@jeffreycao/copilot-api 2.0.1 → 2.1.0

package/README.md

@@ -401,6 +401,11 @@ request_max_retries = 3
 stream_max_retries = 1
 stream_idle_timeout_ms = 300000
 
+[features.code_mode]
+excluded_tool_namespaces = [
+    "mcp__codex_apps__sites"
+]
+
 [features]
 remote_compaction_v2 = true
 
@@ -410,6 +415,8 @@ enabled = false
 
 > [!NOTE]
 > `name` must be set to `"OpenAI"`.
+>
+> For models that do not support `tool_search`, we recommend disabling `mcp__codex_apps__sites` by adding it to `features.code_mode.excluded_tool_namespaces` as shown above. Otherwise, each prompt may consume more than 10,000 additional tokens.
 
 When a Codex client (`User-Agent` starts with `codex`) requests the top-level `GET /v1/models`, the gateway merges native Codex models with models available through the Messages adapter. The latter advertise `use_responses_lite: true`: `/v1/responses` uses **Responses → Messages** for Anthropic providers, while OpenAI-compatible providers and Chat-only Copilot models reuse the existing Messages route for **Responses → Messages → Chat Completions**, then translate streaming or JSON results back to Responses.
 

package/README.zh-CN.md

@@ -429,6 +429,11 @@ request_max_retries = 3
 stream_max_retries = 1
 stream_idle_timeout_ms = 300000
 
+[features.code_mode]
+excluded_tool_namespaces = [
+    "mcp__codex_apps__sites"
+]
+
 [features]
 remote_compaction_v2 = true
 
@@ -438,6 +443,8 @@ enabled = false
 
 > [!NOTE]
 > `name` 一定要配置为 `"OpenAI"`。
+>
+> 对于不支持 `tool_search` 的模型，建议按上面的配置将 `mcp__codex_apps__sites` 添加到 `features.code_mode.excluded_tool_namespaces` 中。否则每次提问可能会额外消耗 1 万多个 tokens。
 
 Codex 客户端（`User-Agent` 以 `codex` 开头）请求顶层 `GET /v1/models` 时，网关会把原生 Codex 模型与可通过 Messages 适配的模型合并返回。后者会声明 `use_responses_lite: true`：调用 `/v1/responses` 后，Anthropic provider 走 **Responses → Messages**，OpenAI 兼容 provider 以及只支持 Chat 的 Copilot 模型则复用现有 Messages 路由继续走 **Responses → Messages → Chat Completions**，最终统一翻译回 Responses（包括流式事件）。
 

package/dist/main.js

@@ -25,7 +25,7 @@ bindElectronFetch();
 const { auth } = await import("./auth-DH-ThnhJ.js");
 const { debug } = await import("./debug-D2giR-Kj.js");
 const { mcp } = await import("./mcp-BG6fpi6q.js");
-const { start } = await import("./start-BZqWghY0.js");
+const { start } = await import("./start-D3oRc7UM.js");
 await runMain(defineCommand({
 	meta: {
 		name: "copilot-api",

package/dist/{server-BDdqK9ps.js → server-DcL9pgmS.js}

@@ -2562,9 +2562,11 @@ async function forwardCodexAlphaSearchRequest(request) {
 	return createProviderProxyResponse(upstreamResponse);
 }
 function createAlphaSearchRequest(request, payload) {
-	return new Request(request, {
+	return new Request(request.url, {
 		body: JSON.stringify(payload),
-		method: "post"
+		headers: request.headers,
+		method: request.method,
+		signal: request.signal
 	});
 }
 async function handleCodexRequest(c, request, resolvedProviderConfig) {
@@ -2609,7 +2611,10 @@ async function handleAlphaSearchRequest(c, resolvedProviderConfig) {
 		if (providerModelAlias.provider === "codex") return await handleCodexRequest(c, createAlphaSearchRequest(c.req.raw, payload));
 	}
 	if (isAlphaSearchCodexPriorityEnabled()) {
-		if (await resolveProviderConfig("codex")) return await forwardCodexAlphaSearchRequest(createAlphaSearchRequest(c.req.raw, payload));
+		if (await resolveProviderConfig("codex")) {
+			if (!payload.model.startsWith("gpt")) payload.model = "gpt-5.6-luna";
+			return await forwardCodexAlphaSearchRequest(createAlphaSearchRequest(c.req.raw, payload));
+		}
 	}
 	const messagesBackedModel = await isMessagesBackedModel(resolvedRequestedModel);
 	if (messagesBackedModel) {
@@ -8134,6 +8139,7 @@ function translateResponsesToMessages(payload, options) {
 		content: MESSAGES_COMPACTION_PROMPT
 	});
 	if (messages.length === 0) throw new ResponsesMessagesTranslationError("Responses input must contain at least one translatable message");
+	applyEphemeralCacheControl(messages, system);
 	const reasoningEffort = translateReasoningEffort(payload.reasoning?.effort);
 	const messagesPayload = {
 		model: options.model,
@@ -8660,6 +8666,24 @@ function resolveMetadataUserId(payload) {
 	if (payload.safety_identifier?.trim()) return payload.safety_identifier;
 	if (payload.prompt_cache_key?.trim()) return payload.prompt_cache_key;
 }
+const EPHEMERAL_CACHE_CONTROL = { type: "ephemeral" };
+function applyEphemeralCacheControl(messages, system) {
+	const lastSystemBlock = system.at(-1);
+	if (lastSystemBlock) lastSystemBlock.cache_control = { ...EPHEMERAL_CACHE_CONTROL };
+	const lastMessage = messages.at(-1);
+	if (!lastMessage) return;
+	if (typeof lastMessage.content === "string") {
+		lastMessage.content = [{
+			type: "text",
+			text: lastMessage.content,
+			cache_control: { ...EPHEMERAL_CACHE_CONTROL }
+		}];
+		return;
+	}
+	const lastBlock = lastMessage.content.at(-1);
+	if (!lastBlock || lastBlock.type === "thinking") return;
+	lastBlock.cache_control = { ...EPHEMERAL_CACHE_CONTROL };
+}
 function appendAssistantBlock(messages, block) {
 	const last = messages.at(-1);
 	if (last?.role === "assistant" && Array.isArray(last.content)) {
@@ -8772,6 +8796,9 @@ var CustomToolInputStreamDecoder = class {
 	prefixTokenOffset = 0;
 	safeInputLength = 0;
 	state = "prefix";
+	trailingDepth = 0;
+	trailingEscaped = false;
+	trailingInString = false;
 	unicodeDigitsRemaining = 0;
 	append(partialJson) {
 		if (this.failed) return this.fail("decoder is already in a failed state");
@@ -8807,8 +8834,42 @@ var CustomToolInputStreamDecoder = class {
 		}
 		if (this.state === "suffix") {
 			if (isJsonWhitespace(char)) return;
-			if (char !== "}") return this.fail("expected \"}\" after the input string");
-			this.state = "done";
+			if (char === "}") {
+				this.state = "done";
+				return;
+			}
+			if (char === ",") {
+				this.state = "trailing";
+				this.trailingDepth = 1;
+				return;
+			}
+			return this.fail("expected \"}\" or \",\" after the input string");
+		}
+		if (this.state === "trailing") {
+			if (this.trailingInString) {
+				if (this.trailingEscaped) {
+					this.trailingEscaped = false;
+					return;
+				}
+				if (char === "\\") {
+					this.trailingEscaped = true;
+					return;
+				}
+				if (char === "\"") this.trailingInString = false;
+				return;
+			}
+			if (char === "\"") {
+				this.trailingInString = true;
+				return;
+			}
+			if (char === "{" || char === "[") {
+				this.trailingDepth += 1;
+				return;
+			}
+			if (char === "}" || char === "]") {
+				this.trailingDepth -= 1;
+				if (this.trailingDepth === 0) this.state = "done";
+			}
 			return;
 		}
 		if (this.state === "done") {
@@ -8869,7 +8930,7 @@ function isJsonWhitespace(char) {
 }
 //#endregion
 //#region src/routes/responses/messages-stream-translation.ts
-const EMPTY_SIGNATURE_ENCRYPTED_CONTENT = "Y29waWxvdC1hcGk6bWVzc2FnZXMtZW1wdHktc2lnbmF0dXJlOnYx";
+const EMPTY_SIGNATURE_ENCRYPTED_CONTENT = "";
 async function* translateMessagesStream(chunks, context) {
 	const state = createTranslationState(context);
 	try {
@@ -10062,4 +10123,4 @@ server.route("/:provider/images", providerImageRoutes);
 //#endregion
 export { server };
 
-//# sourceMappingURL=server-BDdqK9ps.js.map
+//# sourceMappingURL=server-DcL9pgmS.js.map