dsh-provider-pro

dsh-provider-pro

English · 中文

DeepSeek Harness 插件：为自定义供应方（llm-pi-ai 手动添加的 provider）补上 DSH 官方渠道才有的能力，开箱即用、零逐模型配置。

A DeepSeek Harness plugin that gives custom providers (hand-added via the llm-pi-ai provider) four capabilities official channels already have — out of the box, with no per-model configuration.

双版本兼容（0.6.1）：同一代码库同时支撑 DSH Desktop 2.0.15（dsh- 0.1.7-rc.2）* 与最新 0.2.0-rc.2。核验方式：对两版 lib/index.js 做 SHA-256 比对—— dsh-settings、dsh-api-settings-controller、dsh-llm、dsh-credentials、 dsh-attachment、dsh-typert-protocol 逐字节相同，cordis 同为 4.0.4； 仅 dsh-llm-pi-ai（内置供应商表 + anthropic 响应模型记录）与 dsh-config-editor （仅 list() 的 inherited 优化，写路径 edit() 逐字相同）有差异，且都不涉及本插件 使用的接口。因此采用差异收敛而非版本分支：src/compat.ts 是唯一差异缝—— describe 形状归一（宿主裸数组 / RPC namespaces 包装 / 客户端 {ok,value} 信封）、 CAS 重试共用、同时订阅新旧两个 settings 事件名（settings/document-updated 与旧 settings/updated，未发射的事件名天然静默）。未来任一版本演进只改这一个文件。 Dual-version (0.6.1): one codebase serves DSH Desktop 2.0.15 (dsh-* 0.1.7-rc.2) and the latest 0.2.0-rc.2. Verified by SHA-256 over both lib/index.js trees: dsh-settings / dsh-api-settings-controller / dsh-llm / dsh-credentials / dsh-attachment / dsh-typert-protocol are byte-identical and cordis is 4.0.4 in both; only dsh-llm-pi-ai (builtin provider tables, anthropic response-model bookkeeping) and dsh-config-editor (a list()-only inherited optimization; the edit() write path is byte-identical) differ, and neither touches a surface this plugin uses. The strategy is therefore convergence, not version branching: src/compat.ts is the single seam — describe-shape normalization (bare array / namespaces wrapper / {ok,value} envelope), a shared CAS retry, and subscriptions under both settings event names. A future version change touches that one file. 兼容性：需要 DSH Desktop 2.0.15+（0.1.7-rc.2 设置模型：settings.yaml 退役、 per-profile Cordis patch、仅 .volatile() 字段可写）。0.6.0 起逐面核验于 DSH Desktop 2.0.15（cordis 4.0.4、dsh-settings/dsh-llm-pi-ai 0.1.7-rc.2）：describe/mutate 0.1.7 契约（get/section/settings/updated 已移除）、settings.document-updated 事件、settings.section 槽、 ModuleLoader bundle 协议、pi-ai 推理词典语义（reasoningEfforts/compat/ input/contextWindow 均为 schema 声明字段，volatile 子树写入合法）、 llm.stream/discoverModels 的 signal 透传。 插件运行时可变数据（探测 IPC、总开关、UA 覆盖）全部迁入本插件自有设置项 dsh-provider-pro（volatile Schemastery Config）。 Requires DSH Desktop 2.0.15+ (the 0.1.7-rc.2 settings model: settings.yaml retired, per-profile Cordis patches, only .volatile() fields writable). 0.6.0 is verified surface-by-surface against DSH Desktop 2.0.15 (cordis 4.0.4, dsh-settings/dsh-llm-pi-ai 0.1.7-rc.2): the describe/mutate 0.1.7 contract (get/section/settings/updated removed upstream), the settings/document-updated event, the settings.section slot, the ModuleLoader bundle protocol, pi-ai reasoning semantics (reasoningEfforts / compat / input / contextWindow remain schema-declared — writes under the volatile providers dict stay legal), and signal passthrough for llm.stream/discoverModels. All runtime-mutable plugin data (probe IPC, master switch, UA overrides) moved into the plugin's own dsh-provider-pro settings entry (volatile Schemastery Config).

---

中文

功能

1. 自定义 User-Agent — 为每个自定义供应方设置请求 UA（覆盖内置 attribution 头）， 应对按 UA 限流、或禁止非官方终端应用访问的供应方。
2. 推理等级切换 — 与官方渠道一样的思考等级下拉。数据层自动完成： Host 半自动为每个缺少 reasoningEfforts 的自定义模型写入五档词典 （off → null，low/medium/high/max → 同名值）， 对话模型选择器随即出现推理等级行； defaultEffort 不写 → 初始档位为 Default（请求不带思考参数，由供应方 自行决定）。切换发生在聊天里，无需设置。
3. 图片输入声明 — 每个模型卡片内提供「支持图片输入」复选框， 勾选即写入 input: [text, image]，DSH 随即允许向该模型附加图片。
4. 一键全量探测与测活（内置）— 每个模型旁一个「探测」按钮，一次实测三项：
   • 上下文窗口：调 /v1/models 拉取 contextWindow/maxTokens， 缺失字段自动回填（手工值不覆盖）；
   • 角色兼容：pi-ai 的 openai-completions 兼容层默认对推理模型发送 developer 角色（部分上游如 GLM 中转直接拒绝，报 1214「角色信息不正确」）。 探测发极小请求实测；被拒且 system 可用时自动写入 compat.supportsDeveloperRole: false（回退 system），即刻修复；
   • 推理等级：不由探测判断；缺少 reasoningEfforts 的模型由自动补档器 恢复默认五档 off/low/medium/high/max，探测从不覆写该字典；
   • 图片输入 + 延迟：通过 DSH LLM 运行时发送带 1×1 PNG 的真实流式 请求（最后一步，验证修复后的真实配置），报告首 token 延迟与图片接受， 并把结果同步到 input 声明（只增删 image，其他 modality 保留）。 写回（容量回填 + compat + 图片声明）各自在 mutate 前重新读取最新配置， 只写真正变化的字段，互不覆盖；流式探测 30s 预算内无首 token 判为失败， 不会误报存活。wire 检查仅对 openai-completions 路由生效。 供应商卡片头显示 可达/不可达/冷却/未测 徽章，模型行显示对应状态点； 「一键探测」对全部模型走同一全量探测（冷却中的模型跳过，保留上次结果）。

安装

# 方式一：git 安装（需要一次构建许可，见下）
dsh plugin --profile desktop add github:Mortal520/dsh-provider-pro

# 方式二：本地仓库 / tarball（免构建许可）
dsh plugin --profile desktop add ./dsh-provider-pro
dsh plugin --profile desktop add ./dsh-provider-pro-0.5.1.tgz   # 先 pnpm pack

git 安装注意：pnpm ≥10 默认拒绝运行 git 依赖的构建脚本。首次 add 报错时， 把 pnpm 提示的那个包 key 写进 profile 的 pnpm-workspace.yaml：

allowBuilds:
  dsh-provider-pro: true

然后重跑 add。这是对包内代码在你机器上执行的一次性授权——只允许可信来源， 并建议锁定提交（github:Mortal520/dsh-provider-pro#<sha>）。

为避免该步骤，本仓库已提交构建产物 lib/，也可直接用 tarball/本地路径安装。

安装后**重启 dsh web（DSH Desktop 需重启应用）**生效。

用法（设置 → 模型增强）

在「模型」下方新增的原生设置页（与官方页同款令牌样式）：

• 总开关「为全部自定义模型启用推理等级切换」（默认开）：控制功能 2。 关闭时，自动补全的档位会被移除、回到供应方默认；你在别处手动设过的等级不受影响。
• 每个供应方一张 User-Agent 卡：填写后保存，发往该供应方 baseURL 前缀的请求 带上此 UA；留空保存即清除覆盖。

平台红线（第三方插件无法突破）

对话模型选择器内部的档位名是 dsh-llm-pi-ai 硬编码英文 （Off / Minimal / … / Max，reasoningInfo() 在 dsh-llm-pi-ai/lib/index.js 写死）， 第三方插件无法为其改语言，defaultEffort 也不接受中文名值。中文能力只覆盖本插件的 设置页。若需要选择器内中文化，只能改 dsh-llm-pi-ai 本体。

实现原理（给维护者）

• 推理等级（数据层）：完整 reasoningEfforts 词典 → pi-ai resolveModelReasoning → 模型 reasoning = { thinkingLevelMap 全 5 档 } → ModelSelect 的 Effort 行天然出现； defaultEffort 留空 → 选择器自动前置「Default」并默认选中。
• 总开关：本插件自有设置项 dsh-provider-pro 的 autoReasoning （volatile Config 字段，缺省 = 开）。DSH 2.0.15 起设置只认 .volatile() 声明字段，pi-ai 段已是 zod-strict——旧版塞在 llm-pi-ai 用户层里的插件标志一律被剥除。Host 补档器每次扫描前 检查该开关；客户端点开关一次性写开关（自有 entry）+ 移除字节级 相同的自动补档词典（pi-ai entry）。
• UA：pi-ai 会过滤用户 headers 里任意大小写的 user-agent、再追加 attribution 头 （requestHeaders()），改 settings.headers 是死路。Host 半在宿主进程对 globalThis.fetch 做一次 URL 前缀匹配的最小补丁：仅当请求 URL 命中某供应方 baseURL、 且该供应方在本插件自有设置项 userAgents.<route> 配了覆盖值 （DSH 2.0.15 起 pi-ai schema 不再保留 providers.<route>.userAgent 未知字段， UA 存储随 0.6.0 迁入插件自有 entry）时替换（而非追加）User-Agent。
• 区块：settings.section 槽，id: provider-pro、order: 15（官方「模型」=10）。

构建与验证

pnpm install
pnpm run check    # 类型检查 + 构建 + 离线产物校验
pnpm run smoke    # Host 行为冒烟（fetch 补丁 / 补档器 / 总开关 / UA 边界 / 探测消费者）
pnpm run verify   # 上面全部 + 打包（prepack）

lib/ 是发布产物，已提交进仓库（git 安装免重建即可用）。

注：check/smoke 验证的是构建产物与 Host 半逻辑；不覆盖真实 profile 安装、 客户端 UI 渲染或真实网关探测。修改涉及探测或写入路径后，建议对实际网关再跑一次 全量探测，并重启桌面端验证设置页行为。

已知限制

• 选择器内档位名 = dsh-llm-pi-ai 硬编码英文，无法中文化（见「平台红线」）。
• 补档器只在 llm-pi-ai namespace 内、手声明 models[] 的条目生效； 不触碰 modelOverrides 与官方渠道（catalog）。

---

English

Features

1. Custom User-Agent — set a request-level UA per custom provider (overrides DSH's built-in attribution header). Handles providers that rate-limit by UA or reject non-official terminal apps.
2. Reasoning-level switching — the same reasoning picker as the official channels (off / low / medium / high / max). Fully data-driven, no per-model setup: the host fills a five-level reasoningEfforts dictionary for every hand-declared custom model that lacks one and keeps defaultEffort unset, so the picker preselects Default (no thinking parameter is sent; the provider decides). Switch levels any time in chat.
3. Image-input declaration — each model card shows a "Support image input" checkbox; checking it writes input: [text, image] so DSH allows image attachments for that model.
4. One-button full probe & liveness — each model row has a single "Probe" button that measures everything in one pass:
   • Context window: calls /v1/models for contextWindow/maxTokens and backfills missing fields (hand-set values are never overwritten).
   • Message-role compat: pi-ai's openai-completions layer defaults to sending OpenAI's developer role for reasoning-capable models — which some upstreams (GLM behind a relay, error 1214) refuse. A minimal wire request verifies it; when developer is refused while system passes, compat.supportsDeveloperRole: false is written automatically.
   • Reasoning levels: not probed. Models lacking a reasoningEfforts dictionary get the default five-level dict (off / low / medium / high / max) from the auto-fill pass; the probe never overwrites it.
   • Image input + latency: a real stream carrying a 1×1 PNG through DSH's LLM runtime (run last, exercising the exact post-fix config) reports first-token latency and image admission, and syncs the input declaration (adds/removes only image; other declared modalities are preserved). A stream that yields no first token within the 30s budget is an explicit failure, never a false "alive". Each write-back (backfill + compat + image declaration) re-reads the freshest settings right before its mutate and writes only fields that actually changed. Wire checks apply only to openai-completions routes. Provider card headers show an up/down/cooldown/untested badge and each model row a matching status dot, aggregated from the latest results ("Probe all" skips models with an active cooldown and keeps their previous result).

Install

# git install (needs one build allow, see note)
dsh plugin --profile desktop add github:Mortal520/dsh-provider-pro

# or a local checkout / tarball (no build permission needed)
dsh plugin --profile desktop add ./dsh-provider-pro
dsh plugin --profile desktop add ./dsh-provider-pro-0.5.1.tgz   # after pnpm pack

git installs run the package's prepare script. pnpm ≥10 blocks that until you allow it: copy the package key pnpm prints into the profile's pnpm-workspace.yaml (allowBuilds: dsh-provider-pro: true) and re-run add. Only allow packages you trust — and preferably pin a commit (github:Mortal520/dsh-provider-pro#<sha>).

To avoid that step entirely, this repo commits its built lib/, and tarball/local installs need no build permission at all.

Restart dsh web after installing.

Usage (Settings → 模型增强)

A native settings section right below the official 模型 page:

• Master switch "Enable reasoning-level switching for all custom models" (on by default) — controls feature 2. Turning it off strips only the auto-filled dictionaries (byte-identical ones) and leaves any level you set manually untouched.
• One User-Agent card per provider — save to apply the UA to requests whose URL starts with that provider's baseURL; empty + save clears it.

Platform red line

Effort-level names inside the model picker are hardcoded English in dsh-llm-pi-ai and cannot be localized by a third-party plugin; our Chinese labels cover this plugin's own settings section only. In-picker Chinese requires patching dsh-llm-pi-ai itself.

Build & verify

pnpm install
pnpm run check    # typecheck + build + offline artifact validation
pnpm run smoke    # host smoke tests (fetch patch / filler / master switch / UA boundary / probe consumer)
pnpm run verify   # the above plus a tarball pack (prepack)

lib/ is the build artifact and is committed, so a git install works without building.

Note: check/smoke validate the build output and the host-half logic; they do not cover a real profile install, client-UI rendering, or a live gateway probe. After changing the probe or write paths, re-run a full probe against your actual gateway and restart the desktop app to verify the settings page behavior.

Known limitations

• In-picker effort names are English-only (see platform red line).
• The filler only touches hand-declared models[] under the llm-pi-ai namespace; it never touches modelOverrides or catalog models.

---

License

MIT · Changelog