@jeffreycao/copilot-api 2.1.5 → 2.1.7

package/README.md

@@ -398,7 +398,7 @@ requires_openai_auth = true
 supports_websockets = false
 wire_api = "responses"
 request_max_retries = 3
-stream_max_retries = 1
+stream_max_retries = 3
 stream_idle_timeout_ms = 300000
 
 [features]
@@ -417,6 +417,8 @@ enabled = false
 
 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.
 
+For Codex clients, only `gpt-*` Copilot models use the native Responses API; non-GPT Copilot models always go through the adapter, even when they advertise native `/responses` support.
+
 Responses Lite tool definitions are read from `input.additional_tools`, without relying on top-level `tools`. Function, `namespace`, and custom tools are supported; clients must declare `apply_patch` as `type: "custom"`, and it is not handled as a standalone special tool type. Returned calls recover their original `name` and `namespace`. Tools are collected before old history is trimmed, so compaction requests retain them. The Messages fallback does not support Responses `tool_search` mode. Anthropic `output_config.effort` keeps the project's existing valid levels; Responses `minimal` maps to `low`, while `none` omits Anthropic effort.
 
 When Codex uses the top-level GitHub Copilot route with `approvals_reviewer = "auto_review"`, map its internal review model to a Responses-capable Copilot model in the gateway's `config.json`:

package/README.zh-CN.md

@@ -426,7 +426,7 @@ requires_openai_auth = true
 supports_websockets = false
 wire_api = "responses"
 request_max_retries = 3
-stream_max_retries = 1
+stream_max_retries = 3
 stream_idle_timeout_ms = 300000
 
 [features]
@@ -445,6 +445,8 @@ enabled = false
 
 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）。
 
+对 Codex 客户端而言，只有 `gpt-*` Copilot 模型走原生 Responses API；非 GPT Copilot 模型一律走适配路径，即使声明支持原生 `/responses` 也不例外。
+
 Responses Lite 的工具定义从 `input` 中的 `additional_tools` 读取，而不是依赖顶层 `tools`。该适配支持 function、`namespace` 和 custom tool；`apply_patch` 需要由客户端声明为 `type: "custom"`，不会作为独立工具类型特殊处理。工具调用返回时会恢复原始 `name` 与 `namespace`；压缩请求在裁剪旧历史前先保存工具定义，因此压缩期间也不会丢失工具。Messages 回退路径不支持 Responses `tool_search` 模式。Anthropic 的 `output_config.effort` 仍只使用项目既有的合法档位；Responses 的 `minimal` 会降级为 `low`，`none` 则不向 Anthropic 发送 effort。
 
 当 Codex 通过顶层 GitHub Copilot 路由并设置 `approvals_reviewer = "auto_review"` 时，可在网关的 `config.json` 中将内部审核模型映射到一个支持 Responses API 的 Copilot 模型：

package/dist/main.js

@@ -25,7 +25,7 @@ bindElectronFetch();
 const { auth } = await import("./auth-t5Gi1JB9.js");
 const { debug } = await import("./debug-2GkXJBt5.js");
 const { mcp } = await import("./mcp-BG6fpi6q.js");
-const { start } = await import("./start-DzKOeRTn.js");
+const { start } = await import("./start-D5W3xtWv.js");
 await runMain(defineCommand({
 	meta: {
 		name: "copilot-api",

package/dist/{server-DyVtfan1.js → server-s6RzK7Oo.js}

@@ -3458,6 +3458,7 @@ function translateToOpenAI(payload, options = {}) {
 	const model = state.models?.data.find((m) => m.id === modelId);
 	const thinkingBudget = getThinkingBudget(payload, model);
 	const reasoningEffort = getReasoningEffort(payload, options);
+	const promptCacheKey = requestContext.getStore()?.sessionAffinity?.trim();
 	return {
 		model: modelId,
 		messages: translateAnthropicMessagesToOpenAI(payload, modelId, {
@@ -3473,7 +3474,8 @@ function translateToOpenAI(payload, options = {}) {
 		tools: translateAnthropicToolsToOpenAI(payload.tools),
 		tool_choice: translateAnthropicToolChoiceToOpenAI(payload.tool_choice),
 		thinking_budget: thinkingBudget,
-		...reasoningEffort ? { reasoning_effort: reasoningEffort } : {}
+		...reasoningEffort ? { reasoning_effort: reasoningEffort } : {},
+		...promptCacheKey ? { prompt_cache_key: promptCacheKey } : {}
 	};
 }
 function getReasoningEffort(payload, options) {
@@ -6355,7 +6357,7 @@ const CODEX_MODELS = [
 		]
 	},
 	{
-		contextWindow: 105e4,
+		contextWindow: 272e3,
 		id: "gpt-5.6-sol",
 		input: ["text", "image"],
 		maxTokens: 128e3,
@@ -6370,7 +6372,7 @@ const CODEX_MODELS = [
 		]
 	},
 	{
-		contextWindow: 105e4,
+		contextWindow: 272e3,
 		id: "gpt-5.6-terra",
 		input: ["text", "image"],
 		maxTokens: 128e3,
@@ -6385,7 +6387,7 @@ const CODEX_MODELS = [
 		]
 	},
 	{
-		contextWindow: 105e4,
+		contextWindow: 272e3,
 		id: "gpt-5.6-luna",
 		input: ["text", "image"],
 		maxTokens: 128e3,
@@ -8090,15 +8092,20 @@ function getCopilotCodexCandidates() {
 function isCopilotCodexCandidate(model) {
 	return (model.supported_endpoints ?? []).some((endpoint) => endpoint === MESSAGES_ENDPOINT || endpoint === CHAT_COMPLETIONS_ENDPOINT || RESPONSES_ENDPOINTS.has(endpoint)) && model.capabilities.supports.tool_calls !== false;
 }
+function describeCopilotAdapter(model) {
+	const supportsResponses = model.supported_endpoints?.some((endpoint) => RESPONSES_ENDPOINTS.has(endpoint));
+	if (model.id.startsWith("gpt") && supportsResponses) return `${model.name} through the Copilot Responses API.`;
+	if (model.supported_endpoints?.includes(MESSAGES_ENDPOINT)) return `${model.name} through the Copilot Messages adapter.`;
+	if (supportsResponses) return `${model.name} through the Copilot Messages-to-Responses adapter.`;
+	return `${model.name} through the Copilot Messages-to-Chat adapter.`;
+}
 function createCopilotCodexCandidate(model) {
 	const reasoningEfforts = normalizeReasoningEfforts(model.capabilities.supports.reasoning_effort);
-	const usesNativeResponses = model.supported_endpoints?.some((endpoint) => RESPONSES_ENDPOINTS.has(endpoint));
-	const usesNativeMessages = model.supported_endpoints?.includes(MESSAGES_ENDPOINT);
 	return {
 		slug: toClientModelId(model.id),
 		displayName: model.name,
-		description: usesNativeResponses ? `${model.name} through the Copilot Responses API.` : usesNativeMessages ? `${model.name} through the Copilot Messages adapter.` : `${model.name} through the Copilot Messages-to-Chat adapter.`,
-		contextWindow: positiveNumber(model.capabilities.limits.max_context_window_tokens, 256e3),
+		description: describeCopilotAdapter(model),
+		contextWindow: positiveNumber(model.capabilities.limits.max_prompt_tokens, 256e3),
 		maxOutputTokens: positiveNumber(model.capabilities.limits.max_output_tokens, 32e3),
 		inputModalities: model.capabilities.supports.vision ? ["text", "image"] : ["text"],
 		reasoningEfforts,
@@ -8177,15 +8184,16 @@ function normalizeInputModalities(value) {
 	if (!Array.isArray(value)) return [];
 	return [...new Set(value.filter((modality) => modality === "text" || modality === "image"))];
 }
-function resolveInputModalities(providerName, configuredModalities, remoteModalities, builtinModalities) {
-	if (configuredModalities.length > 0) return configuredModalities;
-	if (providerName === "kimi") {
-		const modalities = remoteModalities.length > 0 ? remoteModalities : builtinModalities.length > 0 ? builtinModalities : ["text"];
-		return [...new Set([...modalities, "image"])];
-	}
+function fallbackModalities(remoteModalities, builtinModalities) {
 	if (remoteModalities.length > 0) return remoteModalities;
 	return builtinModalities.length > 0 ? builtinModalities : ["text"];
 }
+function resolveInputModalities(providerName, configuredModalities, remoteModalities, builtinModalities) {
+	if (configuredModalities.length > 0) return configuredModalities;
+	const modalities = fallbackModalities(remoteModalities, builtinModalities);
+	if (providerName === "kimi") return [...new Set([...modalities, "image"])];
+	return modalities;
+}
 function selectDefaultReasoningEffort(efforts, configured) {
 	if (configured && efforts.includes(configured)) return configured;
 	if (efforts.includes("max")) return "max";
@@ -8400,6 +8408,7 @@ function translateResponsesToMessages(payload, options) {
 	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 metadataUserId = resolveMetadataUserId(payload);
 	const messagesPayload = {
 		model: options.model,
 		messages,
@@ -8412,7 +8421,7 @@ function translateResponsesToMessages(payload, options) {
 		tool_choice: translateToolChoice(payload.tool_choice, registry),
 		...reasoningEffort ? { output_config: { effort: reasoningEffort } } : {},
 		...payload.service_tier === "auto" || payload.service_tier === "standard_only" ? { service_tier: payload.service_tier } : {},
-		...resolveMetadataUserId(payload) ? { metadata: { user_id: resolveMetadataUserId(payload) } } : {}
+		...metadataUserId ? { metadata: { user_id: metadataUserId } } : {}
 	};
 	return {
 		compaction: normalized.compaction,
@@ -8920,6 +8929,8 @@ function translateReasoningEffort(effort) {
 	if (effort === "low" || effort === "medium" || effort === "high" || effort === "xhigh" || effort === "max") return effort;
 }
 function resolveMetadataUserId(payload) {
+	const sessionAffinity = requestContext.getStore()?.sessionAffinity?.trim();
+	if (sessionAffinity) return sessionAffinity;
 	const metadataUserId = payload.metadata?.user_id;
 	if (metadataUserId?.trim()) return metadataUserId;
 	if (payload.safety_identifier?.trim()) return payload.safety_identifier;
@@ -10422,4 +10433,4 @@ server.route("/:provider/images", providerImageRoutes);
 //#endregion
 export { server };
 
-//# sourceMappingURL=server-DyVtfan1.js.map
+//# sourceMappingURL=server-s6RzK7Oo.js.map