@codehz/ai 0.2.2 → 0.2.4

package/README.md

@@ -1,11 +1,11 @@
-# nano-ai
+# @codehz/ai
 
-统一流式 AI 客户端，提供一套 canonical API，对接真实模型后端与面向测试的回调驱动 `MockAdapter`（`responses` / `messages` / `chat.completions` / `ollama` / `mock`）。
+统一流式 AI 客户端，提供一套 canonical API，对接真实模型后端与面向测试的回调驱动 `MockAdapter`（`responses` / `messages` / `chat-completions` / `ollama` / `mock`）。
 
 ## 安装
 
 ```bash
-bun add nano-ai
+bun add @codehz/ai
 ```
 
 依赖：Bun（内置 `fetch`、`crypto`），无需额外运行时依赖。
@@ -13,7 +13,7 @@ bun add nano-ai
 ## 快速开始
 
 ```ts
-import { createAIClient, ResponsesAdapter } from "nano-ai";
+import { createAIClient, ResponsesAdapter } from "@codehz/ai";
 
 const client = createAIClient({
   adapter: new ResponsesAdapter({ apiKey: process.env.OPENAI_API_KEY! }),
@@ -69,22 +69,22 @@ response.started → (item.started → item.delta* → item.completed)* → resp
 
 事件类型：
 
-| 事件                                  | 含义                            |
-| ------------------------------------- | ------------------------------- |
-| `response.started`                    | 响应开始                        |
-| `message.{started,delta,completed}`   | 消息输出                        |
-| `reasoning.{started,delta,completed}` | 思维链                          |
-| `tool_call.{started,delta,completed}` | 工具调用                        |
-| `response.warning`                    | 非致命警告                      |
-| `response.auxiliary`                  | usage / billing 辅助信息        |
-| `response.completed`                  | 响应结束，内含完整 `AIResponse` |
+| 事件                                  | 含义                                        |
+| ------------------------------------- | ------------------------------------------- |
+| `response.started`                    | 响应开始                                    |
+| `message.{started,delta,completed}`   | 消息输出                                    |
+| `reasoning.{started,delta,completed}` | 思维链                                      |
+| `tool_call.{started,delta,completed}` | 工具调用                                    |
+| `response.warning`                    | 非致命警告                                  |
+| `response.auxiliary`                  | usage / billing 辅助信息                    |
+| `response.completed`                  | 响应结束，携带 replay、终止原因及最终元数据 |
 
 ### 统一终结结果
 
 流结束后可通过 `collectStream()` 聚合为 `AIResponse`：
 
 ```ts
-import { collectStream } from "nano-ai";
+import { collectStream } from "@codehz/ai";
 
 const response = await collectStream(client.stream({ input }));
 console.log(response.text); // 全部文本
@@ -95,17 +95,22 @@ console.log(response.replay); // 续接材料
 
 `AIResponse` 包含：
 
-| 字段         | 类型             | 说明                     |
-| ------------ | ---------------- | ------------------------ |
-| `output`     | `OutputItem[]`   | 当前轮输出               |
-| `replay`     | `ReplayItem[]`   | 续接材料（下次请求带回） |
-| `text`       | `string`         | 全部文本拼接             |
-| `toolCalls`  | `ToolCallItem[]` | 工具调用                 |
-| `stopReason` | `StopReason`     | 终止原因                 |
-| `usage`      | `Usage`          | token 统计               |
-| `billing`    | `BillingInfo`    | 计费信息                 |
-| `warnings`   | `string[]`       | 非致命警告               |
-| `backend`    | `BackendTrace`   | 调用链路元数据           |
+| 字段         | 类型             | 说明                      |
+| ------------ | ---------------- | ------------------------- |
+| `output`     | `OutputItem[]`   | 当前轮输出                |
+| `replay`     | `ReplayItem[]`   | 续接材料（下次请求带回）  |
+| `text`       | `string`         | 全部文本拼接              |
+| `toolCalls`  | `ToolCallItem[]` | 工具调用                  |
+| `stopReason` | `StopReason?`    | 终止原因（可选）          |
+| `usage`      | `Usage?`         | token 统计（可选）        |
+| `billing`    | `BillingInfo?`   | 计费信息（可选）          |
+| `auxiliary`  | `AuxiliaryInfo?` | Provider 辅助信息（可选） |
+| `warnings`   | `string[]?`      | 非致命警告（可选）        |
+| `backend`    | `BackendTrace`   | 调用链路元数据            |
+
+流式 `message.delta` / `reasoning.delta` 保持后端分片粒度；完成态 `output` 中的
+`message` / `reasoning` 会合并相邻 `text` content blocks（直接拼接且不添加分隔符），
+非文本 block 仍保留原有边界。
 
 ## 后端 Adapter
 
@@ -125,7 +130,7 @@ import {
   OllamaAdapter,
   MockAdapter,
   withMockStreaming,
-} from "nano-ai";
+} from "@codehz/ai";
 
 // OpenAI Responses API
 const responses = new ResponsesAdapter({ apiKey: "sk-..." });
@@ -171,8 +176,8 @@ adapter.capabilities.usage; // "stream" | "final" | "none"
 adapter.capabilities.toolResultOutcomes;
 ```
 
-`nativeStreaming` 已由 `capabilities` 直接替代。响应级 `backend.isSyntheticStream` 根据
-`textStreaming === "synthetic"` 推导；具体响应内容仍应从本次事件流、warning 和 `replay` 判断。
+响应级 `backend.isSyntheticStream` 根据 `textStreaming === "synthetic"` 推导；具体响应内容仍应从
+本次事件流、warning 和 `replay` 判断。
 
 ## Mock 后端
 
@@ -204,7 +209,7 @@ const handler = withMockStreaming(
 - 可用 `assertMockRequest()` 验证调用方是否把上一轮 `replay` 和当前 `tool_result` 正确带回
 
 ```ts
-import { assertMockRequest, createAIClient, MockAdapter } from "nano-ai";
+import { assertMockRequest, createAIClient, MockAdapter } from "@codehz/ai";
 
 const client = createAIClient({
   adapter: new MockAdapter({
@@ -321,7 +326,7 @@ const r2 = await collectStream(client.stream({ input, tools }));
 非流式后端可通过 `syntheticStream()` 包装为规范事件流：
 
 ```ts
-import { syntheticStream } from "nano-ai";
+import { syntheticStream } from "@codehz/ai";
 
 const events = syntheticStream({
   model: "gpt-4o",
@@ -341,7 +346,7 @@ for await (const event of events) {
 `AuxiliaryCollector` 提供分层 best-effort 采集（流事件 → headers → lookup → derived）：
 
 ```ts
-import { AuxiliaryCollector } from "nano-ai";
+import { AuxiliaryCollector } from "@codehz/ai";
 
 const collector = new AuxiliaryCollector();
 collector.recordUsage({ inputTokens: 10, outputTokens: 5 }, "stream");

package/dist/index.d.mts

@@ -88,7 +88,8 @@ type AIRequest = {
   include?: IncludeSettings;
   metadata?: Record<string, string>;
   temperature?: number;
-  maxOutputTokens?: number;
+  maxOutputTokens?: number; /** AbortSignal 用于打断请求。abort 时 fetch 调用会被取消，流迭代器抛出 AbortError。 */
+  signal?: AbortSignal;
 };
 //#endregion
 //#region src/types/response.d.ts
@@ -250,7 +251,8 @@ interface BackendAdapter {
 type CreateAIClientOptions = {
   adapter: BackendAdapter;
   model: string;
-  defaults?: Partial<AIRequest>;
+  defaults?: Partial<AIRequest>; /** 全局默认 AbortSignal，当 request.signal 未设置时生效。 */
+  signal?: AbortSignal;
 };
 interface AIClient {
   stream(request: AIRequest): AsyncIterable<AIStreamEvent>;
@@ -839,7 +841,8 @@ type MockHandlerContext = {
   turnIndex: number;
   previousReplay: ReplayItem[];
   pendingToolCalls: readonly ToolCallItem[];
-  history: readonly MockHistoryRecord[];
+  history: readonly MockHistoryRecord[]; /** 请求的 AbortSignal，handler 可检查 signal.aborted 提前退出。 */
+  signal?: AbortSignal;
 };
 type MockWarningStep = {
   type: "warning";

package/dist/index.mjs

@@ -313,15 +313,29 @@ function normalizeRequest(request, options) {
 //#endregion
 //#region src/core/client.ts
 function createAIClient(options) {
-	const { adapter, model, defaults } = options;
+	const { adapter, model, defaults, signal: defaultSignal } = options;
 	return { stream(request) {
-		const normalized = normalizeRequest(request, {
+		const signal = mergeAbortSignals(defaultSignal, request.signal);
+		const normalized = normalizeRequest({
+			...request,
+			signal
+		}, {
 			model,
 			defaults
 		});
 		return adapter.stream(normalized);
 	} };
 }
+/**
+* 合并多个 AbortSignal：任一 signal abort 即触发。
+* 如果没有 signal 需要合并则返回 undefined。
+*/
+function mergeAbortSignals(...signals) {
+	const valid = signals.filter((s) => s != null);
+	if (valid.length === 0) return void 0;
+	if (valid.length === 1) return valid[0];
+	return AbortSignal.any(valid);
+}
 //#endregion
 //#region src/core/event-factory.ts
 function timestamp() {
@@ -494,12 +508,21 @@ function getActiveItem(state, itemId, expectedType) {
 	if (item.type !== expectedType) throw streamProtocolError(`Item ${itemId} started as ${item.type} but received ${expectedType} event`);
 	return item;
 }
+function coalesceContentBlocks(blocks) {
+	const result = [];
+	for (const block of blocks) {
+		const previous = result[result.length - 1];
+		if (block.type === "text" && previous?.type === "text") previous.text += block.text;
+		else result.push({ ...block });
+	}
+	return result;
+}
 function finalizeMessage(active) {
 	return {
 		type: "message",
 		id: active.id,
 		role: active.role,
-		content: active.content
+		content: coalesceContentBlocks(active.content)
 	};
 }
 function finalizeReasoning(active) {
@@ -507,7 +530,7 @@ function finalizeReasoning(active) {
 		type: "reasoning",
 		id: active.id,
 		visibility: active.visibility,
-		content: active.content
+		content: coalesceContentBlocks(active.content)
 	};
 }
 function finalizeToolCall(active) {
@@ -1067,6 +1090,7 @@ var AdapterBase = class {
 	* 3. 委托 runStream 发射全部流事件（含 response.completed）
 	*/
 	async *stream(request) {
+		request.signal?.throwIfAborted();
 		const factory = createEventFactory({
 			responseId: request.requestId,
 			backend: {
@@ -1750,7 +1774,8 @@ var ResponsesAdapter = class extends AdapterBase {
 					"Content-Type": "application/json",
 					Authorization: `Bearer ${this.apiKey}`
 				},
-				body: JSON.stringify(providerRequest)
+				body: JSON.stringify(providerRequest),
+				signal: request.signal
 			});
 		} catch (err) {
 			throw new AIProviderError(err instanceof Error ? err.message : String(err), "PROVIDER_ERROR");
@@ -2166,7 +2191,8 @@ var MessagesAdapter = class extends AdapterBase {
 					"x-api-key": this.apiKey,
 					"anthropic-version": this.apiVersion
 				},
-				body: JSON.stringify(providerRequest)
+				body: JSON.stringify(providerRequest),
+				signal: request.signal
 			});
 		} catch (err) {
 			throw new AIProviderError(err instanceof Error ? err.message : String(err), "PROVIDER_ERROR");
@@ -2634,7 +2660,8 @@ var ChatCompletionsAdapter = class extends AdapterBase {
 					"Content-Type": "application/json",
 					Authorization: `Bearer ${this.apiKey}`
 				},
-				body: JSON.stringify(providerRequest)
+				body: JSON.stringify(providerRequest),
+				signal: request.signal
 			});
 		} catch (err) {
 			throw new AIProviderError(err instanceof Error ? err.message : String(err), "PROVIDER_ERROR");
@@ -3052,7 +3079,8 @@ var OllamaAdapter = class extends AdapterBase {
 			response = await this.fetchFn(`${this.baseUrl}/api/chat`, {
 				method: "POST",
 				headers,
-				body: JSON.stringify(providerRequest)
+				body: JSON.stringify(providerRequest),
+				signal: request.signal
 			});
 		} catch (err) {
 			throw new AIProviderError(err instanceof Error ? err.message : String(err), "PROVIDER_ERROR");
@@ -3287,7 +3315,7 @@ var MockAdapter = class extends AdapterBase {
 	}
 	async buildRequest(request) {
 		const turnIndex = this.cursor;
-		const context = this.buildHandlerContext(turnIndex);
+		const context = this.buildHandlerContext(turnIndex, request.signal);
 		const remainingPendingToolCalls = consumePendingToolCalls(this.pendingToolCalls, request.input);
 		const handlerResult = this.handler(request, context);
 		this.cursor += 1;
@@ -3306,6 +3334,7 @@ var MockAdapter = class extends AdapterBase {
 			const output = [];
 			let stepCount = 0;
 			for await (const step of mockRequest.handlerResult) {
+				if (request.signal?.aborted) return;
 				stepCount += 1;
 				switch (step.type) {
 					case "warning":
@@ -3425,7 +3454,7 @@ var MockAdapter = class extends AdapterBase {
 			rawResponseId: completion.rawResponseId
 		}, factory);
 	}
-	buildHandlerContext(turnIndex) {
+	buildHandlerContext(turnIndex, signal) {
 		return {
 			turnIndex,
 			previousReplay: this.previousReplay.map(cloneItem),
@@ -3434,7 +3463,8 @@ var MockAdapter = class extends AdapterBase {
 				...record,
 				replay: record.replay.map(cloneItem),
 				toolCalls: record.toolCalls.map(cloneItem)
-			}))
+			})),
+			signal
 		};
 	}
 };