@henjicc/ai-sdk 0.2.8 → 0.3.0

package/examples/llm-chat/README.md

@@ -1,32 +1,32 @@
-# llm-chat
-
-用 `@henjicc/ai-sdk@0.2.8` 的 `llm/streaming` 按需入口走 OpenAI-compatible SSE 流式对话，
-不会把 generation、BigModel preset/models/pricing、Groq 或 LLM modules 打进示例 bundle；
-通用端点身份解析所需的 BigModel profiles 仍会保留。默认目标是
-DeepSeek `deepseek-v4-flash`，提示词要求只回复 `SDK OK`，并限制最多 16 tokens。示例显式传入
-`reasoning: { enabled: false, effort: 'high' }`，避免短输出预算被供应商默认思考过程占用；若流结束后
-没有任何文本 token，示例会明确失败。
-
-```bash
-npm install
-npm run dry-run
-npm run check:bundle
-```
-
-`dry-run` 使用本地 SSE fixture，不访问网络。真实调用前先确认供应商、模型、Base URL 和价格：
-
-```bash
-LLM_API_KEY='你的密钥' \
-LLM_PROVIDER_ID='deepseek' \
-LLM_MODEL_ID='deepseek-v4-flash' \
-LLM_BASE_URL='https://api.deepseek.com' \
-npm start
-```
-
-DeepSeek Flash 当前资料价为缓存未命中输入 ¥1.5/百万 tokens（闲时）或 ¥3/百万（高峰），输出
-¥4.5/百万（闲时）或 ¥9/百万（高峰）。示例按高峰价、并把字符数保守当 token 数输出粗估；实际账单
-以供应商 usage 和账单为准。
-
-示例对 `/chat/completions` 的 POST 设置单次外网闸门，进程内第二次付费请求会在本地被拒绝。
-要取消流式请求，保存 `resolveLlmTaskId(request)` 返回的 `taskId`，再调用
-`cancelLlmChatTask(taskId)`。按需流式入口没有需要宿主释放的客户端实例。
+# llm-chat
+
+用 `@henjicc/ai-sdk@0.3.0` 的 `llm/streaming` 按需入口走 OpenAI-compatible SSE 流式对话，
+不会把 generation、BigModel preset/models/pricing、Groq 或 LLM modules 打进示例 bundle；
+通用端点身份解析所需的 BigModel profiles 仍会保留。默认目标是
+DeepSeek `deepseek-v4-flash`，提示词要求只回复 `SDK OK`，并限制最多 16 tokens。示例显式传入
+`reasoning: { enabled: false, effort: 'high' }`，避免短输出预算被供应商默认思考过程占用；若流结束后
+没有任何文本 token，示例会明确失败。
+
+```bash
+npm install
+npm run dry-run
+npm run check:bundle
+```
+
+`dry-run` 使用本地 SSE fixture，不访问网络。真实调用前先确认供应商、模型、Base URL 和价格：
+
+```bash
+LLM_API_KEY='你的密钥' \
+LLM_PROVIDER_ID='deepseek' \
+LLM_MODEL_ID='deepseek-v4-flash' \
+LLM_BASE_URL='https://api.deepseek.com' \
+npm start
+```
+
+DeepSeek Flash 当前资料价为缓存未命中输入 ¥1.5/百万 tokens（闲时）或 ¥3/百万（高峰），输出
+¥4.5/百万（闲时）或 ¥9/百万（高峰）。示例按高峰价、并把字符数保守当 token 数输出粗估；实际账单
+以供应商 usage 和账单为准。
+
+示例对 `/chat/completions` 的 POST 设置单次外网闸门，进程内第二次付费请求会在本地被拒绝。
+要取消流式请求，保存 `resolveLlmTaskId(request)` 返回的 `taskId`，再调用
+`cancelLlmChatTask(taskId)`。按需流式入口没有需要宿主释放的客户端实例。

package/examples/llm-chat/index.ts

@@ -1,100 +1,100 @@
-import {
-  resolveLlmTaskId,
-  runLlmChatStream,
-  type LlmChatRequestDto,
-  type MediaReader,
-  type RuntimeContext,
-  type Transport,
-} from '@henjicc/ai-sdk/llm/streaming'
-
-const live = process.argv.includes('--live')
-const providerId = process.env.LLM_PROVIDER_ID?.trim() || 'deepseek'
-const modelId = process.env.LLM_MODEL_ID?.trim() || 'deepseek-v4-flash'
-const baseUrl = process.env.LLM_BASE_URL?.trim() || 'https://api.deepseek.com'
-
-class SingleChatTransport implements Transport {
-  private paidRequests = 0
-
-  async fetch(url: string, init?: RequestInit): Promise<Response> {
-    if ((init?.method ?? 'GET').toUpperCase() === 'POST' && new URL(url).pathname.endsWith('/chat/completions')) {
-      this.paidRequests += 1
-      if (this.paidRequests > 1) throw new Error('Blocked a repeated paid LLM request')
-    }
-    return await fetch(url, init)
-  }
-}
-
-class DryRunTransport implements Transport {
-  calls = 0
-
-  async fetch(): Promise<Response> {
-    this.calls += 1
-    return new Response(
-      'data: {"choices":[{"delta":{"content":"SDK OK"}}]}\n\ndata: [DONE]\n\n',
-      { status: 200, headers: { 'Content-Type': 'text/event-stream' } },
-    )
-  }
-}
-
-const unusedMedia: MediaReader = {
-  async read() {
-    throw new Error('This text-only example does not read media')
-  },
-}
-
-async function main(): Promise<void> {
-  const dryTransport = new DryRunTransport()
-  const apiKey = live ? process.env.LLM_API_KEY?.trim() : 'dry-run-not-a-secret'
-  if (!apiKey) throw new Error('Missing LLM_API_KEY')
-
-  const runtime: RuntimeContext = {
-    transport: live ? new SingleChatTransport() : dryTransport,
-    credentials: {
-      get: (scope, requestedProviderId) => scope === 'llm' && requestedProviderId === providerId
-        ? apiKey
-        : undefined,
-    },
-    media: unusedMedia,
-  }
-  const prompt = '只回复两个英文单词：SDK OK'
-  const request: LlmChatRequestDto = {
-    requestId: live ? `example-llm-${Date.now()}` : 'example-llm-dry-run',
-    providerId,
-    modelId,
-    adapter: providerId === 'deepseek' ? 'deepseek' : 'openai',
-    baseUrl,
-    messages: [{ role: 'user', content: prompt }],
-    capabilities: { reasoning: providerId === 'deepseek' },
-    reasoning: providerId === 'deepseek'
-      ? { enabled: false, effort: 'high' }
-      : undefined,
-    policy: { max_tokens: 16 },
-  }
-  const taskId = resolveLlmTaskId(request)
-  let output = ''
-
-  const outcome = await runLlmChatStream(request, taskId, (event) => {
-    if (event.type === 'Token') output += event.data
-  }, runtime)
-
-  if (!output.trim()) {
-    throw new Error('LLM completed without a text token')
-  }
-
-  const conservativeCny = ((outcome.inputChars * 3) + (outcome.outputChars * 9)) / 1_000_000
-  console.log(JSON.stringify({
-    mode: live ? 'live' : 'dry-run',
-    networkCalls: live ? 1 : 0,
-    interceptedRequests: live ? undefined : dryTransport.calls,
-    providerId,
-    modelId,
-    output,
-    elapsedMs: outcome.elapsedMs,
-    conservativeCnyEstimate: Number(conservativeCny.toFixed(6)),
-  }))
-}
-
-void main().catch((error: unknown) => {
-  console.error(error instanceof Error ? error.message : String(error))
-  process.exitCode = 1
-})
+import {
+  resolveLlmTaskId,
+  runLlmChatStream,
+  type LlmChatRequestDto,
+  type MediaReader,
+  type RuntimeContext,
+  type Transport,
+} from '@henjicc/ai-sdk/llm/streaming'
+
+const live = process.argv.includes('--live')
+const providerId = process.env.LLM_PROVIDER_ID?.trim() || 'deepseek'
+const modelId = process.env.LLM_MODEL_ID?.trim() || 'deepseek-v4-flash'
+const baseUrl = process.env.LLM_BASE_URL?.trim() || 'https://api.deepseek.com'
+
+class SingleChatTransport implements Transport {
+  private paidRequests = 0
+
+  async fetch(url: string, init?: RequestInit): Promise<Response> {
+    if ((init?.method ?? 'GET').toUpperCase() === 'POST' && new URL(url).pathname.endsWith('/chat/completions')) {
+      this.paidRequests += 1
+      if (this.paidRequests > 1) throw new Error('Blocked a repeated paid LLM request')
+    }
+    return await fetch(url, init)
+  }
+}
+
+class DryRunTransport implements Transport {
+  calls = 0
+
+  async fetch(): Promise<Response> {
+    this.calls += 1
+    return new Response(
+      'data: {"choices":[{"delta":{"content":"SDK OK"}}]}\n\ndata: [DONE]\n\n',
+      { status: 200, headers: { 'Content-Type': 'text/event-stream' } },
+    )
+  }
+}
+
+const unusedMedia: MediaReader = {
+  async read() {
+    throw new Error('This text-only example does not read media')
+  },
+}
+
+async function main(): Promise<void> {
+  const dryTransport = new DryRunTransport()
+  const apiKey = live ? process.env.LLM_API_KEY?.trim() : 'dry-run-not-a-secret'
+  if (!apiKey) throw new Error('Missing LLM_API_KEY')
+
+  const runtime: RuntimeContext = {
+    transport: live ? new SingleChatTransport() : dryTransport,
+    credentials: {
+      get: (scope, requestedProviderId) => scope === 'llm' && requestedProviderId === providerId
+        ? apiKey
+        : undefined,
+    },
+    media: unusedMedia,
+  }
+  const prompt = '只回复两个英文单词：SDK OK'
+  const request: LlmChatRequestDto = {
+    requestId: live ? `example-llm-${Date.now()}` : 'example-llm-dry-run',
+    providerId,
+    modelId,
+    adapter: providerId === 'deepseek' ? 'deepseek' : 'openai',
+    baseUrl,
+    messages: [{ role: 'user', content: prompt }],
+    capabilities: { reasoning: providerId === 'deepseek' },
+    reasoning: providerId === 'deepseek'
+      ? { enabled: false, effort: 'high' }
+      : undefined,
+    policy: { max_tokens: 16 },
+  }
+  const taskId = resolveLlmTaskId(request)
+  let output = ''
+
+  const outcome = await runLlmChatStream(request, taskId, (event) => {
+    if (event.type === 'Token') output += event.data
+  }, runtime)
+
+  if (!output.trim()) {
+    throw new Error('LLM completed without a text token')
+  }
+
+  const conservativeCny = ((outcome.inputChars * 3) + (outcome.outputChars * 9)) / 1_000_000
+  console.log(JSON.stringify({
+    mode: live ? 'live' : 'dry-run',
+    networkCalls: live ? 1 : 0,
+    interceptedRequests: live ? undefined : dryTransport.calls,
+    providerId,
+    modelId,
+    output,
+    elapsedMs: outcome.elapsedMs,
+    conservativeCnyEstimate: Number(conservativeCny.toFixed(6)),
+  }))
+}
+
+void main().catch((error: unknown) => {
+  console.error(error instanceof Error ? error.message : String(error))
+  process.exitCode = 1
+})

package/examples/llm-chat/package.json

@@ -1,19 +1,19 @@
-{
-  "name": "@henjicc/ai-sdk-example-llm-chat",
-  "private": true,
-  "type": "module",
-  "scripts": {
-    "build": "tsc -p tsconfig.json",
-    "dry-run": "npm run build && node dist/index.js --dry-run",
-    "check:bundle": "node scripts/check-bundle.mjs",
-    "start": "npm run build && node dist/index.js --live"
-  },
-  "dependencies": {
-    "@henjicc/ai-sdk": "0.2.8"
-  },
-  "devDependencies": {
-    "@types/node": "^24.0.0",
-    "esbuild": "^0.25.9",
-    "typescript": "^5.9.2"
-  }
-}
+{
+  "name": "@henjicc/ai-sdk-example-llm-chat",
+  "private": true,
+  "type": "module",
+  "scripts": {
+    "build": "tsc -p tsconfig.json",
+    "dry-run": "npm run build && node dist/index.js --dry-run",
+    "check:bundle": "node scripts/check-bundle.mjs",
+    "start": "npm run build && node dist/index.js --live"
+  },
+  "dependencies": {
+    "@henjicc/ai-sdk": "0.3.0"
+  },
+  "devDependencies": {
+    "@types/node": "^24.0.0",
+    "esbuild": "^0.25.9",
+    "typescript": "^5.9.2"
+  }
+}

package/examples/llm-chat/tsconfig.json

@@ -1,14 +1,14 @@
-{
-  "compilerOptions": {
-    "target": "ES2022",
-    "lib": ["ES2022", "DOM"],
-    "module": "NodeNext",
-    "moduleResolution": "NodeNext",
-    "strict": true,
-    "skipLibCheck": true,
-    "outDir": "dist",
-    "rootDir": ".",
-    "types": ["node"]
-  },
-  "include": ["index.ts"]
-}
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "lib": ["ES2022", "DOM"],
+    "module": "NodeNext",
+    "moduleResolution": "NodeNext",
+    "strict": true,
+    "skipLibCheck": true,
+    "outDir": "dist",
+    "rootDir": ".",
+    "types": ["node"]
+  },
+  "include": ["index.ts"]
+}

package/examples/minimal-node/README.md

@@ -1,31 +1,31 @@
-# minimal-node
-
-用 `@henjicc/ai-sdk@0.2.8` 的 `generation` 与 `runtime` 按需入口调用 KIE `z-image`。
-它保留完整 generation catalog，但不会带入 LLM/BigModel。示例不会下载结果文件；成功后只输出供应商返回的 URL。
-
-## 安装
-
-SDK 公开发布在 npm，无需配置 registry 或访问令牌。
-
-```bash
-npm install
-npm run dry-run
-```
-
-`dry-run` 会走真实 catalog、builder、预处理和 KIE provider，但用本地 `Response` 截住请求，
-输出中的 `networkCalls` 必须是 `0`。
-
-## 一次真实调用
-
-当前模型目录估价是 **$0.004/张**。确认账户有余额后，仅在当前终端注入密钥：
-
-```bash
-KIE_API_KEY='你的密钥' npm start
-```
-
-示例在付费创建端点前有一次性闸门：即使 SDK 的“安全预连接重试”判断被触发，也会在第二次 POST
-到达网络前本地拒绝，因此一次进程最多发出一个 KIE 创建请求。状态轮询是只读 GET，不会重复创建任务。
-
-`generate()` 返回 `pending + taskId` 时，示例继续调用 `continuePolling()` 直到得到 URL；SDK 不负责把
-URL 下载到磁盘。取消时应保留本次 `requestId` 或供应商 `taskId`，调用
-`client.cancel({ namespace: 'generation', taskId })`。
+# minimal-node
+
+用 `@henjicc/ai-sdk@0.3.0` 的 `generation` 与 `runtime` 按需入口调用 KIE `z-image`。
+它保留完整 generation catalog，但不会带入 LLM/BigModel。示例不会下载结果文件；成功后只输出供应商返回的 URL。
+
+## 安装
+
+SDK 公开发布在 npm，无需配置 registry 或访问令牌。
+
+```bash
+npm install
+npm run dry-run
+```
+
+`dry-run` 会走真实 catalog、builder、预处理和 KIE provider，但用本地 `Response` 截住请求，
+输出中的 `networkCalls` 必须是 `0`。
+
+## 一次真实调用
+
+当前模型目录估价是 **$0.004/张**。确认账户有余额后，仅在当前终端注入密钥：
+
+```bash
+KIE_API_KEY='你的密钥' npm start
+```
+
+示例在付费创建端点前有一次性闸门：即使 SDK 的“安全预连接重试”判断被触发，也会在第二次 POST
+到达网络前本地拒绝，因此一次进程最多发出一个 KIE 创建请求。状态轮询是只读 GET，不会重复创建任务。
+
+`generate()` 返回 `pending + taskId` 时，示例继续调用 `continuePolling()` 直到得到 URL；SDK 不负责把
+URL 下载到磁盘。取消时应保留本次 `requestId` 或供应商 `taskId`，调用
+`client.cancel({ namespace: 'generation', taskId })`。