 
 | 字段 | 默认 | 说明 |
 |---|---|---|
-| `baseURL` | `https://llm.pko.svcfusion.com` | 站点根地址，不带尾斜杠 |
+| `baseURL` | `https://neartoken.cc` | 站点根地址，不带尾斜杠 |
 | `apiKeyEnv` | `NEARTOKEN_API_KEY` | 授权写入、adapter 读取的凭据引用名 |
 | `catalogTtlMs` | `300000` | `/api/pricing` 目录缓存时长 |
 | `defaultContextWindow` | `200000` | 没有任何来源能说明某模型上下文窗口时上报的值 |
@@ -85,6 +85,29 @@
         reasoning: [low, medium, high, max]
 ```
 
+### 站点地址迁移（0.17.0）
+
+站点从 `https://llm.pko.svcfusion.com` 迁到 `https://neartoken.cc`，`baseURL` 的默认值随之改变。凭据里存的 key 不变：同一套 New API 数据下，升级后不需要重新登录。
+
+**旧域名要给还在跑旧版插件的客户端继续用，只能做反向代理，不能做 302/307 重定向**：
+
+- **302 把 POST 降级成 GET 并丢掉请求体**：登录（`/api/dsh/device/*`）、出字（`/v1/*`）、生图（`/v1/images/generations`）、充值全部失败。只剩匿名的 `GET /api/pricing` 和 `GET /api/status` 还能过，现场表现是「模型列表能显示，一发消息就报错」。
+- **跨 origin 的重定向会剥掉 `Authorization` 头**（302 / 307 / 308 一样，是 fetch 规范行为）：凡是带 key 的请求落到新域名都成了未认证请求，一律 401。307 只救得回匿名端点，结果更迷惑 —— 登录能成功，之后每个模型都 401。
+
+反向代理保留方法、请求体和请求头，SSE 也不会断：
+
+```nginx
+location / {
+  proxy_pass https://neartoken.cc;
+  proxy_http_version 1.1;
+  proxy_set_header Host $host;
+  proxy_buffering off;        # 不关会把流式出字整段缓冲
+  proxy_read_timeout 600s;    # 生图有 114 秒的先例
+}
+```
+
+旧域名退役前，新版插件照常直连 `https://neartoken.cc`，两者互不影响。
+
 ### 图片压缩
 
 **图片在离开本机前先被重编码**，不是原图直传：请求前把每张图交给 harness 的附件服务（`attachments.readImageRequest`）按 `imageMaxEdge` / `imageMaxBytes` 生成一份请求版，sharp 缩放 + 质量阶梯重编码（有 alpha 走 WebP、无 alpha 走 JPEG），结果按「附件 + 目标尺寸 + 字节预算」缓存，同一张图重复发不重复编码。已经被上游视觉网格丢掉的分辨率不再上传，也就不再计费。
@@ -204,6 +227,7 @@ Gemini 系走原生 surface 而不是 openai 兼容层：同模型同提示词
 ## 已知限制
 
 - 四套 wire（chat / anthropic / genai / responses）各自解析 SSE；**图片按各协议发送** —— chat 用 `image_url` part、anthropic 用 `source.base64`、genai 用 `inlineData`、responses 用 `input_image`。发送前先按 `imageMaxEdge` / `imageMaxBytes` 重编码（见「图片压缩」），所以线上字节与媒体类型未必等于附件里的原图。**工具结果里的图片也发**（`read_image` 的结果就是一段文本信封 + 一张图）：anthropic 把图塞回同一个 `tool_result` 的 content 块数组；genai 在同一 user turn 的 parts 里补 `inlineData`；chat 的 tool 消息只能放字符串、responses 的 `function_call_output` 同理，所以图跟在**整段连续工具结果之后**的一条 user 消息 / 条目里（并行调用时不会把 tool 消息切断）。字节由 `attachments.readImage` 在请求前内联；没有 attachments 服务或读不出来会**让请求失败**，不再静默降级成纯文本（曾经就是这样：贴的截图模型根本收不到）。会话里被 image-offload 标记的图片（`offloaded: true`）不发字节；插件也不替它生成占位文本，所以那一次图片对模型完全不可见。文件块（非图片附件）仍不发；`/model` 弹窗仍是内置平铺列表；箭头键盘导航未接。
+- **genai 线的并行工具调用必须收在同一个 user turn 里**：Gemini 校验「一个 functionCall turn ↔ 紧随其后那条 user turn 里等量的 functionResponse parts」，每条工具结果各开一条 user turn 会被上游拒成 400 `Please ensure that the number of function response parts is equal to the number of function call parts of the function call turn`。它是**确定性**错误：历史不变就必然复现，重试、换分组都无用，只能新开会话（0.16.1 及以前踩中过一次：一轮发了 6 个并行 `read`，整轮就此作废）。0.16.2 起把整段连续的工具结果折进同一个 user turn；Gemini ≥ 3 另给 `functionCall` / `functionResponse` 带上 `id`（同一个函数被并行调多次时才分得清谁答谁），`gemma*` 与更早的 Gemini 不发该字段，只按函数名匹配。
 - 四套 wire 的 SSE 解析都有「帧样本 → 期望 chunk 序列」回归测试（`src/host/wire-parse.spec.ts`），帧样本取自生产网关的真实响应。网关在内容帧上发 `"usage": null`，只有末帧带真值：任何一处把 `usage` 当对象直接读都会让整条流中途崩掉（曾导致 claude 模型全不可用）。
 - **每个请求都带会话亲和键**：chat 与 responses 发 `prompt_cache_key`，anthropic 发 `metadata.user_id`，取值都是 DSH 的 `GenerateOptions.sessionId`。站点就是靠这个值挑上游渠道的（New API 的「codex cli trace」规则：`^gpt-.*$` + `/v1/responses`，键取自 body 的 `prompt_cache_key`，取不到就把整条规则跳过），而提示词缓存按**上游账号**存 —— 不带键时每个请求随机换账号，缓存永远不中。实测同一 6k 前缀：不带键连发 4 次 `cached_tokens` 全 0；带键后第 2 次起 5888/6011 ≈ 98%。`/v1/chat/completions` 上目前没有对应规则，chat 线即使带键也只是偶发命中，要彻底解决需要站点加一条匹配该路径的亲和规则；anthropic 线的 `metadata.user_id` 同理（规则只作用于 `/v1/messages`，这一条已经匹配）。
 - **anthropic 线显式打缓存断点**：Anthropic 只在 `cache_control` 断点处写缓存，没有断点就整段 prompt 按新输入计价（亲和键只保证落在同一上游账号，不产生缓存）。每个请求打 4 个断点：`tools` 末项、`system` 末块、最后两条消息各自末块 —— harness 每轮全量重放历史，尾部双断点保证下一轮几乎全命中，每轮只写新增片段。实测（Claude Max 超惠分组，~6.3k token 前缀）：无断点连发两次 `cache_read` 全 0；打点后第 2 次起 `in=4, cr=6316`，三轮增长历史 `cr` 累积到 10811、`in` 恒为个位数。断点也覆盖 `tool_result` 结尾的轮次（DSH 会话的常态形态）。
@@ -215,7 +239,7 @@ Gemini 系走原生 surface 而不是 openai 兼容层：同模型同提示词
 - `-c` 后缀是站点自己的约定（带该后缀的模型工具调用不可靠）：站点声明了端点类型时插件按声明提供，声明里没有它就不会出现；没有声明时按名字隐藏它的无后缀孪生之外的变体。
 - 站点上新但还没在后台填能力的模型：只要声明了端点类型就会提供，容量走 `defaultContextWindow`，模态按纯文本、没有推理档位（除非在 `modelOverrides` 里写明）。这不是 bug，是「站点是唯一真源」的直接后果——补能力请去后台填，不必等插件发版。
 - pricing 加载失败时沿用上一次成功的结果，从未成功过则目录为空（失败不写缓存，下次调用会重试）。
-- **上游错误映射到共享重试码表**（`src/host/wire-error.ts`）：429→`RATE_LIMIT`、5xx/529/流中途失败→`SERVER`、网络层断连→`TRANSPORT`（fetch 抛错但非用户取消），这三类是 base 组合里 `dsh-llm-retry` 默认策略（5 次、指数退避、读 `retry-after`）会等过去重试的；401/403→`AUTH`、402/余额耗尽→`QUOTA`、400/413→`INVALID_REQUEST`、上下文超限→`CONTEXT_WINDOW_EXCEEDED`、其余→`HTTP_<status>` 保持不可重试。曾经全部抛 `PROVIDER_ERROR`——不在默认可重试集合里，网关一次瞬断就终止整轮（「一遇错误立即停止」的根因）。
+- **上游错误映射到共享重试码表**（`src/host/wire-error.ts`）：429→`RATE_LIMIT`、5xx/529/流中途失败→`SERVER`、网络层断连→`TRANSPORT`（fetch 抛错但非用户取消），这三类是 base 组合里 `dsh-llm-retry` 默认策略（5 次、指数退避、读 `retry-after`）会等过去重试的；401/403→`AUTH`、402/余额耗尽→`QUOTA`、400/413→`INVALID_REQUEST`、上下文超限→`CONTEXT_WINDOW_EXCEEDED`、其余→`HTTP_<status>` 保持不可重试。曾经全部抛 `PROVIDER_ERROR`——不在默认可重试集合里，网关一次瞬断就终止整轮（「一遇错误立即停止」的根因）。识别到上面那句 function response 计数对不上的文案时，错误消息里会直接带一句中文处置提示（码仍是 `INVALID_REQUEST`，不进重试集合）。
 - 账号卡片的钱是**两次读**拼接的：授权那一刻 grant 里的 `user` 给出身份事实，余额/已用/代金券则来自站点 `/api/dsh/device/wallet`——Host 每 5 分钟读一次，页面在余额旁另有**立即刷新**按钮（`POST /neartoken/refresh`，读完后把同一份状态回给页面），所以不必重连也能看到新数字。落盘到 profile 下的 `neartoken-account.json` 只作重启回退。站点若没有该端点或读不到，卡片保留上一次的数字、不会清空。代金券为空时卡片给一句去「邀请计划」（站点 `/wallet` 页）领免费代金券的提示。
 - 白名单按裸模型名过滤：勾掉某模型会把它的所有分组一起隐藏；某个正在使用的模型被勾掉后，会话里已选中的 id 仍在，但选择器里不再列出它。
 - 白名单写入走 DSH 的 Settings：profile 里该行只保留 volatile 字段，所以别把 `baseURL` 等非 volatile 字段写进**同一个 profile 行**（会在此后第一次保存白名单时被 Settings 的 strip 规则清掉）。

package/lib/client.js

@@ -401,23 +401,23 @@ window.__ModuleLoader__.load({
 			document.head.appendChild(tag);
 		}
 		var group_tags_module_css_default = {
-			"toneAmber": "FotuUq_toneAmber",
-			"toneOrange": "FotuUq_toneOrange",
-			"toneEmerald": "FotuUq_toneEmerald",
-			"tonePink": "FotuUq_tonePink",
 			"toneCyan": "FotuUq_toneCyan",
-			"ratioChip": "FotuUq_ratioChip",
+			"toneOrange": "FotuUq_toneOrange",
+			"toneAmber": "FotuUq_toneAmber",
+			"toneRose": "FotuUq_toneRose",
+			"ratioCheap": "FotuUq_ratioCheap",
 			"toneLime": "FotuUq_toneLime",
+			"ratioChip": "FotuUq_ratioChip",
+			"toneEmerald": "FotuUq_toneEmerald",
+			"toneViolet": "FotuUq_toneViolet",
+			"toneIndigo": "FotuUq_toneIndigo",
 			"toneTeal": "FotuUq_toneTeal",
+			"ratioPricey": "FotuUq_ratioPricey",
+			"ratioBase": "FotuUq_ratioBase",
 			"toneSky": "FotuUq_toneSky",
 			"tag": "FotuUq_tag",
-			"ratioCheap": "FotuUq_ratioCheap",
 			"toneFuchsia": "FotuUq_toneFuchsia",
-			"ratioPricey": "FotuUq_ratioPricey",
-			"ratioBase": "FotuUq_ratioBase",
-			"toneRose": "FotuUq_toneRose",
-			"toneViolet": "FotuUq_toneViolet",
-			"toneIndigo": "FotuUq_toneIndigo"
+			"tonePink": "FotuUq_tonePink"
 		};
 		//#endregion
 		//#region src/client/group-tags.tsx
@@ -472,39 +472,39 @@ window.__ModuleLoader__.load({
 			document.head.appendChild(tag);
 		}
 		var ModelGroupSelect_module_css_default = {
-			"selected": "_2pt-bW_selected",
+			"root": "_2pt-bW_root",
+			"warning": "_2pt-bW_warning",
 			"optionTrailing": "_2pt-bW_optionTrailing",
-			"searchRow": "_2pt-bW_searchRow",
-			"retry": "_2pt-bW_retry",
-			"variantCount": "_2pt-bW_variantCount",
 			"cellLabel": "_2pt-bW_cellLabel",
-			"error": "_2pt-bW_error",
-			"searchClear": "_2pt-bW_searchClear",
-			"check": "_2pt-bW_check",
-			"triggerEffort": "_2pt-bW_triggerEffort",
-			"modelOption": "_2pt-bW_modelOption",
+			"cellValue": "_2pt-bW_cellValue",
 			"optionCopy": "_2pt-bW_optionCopy",
-			"chevron": "_2pt-bW_chevron",
-			"groupOptionLabel": "_2pt-bW_groupOptionLabel",
-			"toast": "_2pt-bW_toast",
-			"triggerLabel": "_2pt-bW_triggerLabel",
+			"groups": "_2pt-bW_groups",
+			"groupName": "_2pt-bW_groupName",
+			"searchRow": "_2pt-bW_searchRow",
+			"status": "_2pt-bW_status",
+			"triggerIcon": "_2pt-bW_triggerIcon",
 			"menu": "_2pt-bW_menu",
-			"cell": "_2pt-bW_cell",
+			"searchClear": "_2pt-bW_searchClear",
 			"cellChevron": "_2pt-bW_cellChevron",
-			"status": "_2pt-bW_status",
-			"empty": "_2pt-bW_empty",
-			"search": "_2pt-bW_search",
 			"trigger": "_2pt-bW_trigger",
-			"warning": "_2pt-bW_warning",
-			"modelName": "_2pt-bW_modelName",
-			"groups": "_2pt-bW_groups",
-			"cellValue": "_2pt-bW_cellValue",
-			"triggerIcon": "_2pt-bW_triggerIcon",
-			"optionActive": "_2pt-bW_optionActive",
-			"root": "_2pt-bW_root",
-			"groupName": "_2pt-bW_groupName",
 			"groupInline": "_2pt-bW_groupInline",
-			"option": "_2pt-bW_option"
+			"optionActive": "_2pt-bW_optionActive",
+			"chevron": "_2pt-bW_chevron",
+			"modelName": "_2pt-bW_modelName",
+			"triggerEffort": "_2pt-bW_triggerEffort",
+			"triggerLabel": "_2pt-bW_triggerLabel",
+			"retry": "_2pt-bW_retry",
+			"error": "_2pt-bW_error",
+			"variantCount": "_2pt-bW_variantCount",
+			"selected": "_2pt-bW_selected",
+			"check": "_2pt-bW_check",
+			"search": "_2pt-bW_search",
+			"toast": "_2pt-bW_toast",
+			"groupOptionLabel": "_2pt-bW_groupOptionLabel",
+			"cell": "_2pt-bW_cell",
+			"option": "_2pt-bW_option",
+			"modelOption": "_2pt-bW_modelOption",
+			"empty": "_2pt-bW_empty"
 		};
 		//#endregion
 		//#region src/client/ModelGroupSelect.tsx
@@ -1070,77 +1070,77 @@ window.__ModuleLoader__.load({
 			document.head.appendChild(tag);
 		}
 		var NearTokenSection_module_css_default = {
-			"metaCheck": "_02HI2q_metaCheck",
-			"fact": "_02HI2q_fact",
-			"linkRow": "_02HI2q_linkRow",
-			"search": "_02HI2q_search",
+			"badgeOn": "_02HI2q_badgeOn",
+			"factValue": "_02HI2q_factValue",
+			"notice": "_02HI2q_notice",
+			"tipRow": "_02HI2q_tipRow",
+			"cardToggle": "_02HI2q_cardToggle",
+			"tipGroup": "_02HI2q_tipGroup",
+			"count": "_02HI2q_count",
+			"voucherGroup": "_02HI2q_voucherGroup",
+			"select": "_02HI2q_select",
 			"metaHint": "_02HI2q_metaHint",
+			"list": "_02HI2q_list",
+			"form": "_02HI2q_form",
+			"voucherEmpty": "_02HI2q_voucherEmpty",
+			"topup": "_02HI2q_topup",
 			"facts": "_02HI2q_facts",
+			"fact": "_02HI2q_fact",
+			"badge": "_02HI2q_badge",
+			"codeRow": "_02HI2q_codeRow",
+			"metaActions": "_02HI2q_metaActions",
+			"buttonPrimary": "_02HI2q_buttonPrimary",
+			"linkRow": "_02HI2q_linkRow",
+			"linkLine": "_02HI2q_linkLine",
+			"card": "_02HI2q_card",
+			"rowName": "_02HI2q_rowName",
+			"factLabel": "_02HI2q_factLabel",
+			"metaCheck": "_02HI2q_metaCheck",
+			"intro": "_02HI2q_intro",
 			"label": "_02HI2q_label",
+			"handle": "_02HI2q_handle",
+			"tipGroupName": "_02HI2q_tipGroupName",
+			"search": "_02HI2q_search",
+			"voucherList": "_02HI2q_voucherList",
+			"chevron": "_02HI2q_chevron",
+			"chevronOpen": "_02HI2q_chevronOpen",
+			"topupBar": "_02HI2q_topupBar",
+			"metaBtnOn": "_02HI2q_metaBtnOn",
+			"row": "_02HI2q_row",
+			"voucherAmount": "_02HI2q_voucherAmount",
+			"topupChipActive": "_02HI2q_topupChipActive",
 			"button": "_02HI2q_button",
+			"hint": "_02HI2q_hint",
+			"autosave": "_02HI2q_autosave",
+			"metaEditor": "_02HI2q_metaEditor",
 			"metaIntro": "_02HI2q_metaIntro",
-			"card": "_02HI2q_card",
-			"topupInput": "_02HI2q_topupInput",
-			"cardHead": "_02HI2q_cardHead",
+			"link": "_02HI2q_link",
 			"section": "_02HI2q_section",
-			"valueRow": "_02HI2q_valueRow",
 			"topupChips": "_02HI2q_topupChips",
+			"title": "_02HI2q_title",
+			"topupChip": "_02HI2q_topupChip",
+			"rowMeta": "_02HI2q_rowMeta",
+			"name": "_02HI2q_name",
+			"valueRow": "_02HI2q_valueRow",
 			"toolbar": "_02HI2q_toolbar",
-			"rowName": "_02HI2q_rowName",
-			"count": "_02HI2q_count",
-			"voucherList": "_02HI2q_voucherList",
 			"actions": "_02HI2q_actions",
-			"hint": "_02HI2q_hint",
-			"rowMeta": "_02HI2q_rowMeta",
-			"voucherFact": "_02HI2q_voucherFact",
-			"metaBtnOn": "_02HI2q_metaBtnOn",
-			"linkLine": "_02HI2q_linkLine",
-			"link": "_02HI2q_link",
-			"metaLabel": "_02HI2q_metaLabel",
-			"handle": "_02HI2q_handle",
-			"voucherEmpty": "_02HI2q_voucherEmpty",
+			"tip": "_02HI2q_tip",
 			"code": "_02HI2q_code",
-			"list": "_02HI2q_list",
-			"chevronOpen": "_02HI2q_chevronOpen",
-			"voucherAmount": "_02HI2q_voucherAmount",
-			"topupBar": "_02HI2q_topupBar",
+			"voucherRow": "_02HI2q_voucherRow",
 			"tipTitle": "_02HI2q_tipTitle",
+			"metaTitle": "_02HI2q_metaTitle",
 			"metaField": "_02HI2q_metaField",
-			"row": "_02HI2q_row",
-			"countDot": "_02HI2q_countDot",
-			"buttonPrimary": "_02HI2q_buttonPrimary",
-			"tipGroup": "_02HI2q_tipGroup",
-			"factValue": "_02HI2q_factValue",
-			"invite": "_02HI2q_invite",
-			"badge": "_02HI2q_badge",
-			"topupChip": "_02HI2q_topupChip",
-			"select": "_02HI2q_select",
-			"topupChipActive": "_02HI2q_topupChipActive",
-			"metaInput": "_02HI2q_metaInput",
-			"tip": "_02HI2q_tip",
-			"title": "_02HI2q_title",
-			"metaActions": "_02HI2q_metaActions",
-			"autosave": "_02HI2q_autosave",
-			"voucherRow": "_02HI2q_voucherRow",
-			"notice": "_02HI2q_notice",
-			"intro": "_02HI2q_intro",
-			"rowRatio": "_02HI2q_rowRatio",
-			"form": "_02HI2q_form",
-			"codeRow": "_02HI2q_codeRow",
 			"refresh": "_02HI2q_refresh",
-			"topup": "_02HI2q_topup",
-			"tipGroupName": "_02HI2q_tipGroupName",
-			"tipRow": "_02HI2q_tipRow",
-			"metaEditor": "_02HI2q_metaEditor",
-			"voucherGroup": "_02HI2q_voucherGroup",
-			"metaTitle": "_02HI2q_metaTitle",
 			"countPicked": "_02HI2q_countPicked",
-			"name": "_02HI2q_name",
-			"cardToggle": "_02HI2q_cardToggle",
+			"rowRatio": "_02HI2q_rowRatio",
+			"metaInput": "_02HI2q_metaInput",
+			"invite": "_02HI2q_invite",
+			"topupInput": "_02HI2q_topupInput",
+			"metaLabel": "_02HI2q_metaLabel",
+			"voucherFact": "_02HI2q_voucherFact",
 			"metaBtn": "_02HI2q_metaBtn",
-			"factLabel": "_02HI2q_factLabel",
-			"badgeOn": "_02HI2q_badgeOn",
-			"chevron": "_02HI2q_chevron"
+			"cardHead": "_02HI2q_cardHead",
+			"countDot": "_02HI2q_countDot"
 		};
 		//#endregion
 		//#region src/client/NearTokenSection.tsx

package/lib/index.js

@@ -2016,6 +2016,17 @@ function parseJsonObject(payload) {
 *
 * @module
 */
+/**
+* Wording the gateway uses when a functionCall turn was not answered by the same
+* number of functionResponse parts.
+*
+* A Gemini batch lands here whenever its results were sent one user turn each;
+* no retry clears it, because the conversation that produced the refusal is
+* replayed unchanged.
+*/
+const FUNCTION_RESPONSE_MISMATCH = /number of function response parts/iu;
+/** What the user can do about a refusal the plugin itself cannot repair. */
+const FUNCTION_RESPONSE_MISMATCH_HINT = " 提示：Gemini 原生协议下这批工具结果与调用对不上，重试不会成功；请升级 neartoken 插件，或换用别的模型后新开会话";
 /** The `error` object of a gateway error body, when the body carries one. */
 function errorObjectOf(body) {
 	const parsed = parseJsonObject(body);
@@ -2077,7 +2088,8 @@ function gatewayError(body, status, headers) {
 	const retry = headers?.get("retry-after");
 	const delay = retry == null ? NaN : /^\d+(?:\.\d+)?$/u.test(retry) ? Number(retry) * 1e3 : Date.parse(retry) - Date.now();
 	const id = headers?.get("request-id") ?? headers?.get("x-request-id");
-	return new LlmError(`NearToken answered ${status}: ${(embeddedMessage(body) ?? body).slice(0, 500)}`, code, {
+	const hint = code === "INVALID_REQUEST" && FUNCTION_RESPONSE_MISMATCH.test(detail) ? FUNCTION_RESPONSE_MISMATCH_HINT : "";
+	return new LlmError(`NearToken answered ${status}: ${(embeddedMessage(body) ?? body).slice(0, 500)}${hint}`, code, {
 		status,
 		...id == null ? {} : { requestId: ProviderRequestId(id) },
 		...Number.isFinite(delay) && delay > 0 ? { providerRetryAfterMs: delay } : {}
@@ -2782,9 +2794,11 @@ function toolNamesOf(messages) {
 * Map one harness message onto its genai form.
 * @param message - one conversation message.
 * @param toolNames - call id to function name, for tool results.
+* @param images - the request's inlined images.
+* @param callIds - whether the model matches a response to its call by id.
 * @returns the wire content, or undefined when it carries nothing.
 */
-function serializeMessage$2(message, toolNames, images = /* @__PURE__ */ new Map()) {
+function serializeMessage$2(message, toolNames, images = /* @__PURE__ */ new Map(), callIds = false) {
 	const content = message.content;
 	switch (message.role) {
 		case "user":
@@ -2819,35 +2833,52 @@ function serializeMessage$2(message, toolNames, images = /* @__PURE__ */ new Map
 				} catch {
 					args = {};
 				}
-				parts.push({ functionCall: {
+				const functionCall = {
 					name: call.name,
 					args
-				} });
+				};
+				if (callIds && call.id !== void 0) functionCall.id = String(call.id);
+				parts.push({ functionCall });
 			}
 			return parts.length === 0 ? void 0 : {
 				role: "model",
 				parts
 			};
 		}
-		case "tool": {
-			const text = textOf$2(content);
-			const parts = [{ functionResponse: {
-				name: toolNames.get(String(message.toolCallId)) ?? String(message.toolCallId),
-				response: { result: text }
-			} }];
-			for (const image of imagesIn(content, images)) parts.push({ inlineData: {
-				mimeType: image.mediaType,
-				data: image.data
-			} });
-			return {
-				role: "user",
-				parts
-			};
-		}
+		case "tool": return {
+			role: "user",
+			parts: toolPartsOf(message, toolNames, images, callIds)
+		};
 		default: return;
 	}
 }
 /**
+* The parts one tool result contributes to the user turn that answers a call.
+*
+* A functionResponse carries JSON, not bytes: a picture a tool returned follows
+* as its own inlineData part in the same user turn.
+* @param message - the harness tool message.
+* @param toolNames - call id to function name.
+* @param images - the request's inlined images.
+* @param callIds - whether the model matches a response to its call by id.
+* @returns the parts, starting with the functionResponse.
+*/
+function toolPartsOf(message, toolNames, images, callIds) {
+	const content = message.content;
+	const callId = String(message.toolCallId);
+	const functionResponse = {
+		name: toolNames.get(callId) ?? callId,
+		response: { result: textOf$2(content) }
+	};
+	if (callIds) functionResponse.id = callId;
+	const parts = [{ functionResponse }];
+	for (const image of imagesIn(content, images)) parts.push({ inlineData: {
+		mimeType: image.mediaType,
+		data: image.data
+	} });
+	return parts;
+}
+/**
 * Thinking budget for one harness reasoning effort.
 *
 * genai takes a token budget rather than an effort name. Levels are only
@@ -2880,6 +2911,20 @@ function usesThinkingLevel(model) {
 	return major !== void 0 && Number(major) >= 4;
 }
 /**
+* Whether a model takes an id on its function calls and responses.
+*
+* Gemini 3 added `id` to both, which is what tells two calls of the same
+* function apart inside one batch; earlier generations match on the function
+* name alone and refuse a member they do not declare. Gemma carries no version
+* in its name and stays on the older form.
+* @param model - the bare model id.
+* @returns true for the generations that declare the field.
+*/
+function usesCallIds(model) {
+	const major = /gemini-(\d+)/i.exec(model)?.[1];
+	return major !== void 0 && Number(major) >= 3;
+}
+/**
 * The `thinkingConfig` for one request, in the vocabulary the model speaks.
 * @param model - the bare model id.
 * @param effort - the harness reasoning effort, when one was chosen.
@@ -2906,9 +2951,23 @@ function thinkingConfigOf(model, effort) {
 */
 function serializeRequest$2(options, model, images = /* @__PURE__ */ new Map()) {
 	const toolNames = toolNamesOf(options.messages);
+	const callIds = usesCallIds(model);
 	const contents = [];
+	let batch = null;
 	for (const message of options.messages) {
-		const wire = serializeMessage$2(message, toolNames, images);
+		if (message.role === "tool") {
+			if (batch === null) {
+				batch = [];
+				contents.push({
+					role: "user",
+					parts: batch
+				});
+			}
+			batch.push(...toolPartsOf(message, toolNames, images, callIds));
+			continue;
+		}
+		batch = null;
+		const wire = serializeMessage$2(message, toolNames, images, callIds);
 		if (wire !== void 0) contents.push(wire);
 	}
 	const generationConfig = {};
@@ -4393,7 +4452,7 @@ var PricingDirectory = class {
 	inflight;
 	/**
 	* @param ctx - the owning plugin context (logging only).
-	* @param baseURL - the site root, e.g. `https://llm.pko.svcfusion.com`.
+	* @param baseURL - the site root, e.g. `https://neartoken.cc`.
 	* @param ttlMs - how long one load stays fresh.
 	*/
 	constructor(ctx, baseURL, ttlMs) {
@@ -4961,7 +5020,7 @@ const modelOverride = z.object({
 /** Configuration surface; prototype keeps every knob optional. */
 const Config = z.object({
 	/** Site root without a trailing slash. */
-	baseURL: z.string().default("https://llm.pko.svcfusion.com"),
+	baseURL: z.string().default("https://neartoken.cc"),
 	/** Credential reference the sign-in writes and the adapter resolves. */
 	apiKeyEnv: z.string().default("NEARTOKEN_API_KEY"),
 	/** How long one pricing load stays fresh, in milliseconds. */

package/lib/types/host/pricing.d.ts

@@ -84,7 +84,7 @@ export declare class PricingDirectory {
     private inflight;
     /**
      * @param ctx - the owning plugin context (logging only).
-     * @param baseURL - the site root, e.g. `https://llm.pko.svcfusion.com`.
+     * @param baseURL - the site root, e.g. `https://neartoken.cc`.
      * @param ttlMs - how long one load stays fresh.
      */
     constructor(ctx: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-neartoken",
-  "version": "0.16.1",
+  "version": "0.17.0",
   "publishConfig": {
     "access": "public"
   },
@@ -42,13 +42,6 @@
       "platform": "web"
     }
   },
-  "scripts": {
-    "build": "tsc -p tsconfig.build-host.json && tsc -p tsconfig.build-client.json && tsdown",
-    "bundle": "tsdown",
-    "watch": "tsdown --watch",
-    "typecheck": "tsc -p tsconfig.host.json --noEmit && tsc -p tsconfig.client.json --noEmit",
-    "test": "vitest run"
-  },
   "peerDependencies": {
     "@deepseek-ai/cordis": "~4.0.5-alpha.2",
     "@deepseek-ai/dsh-api-remotes": "^0.2.1-alpha.5",
@@ -173,5 +166,12 @@
     "tsdown": "^0.22.2",
     "typescript": "^5.9.0",
     "vitest": "^4.1.11"
+  },
+  "scripts": {
+    "build": "tsc -p tsconfig.build-host.json && tsc -p tsconfig.build-client.json && tsdown",
+    "bundle": "tsdown",
+    "watch": "tsdown --watch",
+    "typecheck": "tsc -p tsconfig.host.json --noEmit && tsc -p tsconfig.client.json --noEmit",
+    "test": "vitest run"
   }
-}
+}