copilot-api-plus 1.0.41 → 1.0.43

package/README.md

@@ -1,7 +1,5 @@
 # Copilot API Plus
 
-> **Fork of [ericc-ch/copilot-api](https://github.com/ericc-ch/copilot-api)** with bug fixes and improvements.
-
 将 GitHub Copilot、OpenCode Zen、Google Antigravity 等 AI 服务转换为 **OpenAI** 和 **Anthropic** 兼容 API，支持与 [Claude Code](https://docs.anthropic.com/en/docs/claude-code/overview)、[opencode](https://github.com/sst/opencode) 等工具无缝集成。
 
 ---
@@ -18,7 +16,9 @@
 - [Claude Code 集成](#-claude-code-集成)
 - [opencode 集成](#-opencode-集成)
 - [API 端点](#-api-端点)
-- [命令行参考](#-命令行参考)
+- [API Key 认证](#-api-key-认证)
+- [技术细节](#-技术细节)
+- [命令行参考](#️-命令行参考)
 - [Docker 部署](#-docker-部署)
 - [常见问题](#-常见问题)
 
@@ -36,6 +36,10 @@
 | ⚡ **速率限制** | 内置请求频率控制，避免触发限制 |
 | 🌐 **代理支持** | 支持 HTTP/HTTPS 代理，配置持久化 |
 | 🐳 **Docker 支持** | 提供完整的 Docker 部署方案 |
+| 🔑 **API Key 认证** | 可选的 API Key 鉴权，保护公开部署的服务 |
+| ✂️ **智能上下文压缩** | Prompt 超过模型 Token 限制时自动截断，保留系统消息和最近对话 |
+| 🔍 **智能模型匹配** | 自动处理模型名格式差异（日期后缀、dash/dot 版本号等） |
+| 🔁 **Antigravity 端点容错** | 双端点自动切换，按模型族追踪速率限制，指数退避重试 |
 
 ---
 
@@ -558,6 +562,7 @@ curl http://localhost:4141/v1/messages \
 | `--github-token` | `-g` | - | 直接提供 GitHub Token |
 | `--show-token` | - | false | 显示 Token 信息 |
 | `--proxy-env` | - | false | 从环境变量读取代理 |
+| `--api-key` | - | - | API Key 鉴权（可多次指定） |
 
 ### proxy 命令参数
 
@@ -604,6 +609,72 @@ npx copilot-api-plus@latest antigravity clear    # 清除所有账户
 
 ---
 
+## 🔑 API Key 认证
+
+如果你将服务暴露到公网，可以启用 API Key 认证来保护接口：
+
+```bash
+# 单个 Key
+npx copilot-api-plus@latest start --api-key my-secret-key
+
+# 多个 Key
+npx copilot-api-plus@latest start --api-key key1 --api-key key2
+```
+
+启用后，所有请求需要携带 API Key：
+
+```bash
+# OpenAI 格式 - 通过 Authorization header
+curl http://localhost:4141/v1/chat/completions \
+  -H "Authorization: Bearer my-secret-key" \
+  -H "Content-Type: application/json" \
+  -d '{"model": "claude-sonnet-4", "messages": [{"role": "user", "content": "Hello"}]}'
+
+# Anthropic 格式 - 通过 x-api-key header
+curl http://localhost:4141/v1/messages \
+  -H "x-api-key: my-secret-key" \
+  -H "Content-Type: application/json" \
+  -d '{"model": "claude-sonnet-4", "max_tokens": 1024, "messages": [{"role": "user", "content": "Hello"}]}'
+```
+
+在 Claude Code 中使用时，将 `ANTHROPIC_AUTH_TOKEN` 设为你的 API Key 即可。
+
+---
+
+## 🔧 技术细节
+
+### 智能上下文压缩
+
+当 Prompt Token 数量超过模型的上下文窗口限制时，代理层会自动截断消息以避免上游 API 返回 400 错误：
+
+- **保留系统/开发者消息**：system 和 developer 角色的消息始终保留
+- **保留最近对话**：优先丢弃最早的消息，保留最近的上下文
+- **工具调用分组**：assistant 的 tool_calls 和对应的 tool result 消息作为一组，不会被拆散
+- **5% 安全余量**：实际限制为模型上下文窗口的 95%，避免边界情况
+
+### 智能模型名匹配
+
+Anthropic 格式的模型名（如 `claude-opus-4-6`）和 Copilot 的模型列表 ID 可能存在格式差异。代理使用多策略精确匹配：
+
+| 策略 | 示例 |
+|------|------|
+| 精确匹配 | `claude-opus-4-6` → `claude-opus-4-6` |
+| 去日期后缀 | `claude-opus-4-6-20251101` → `claude-opus-4-6` |
+| Dash → Dot | `claude-opus-4-5` → `claude-opus-4.5` |
+| Dot → Dash | `claude-opus-4.5` → `claude-opus-4-5` |
+
+对于 Anthropic 端点（`/v1/messages`），还会先通过 `translateModelName` 做格式转换（包括旧格式 `claude-3-5-sonnet` → `claude-sonnet-4.5` 的映射），再通过上述策略匹配。
+
+### Antigravity 端点容错
+
+Google Antigravity 模式内置了可靠性保障：
+
+- **双端点自动切换**：daily sandbox 和 production 两个端点，一个失败自动切换到另一个
+- **按模型族速率追踪**：分别追踪 Gemini 和 Claude 模型族的速率限制状态
+- **指数退避重试**：429/503 等限流错误自动退避重试，短间隔走同端点，长间隔切换端点
+
+---
+
 ## 🐳 Docker 部署
 
 ### 快速启动

package/dist/main.js

@@ -1194,6 +1194,15 @@ const apiKeyAuthMiddleware = async (c, next) => {
 //#endregion
 //#region src/lib/model-logger.ts
 /**
+* Global token usage store for passing usage info from handlers to logger.
+* Handlers call setTokenUsage() when usage is available,
+* logger reads and clears it after await next().
+*/
+let pendingTokenUsage;
+function setTokenUsage(usage) {
+	pendingTokenUsage = usage;
+}
+/**
 * Get timestamp string in format HH:mm:ss
 */
 function getTime() {
@@ -1207,6 +1216,15 @@ function formatDuration(ms) {
 	return `${(ms / 1e3).toFixed(1)}s`;
 }
 /**
+* Format token usage for log output
+*/
+function formatTokenUsage(usage) {
+	const parts = [`in:${usage.inputTokens}`, `out:${usage.outputTokens}`];
+	if (usage.cacheReadTokens) parts.push(`cache_read:${usage.cacheReadTokens}`);
+	if (usage.cacheCreationTokens) parts.push(`cache_create:${usage.cacheCreationTokens}`);
+	return parts.join(" ");
+}
+/**
 * Extract model name from request body
 */
 async function extractModel(c) {
@@ -1221,7 +1239,7 @@ async function extractModel(c) {
 *
 * Output format:
 * [model] HH:mm:ss <-- METHOD /path
-* [model] HH:mm:ss --> METHOD /path STATUS DURATION
+* [model] HH:mm:ss --> METHOD /path STATUS DURATION [in:N out:N]
 */
 function modelLogger() {
 	return async (c, next) => {
@@ -1233,12 +1251,16 @@ function modelLogger() {
 		if (method === "POST" && c.req.header("content-type")?.includes("json")) model = await extractModel(c);
 		const modelPrefix = model ? `[${model}] ` : "";
 		const startTime = getTime();
+		pendingTokenUsage = void 0;
 		console.log(`${modelPrefix}${startTime} <-- ${method} ${fullPath}`);
 		const start$1 = Date.now();
 		await next();
 		const duration = Date.now() - start$1;
 		const endTime = getTime();
-		console.log(`${modelPrefix}${endTime} --> ${method} ${fullPath} ${c.res.status} ${formatDuration(duration)}`);
+		const usage = pendingTokenUsage;
+		pendingTokenUsage = void 0;
+		const usageSuffix = usage ? ` [${formatTokenUsage(usage)}]` : "";
+		console.log(`${modelPrefix}${endTime} --> ${method} ${fullPath} ${c.res.status} ${formatDuration(duration)}${usageSuffix}`);
 	};
 }
 
@@ -2935,11 +2957,27 @@ const createChatCompletions = async (payload) => {
 		model: payload.model,
 		endpoint: `${copilotBaseUrl(state)}/chat/completions`
 	});
-	const response = await fetch(`${copilotBaseUrl(state)}/chat/completions`, {
+	const url = `${copilotBaseUrl(state)}/chat/completions`;
+	const fetchOptions = {
 		method: "POST",
 		headers,
 		body: JSON.stringify(payload)
-	});
+	};
+	const maxRetries = 2;
+	let lastError;
+	let response;
+	for (let attempt = 0; attempt <= maxRetries; attempt++) try {
+		response = await fetch(url, fetchOptions);
+		break;
+	} catch (error) {
+		lastError = error;
+		if (attempt < maxRetries) {
+			const delay = 1e3 * (attempt + 1);
+			consola.warn(`Network error on attempt ${attempt + 1}/${maxRetries + 1}, retrying in ${delay}ms:`, error instanceof Error ? error.message : error);
+			await new Promise((r) => setTimeout(r, delay));
+		}
+	}
+	if (!response) throw lastError;
 	if (!response.ok) {
 		const errorBody = await response.text();
 		consola.error("Failed to create chat completions", {
@@ -2994,12 +3032,28 @@ async function handleCompletion$1(c) {
 	const response = await createChatCompletions(payload);
 	if (isNonStreaming$1(response)) {
 		consola.debug("Non-streaming response:", JSON.stringify(response));
+		if (response.usage) setTokenUsage({
+			inputTokens: response.usage.prompt_tokens,
+			outputTokens: response.usage.completion_tokens,
+			cacheReadTokens: response.usage.prompt_tokens_details?.cached_tokens
+		});
 		return c.json(response);
 	}
 	consola.debug("Streaming response");
 	return streamSSE(c, async (stream) => {
 		for await (const chunk of response) {
 			consola.debug("Streaming chunk:", JSON.stringify(chunk));
+			try {
+				const sseChunk = chunk;
+				if (sseChunk.data && sseChunk.data !== "[DONE]") {
+					const parsed = JSON.parse(sseChunk.data);
+					if (parsed.usage) setTokenUsage({
+						inputTokens: parsed.usage.prompt_tokens ?? 0,
+						outputTokens: parsed.usage.completion_tokens ?? 0,
+						cacheReadTokens: parsed.usage.prompt_tokens_details?.cached_tokens
+					});
+				}
+			} catch {}
 			await stream.writeSSE(chunk);
 		}
 	});
@@ -3446,6 +3500,12 @@ async function handleCompletion(c) {
 	const response = await createChatCompletions(openAIPayload);
 	if (isNonStreaming(response)) {
 		const anthropicResponse = translateToAnthropic(response);
+		setTokenUsage({
+			inputTokens: anthropicResponse.usage.input_tokens,
+			outputTokens: anthropicResponse.usage.output_tokens,
+			cacheReadTokens: anthropicResponse.usage.cache_read_input_tokens,
+			cacheCreationTokens: anthropicResponse.usage.cache_creation_input_tokens
+		});
 		return c.json(anthropicResponse);
 	}
 	return streamSSE(c, async (stream) => {
@@ -3460,6 +3520,11 @@ async function handleCompletion(c) {
 			if (!rawEvent.data) continue;
 			const chunk = JSON.parse(rawEvent.data);
 			const events$1 = translateChunkToAnthropicEvents(chunk, streamState);
+			if (chunk.usage) setTokenUsage({
+				inputTokens: chunk.usage.prompt_tokens - (chunk.usage.prompt_tokens_details?.cached_tokens ?? 0),
+				outputTokens: chunk.usage.completion_tokens,
+				cacheReadTokens: chunk.usage.prompt_tokens_details?.cached_tokens
+			});
 			for (const event of events$1) await stream.writeSSE({
 				event: event.type,
 				data: JSON.stringify(event)