@jeffreycao/copilot-api 2.1.7 → 2.1.9

package/dist/{server-s6RzK7Oo.js → server-tlEQmO0Y.js}

@@ -1,7 +1,7 @@
-import { A as isResponsesApiWebSocketEnabled, C as getClaudeTokenMultiplier, D as isAlphaSearchCodexPriorityEnabled, E as getResponsesTransportConfig, M as PATHS, O as isMessagesApiEnabled, S as getClaudeAutoModel, T as getMessageApiWebSearchModel, _ as resolveMappedModel, b as getAlphaSearchModel, d as getModelResponsesApiCompactThreshold$1, f as getReasoningEffortForModel, g as isGpt56OrAbove, h as isContextManagementEnabledForResponses, i as listEnabledProviders, k as isResponsesApiWebSearchEnabled, l as getExtraPromptForModel, m as isContextManagementEnabledForMessages, n as getRawProviderConfig, o as resolveEffectiveProviderType, p as getSmallModel, s as resolveProviderAuthType, t as getProviderConfig, u as getModelMappings, v as setModelMappings, w as getConfig, x as getAnthropicApiKey } from "./config-DMrKDHiF.js";
-import { A as buildCodexRequestHeaders, B as createPooledWebSocketStream, C as compactAutoContinuePromptStarts, D as compactTextOnlyGuard, E as compactSystemPromptStarts, F as createResponsesHttpEventStream, H as state, I as fetchResponsesWithLifecycle, L as createResponsesSafeStream, M as generateTraceId, N as requestContext, P as resolveTraceId$1, R as encodePoolKeyPart, U as HTTPError, V as createWebSocketUrl, W as forwardError, b as prepareInteractionHeaders, c as getUUID, d as isResponsesStream, f as parseUserIdMetadata, g as copilotHeaders, h as copilotBaseUrl, j as forwardCodexResponses, k as CODEX_API_BASE_URL, l as isAsyncIterable, o as generateRequestIdFromPayload, p as getCopilotUsage, r as setupCodexToken, s as getRootSessionId, u as isNullish, v as copilotWebSocketHeaders, w as compactMessageSections, x as prepareMessageProxyHeaders, y as prepareForCompact, z as isTerminalResponsesStreamChunk } from "./token-DmNSTe37.js";
+import { A as isResponsesApiWebSocketEnabled, C as getClaudeTokenMultiplier, D as isAlphaSearchCodexPriorityEnabled, E as getResponsesTransportConfig, M as PATHS, O as isMessagesApiEnabled, S as getClaudeAutoModel, T as getMessageApiWebSearchModel, _ as resolveMappedModel, b as getAlphaSearchModel, d as getModelResponsesApiCompactThreshold$1, f as getReasoningEffortForModel, g as isGpt56OrAbove, h as isContextManagementEnabledForResponses, i as listEnabledProviders, k as isResponsesApiWebSearchEnabled, l as getExtraPromptForModel, m as isContextManagementEnabledForMessages, n as getRawProviderConfig, o as resolveEffectiveProviderType, p as getSmallModel, s as resolveProviderAuthType, t as getProviderConfig, u as getModelMappings, v as setModelMappings, w as getConfig, x as getAnthropicApiKey } from "./config-CHzejzwY.js";
+import { A as buildCodexRequestHeaders, B as createPooledWebSocketStream, C as compactAutoContinuePromptStarts, D as compactTextOnlyGuard, E as compactSystemPromptStarts, F as createResponsesHttpEventStream, H as state, I as fetchResponsesWithLifecycle, L as createResponsesSafeStream, M as generateTraceId, N as requestContext, P as resolveTraceId$1, R as encodePoolKeyPart, U as HTTPError, V as createWebSocketUrl, W as forwardError, b as prepareInteractionHeaders, c as getUUID, d as isResponsesStream, f as parseUserIdMetadata, g as copilotHeaders, h as copilotBaseUrl, j as forwardCodexResponses, k as CODEX_API_BASE_URL, l as isAsyncIterable, o as generateRequestIdFromPayload, p as getCopilotUsage, r as setupCodexToken, s as getRootSessionId, u as isNullish, v as copilotWebSocketHeaders, w as compactMessageSections, x as prepareMessageProxyHeaders, y as prepareForCompact, z as isTerminalResponsesStreamChunk } from "./token-Ctj6sfG4.js";
 import { a as isDeferredToolName, c as parseMcpToolSearchSentinel, d as shouldEnableResponsesToolSearch, i as isBridgeToolSearchName, l as resolveBridgeToolSearchName, o as listDeferredToolNames, r as formatToolSearchBridgeArguments, s as normalizeToolSearchBridgeArguments, t as BRIDGE_TOOL_SEARCH_NAME, u as selectDeferredToolsByNames } from "./tool-search-Ds1vbmGG.js";
-import { i as toClientModelId, r as normalizeSdkModelId, t as findEndpointModel } from "./models-Do0TBzAz.js";
+import { i as toClientModelId, r as normalizeSdkModelId, t as findEndpointModel } from "./models-CJuY4JZ5.js";
 import consola from "consola";
 import { createHash } from "node:crypto";
 import fs, { readFileSync } from "node:fs";
@@ -640,8 +640,19 @@ const builtinProviderModelRegistry = new class BuiltinProviderModelRegistry {
 					output: 4.4
 				}
 			},
+			"glm-5.3": {
+				contextWindow: 1e6,
+				inputModalities: ["text"],
+				maxOutputTokens: 64e3,
+				pricing: {
+					cachedInput: .26,
+					input: 1.4,
+					output: 4.4
+				}
+			},
 			"grok-4.5": {
 				contextWindow: 5e5,
+				defaultReasoningEffort: "high",
 				inputModalities: ["text", "image"],
 				maxOutputTokens: 64e3,
 				pricing: { tiers: [{
@@ -653,7 +664,12 @@ const builtinProviderModelRegistry = new class BuiltinProviderModelRegistry {
 					cachedInput: 1,
 					input: 4,
 					output: 12
-				}] }
+				}] },
+				reasoningEfforts: [
+					"low",
+					"medium",
+					"high"
+				]
 			},
 			"deepseek-v4-flash": {
 				contextWindow: 1e6,
@@ -2892,6 +2908,21 @@ const setContextCacheControl = (part) => {
 };
 //#endregion
 //#region src/routes/provider/utils.ts
+const RESPONSES_REASONING_EFFORTS = new Set([
+	"none",
+	"minimal",
+	"low",
+	"medium",
+	"high",
+	"xhigh",
+	"max"
+]);
+const REASONING_EFFORT_ALIASES = {
+	minimal: "low",
+	xhigh: "high",
+	max: "high",
+	ultra: "high"
+};
 const applyModelDefaults = (payload, modelConfig) => {
 	payload.temperature ??= modelConfig?.temperature;
 	payload.top_p ??= modelConfig?.topP;
@@ -2900,6 +2931,24 @@ const applyModelDefaults = (payload, modelConfig) => {
 const applyMissingExtraBody = (payload, options) => {
 	for (const [key, value] of Object.entries(options.extraBody ?? {})) if (!Object.hasOwn(payload, key)) payload[key] = value;
 };
+const normalizeProviderResponsesReasoningEffort = (payload, providerConfig) => {
+	const currentEffort = payload.reasoning?.effort;
+	if (typeof currentEffort !== "string") return;
+	const modelConfig = providerConfig.models?.[payload.model];
+	const builtinModelConfig = builtinProviderModelRegistry.getModelConfig(providerConfig.name, payload.model);
+	const configuredEfforts = normalizeResponsesReasoningEfforts(modelConfig?.reasoningEfforts);
+	const builtinEfforts = normalizeResponsesReasoningEfforts(builtinModelConfig?.reasoningEfforts);
+	const supportedEfforts = configuredEfforts.length > 0 ? configuredEfforts : builtinEfforts;
+	if (supportedEfforts.length === 0 || supportedEfforts.includes(currentEffort)) return;
+	const aliasedEffort = REASONING_EFFORT_ALIASES[currentEffort];
+	const fallbackEffort = aliasedEffort && supportedEfforts.includes(aliasedEffort) ? aliasedEffort : selectSupportedReasoningEffort(supportedEfforts, modelConfig?.defaultReasoningEffort, builtinModelConfig?.defaultReasoningEffort);
+	if (payload.reasoning && fallbackEffort) payload.reasoning.effort = fallbackEffort;
+};
+const normalizeResponsesReasoningEfforts = (efforts) => (efforts ?? []).filter((effort) => RESPONSES_REASONING_EFFORTS.has(effort));
+const selectSupportedReasoningEffort = (supportedEfforts, ...defaults) => {
+	for (const defaultEffort of defaults) if (defaultEffort && supportedEfforts.includes(defaultEffort)) return defaultEffort;
+	return supportedEfforts[0];
+};
 //#endregion
 //#region src/routes/provider/chat-completions/handler.ts
 const logger$14 = createHandlerLogger("provider-chat-completions-handler");
@@ -3449,16 +3498,15 @@ function mapOpenAIStopReasonToAnthropic(finishReason) {
 		content_filter: "end_turn"
 	}[finishReason];
 }
-//#endregion
-//#region src/routes/messages/non-stream-translation.ts
-const THINKING_TEXT = "Thinking...";
 const COPILOT_TOOL_CONTENT_SUPPORT_TYPE = ["array", "image"];
 function translateToOpenAI(payload, options = {}) {
 	const modelId = payload.model;
 	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();
+	const { sessionId: metadataPromptCacheKey } = parseUserIdMetadata(payload.metadata?.user_id);
+	const sessionAffinity = requestContext.getStore()?.sessionAffinity?.trim() || null;
+	const promptCacheKey = metadataPromptCacheKey ?? sessionAffinity;
 	return {
 		model: modelId,
 		messages: translateAnthropicMessagesToOpenAI(payload, modelId, {
@@ -3600,7 +3648,7 @@ function handleAssistantMessage(message, modelId, capabilities) {
 	if (message.content.length === 0) return [];
 	const toolUseBlocks = message.content.filter((block) => block.type === "tool_use");
 	let thinkingBlocks = message.content.filter((block) => block.type === "thinking");
-	if (modelId.startsWith("claude")) thinkingBlocks = thinkingBlocks.filter((b) => b.thinking && b.thinking !== "Thinking..." && b.signature && !b.signature.includes("@"));
+	if (modelId.startsWith("claude")) thinkingBlocks = thinkingBlocks.filter((b) => b.thinking !== "Thinking..." && b.signature && !b.signature.includes("@"));
 	const thinkingContents = thinkingBlocks.filter((b) => b.thinking && b.thinking !== "Thinking...").map((b) => b.thinking);
 	const allThinkingContent = thinkingContents.length > 0 ? thinkingContents.join("\n\n") : void 0;
 	const signature = thinkingBlocks.find((b) => b.signature)?.signature;
@@ -3762,7 +3810,7 @@ function getAnthropicThinkBlocks(reasoningText, reasoningOpaque) {
 	}];
 	if (reasoningOpaque && reasoningOpaque.length > 0) return [{
 		type: "thinking",
-		thinking: THINKING_TEXT,
+		thinking: "",
 		signature: reasoningOpaque
 	}];
 	return [];
@@ -4612,7 +4660,7 @@ function handleReasoningOpaque(delta, events, state) {
 			index: state.contentBlockIndex,
 			delta: {
 				type: "thinking_delta",
-				thinking: THINKING_TEXT
+				thinking: ""
 			}
 		}, {
 			type: "content_block_delta",
@@ -5193,7 +5241,7 @@ const createCompactionThinkingBlock = (item) => {
 	if (!item.id || !item.encrypted_content) return null;
 	return {
 		type: "thinking",
-		thinking: THINKING_TEXT,
+		thinking: "",
 		signature: encodeCompactionCarrierSignature({
 			id: item.id,
 			encrypted_content: item.encrypted_content
@@ -5734,7 +5782,6 @@ const openFunctionCallBlock = (state, params) => {
 	}
 	const { blockIndex } = functionCallState;
 	if (!state.openBlocks.has(blockIndex)) {
-		closeOpenBlocks(state, events);
 		events.push({
 			type: "content_block_start",
 			index: blockIndex,
@@ -5964,6 +6011,7 @@ const createCompactionContextManagement = (compactThreshold) => [{
 	compact_threshold: compactThreshold
 }];
 const applyResponsesApiContextManagement = (payload, maxPromptTokens, options) => {
+	if (!payload.model.startsWith("gpt")) return false;
 	if (responsesUtilsDependencies.isGpt56OrAbove(payload.model)) return false;
 	if (hasTerminalCompactionTrigger(payload)) return isContextManagementEnabledForSource(options.source);
 	if (payload.context_management !== void 0) return true;
@@ -6561,6 +6609,7 @@ async function handleProviderMessagesForProvider(c, options) {
 const handleOpenAIResponsesProviderWebSearchMessages = async (c, options) => {
 	const { modelConfig, payload, provider, providerConfig, usageEndpoint } = options;
 	const responsesPayload = prepareWebSearchResponsesPayload(payload);
+	normalizeProviderResponsesReasoningEffort(responsesPayload, providerConfig);
 	debugJson(logger$10, "provider.messages.responses.web_search.request", {
 		payload: responsesPayload,
 		provider
@@ -6624,6 +6673,7 @@ const handleOpenAIResponsesProviderMessages = async (c, options) => {
 	const selectedModel = providerConfig.name === "codex" ? getModels().data.find((model) => model.id === payload.model) : void 0;
 	const wantsStream = payload.stream === true;
 	const responsesPayload = translateAnthropicMessagesToResponsesPayload(payload);
+	normalizeProviderResponsesReasoningEffort(responsesPayload, providerConfig);
 	if (providerConfig.name === "codex" && !wantsStream) responsesPayload.stream = true;
 	if (applyResponsesApiContextManagement(responsesPayload, selectedModel?.capabilities.limits.max_prompt_tokens, { source: "messages" })) compactInputByLatestCompaction(responsesPayload);
 	debugJson(logger$10, "provider.messages.responses.request", {
@@ -8142,19 +8192,27 @@ async function getProviderCodexCandidates(provider, requestHeaders) {
 function isMessagesFallbackProviderType(type) {
 	return type === "anthropic" || type === "openai-compatible";
 }
+function describeProviderAdapter(type) {
+	if (type === "anthropic") return "Messages";
+	if (type === "openai-responses") return "Messages-to-Responses";
+	return "Messages-to-Chat";
+}
 function createProviderCodexCandidate(providerConfig, modelId, remoteModel, modelConfig, effectiveType) {
 	const builtinModelConfig = builtinProviderModelRegistry.getModelConfig(providerConfig.name, modelId);
 	const configuredReasoningEfforts = normalizeReasoningEfforts(modelConfig?.reasoningEfforts);
-	const reasoningEfforts = configuredReasoningEfforts.length > 0 ? configuredReasoningEfforts : normalizeRemoteReasoningEfforts(remoteModel);
+	const remoteReasoningEfforts = normalizeRemoteReasoningEfforts(remoteModel);
+	const builtinReasoningEfforts = normalizeReasoningEfforts(builtinModelConfig?.reasoningEfforts);
+	const reasoningEfforts = configuredReasoningEfforts.length > 0 ? configuredReasoningEfforts : remoteReasoningEfforts.length > 0 ? remoteReasoningEfforts : builtinReasoningEfforts;
 	const configuredModalities = normalizeInputModalities(modelConfig?.inputModalities);
 	const remoteModalities = normalizeInputModalities(remoteModel?.input_modalities ?? remoteModel?.modalities);
 	const builtinModalities = normalizeInputModalities(builtinModelConfig?.inputModalities);
 	const displayName = getStringField$1(remoteModel ?? {}, "display_name") ?? getStringField$1(remoteModel ?? {}, "name") ?? modelId;
-	const adapterName = effectiveType === "anthropic" ? "Messages" : "Messages-to-Chat";
+	const adapterName = describeProviderAdapter(effectiveType);
+	const requiresCatalogMatch = effectiveType === "openai-responses" && modelId.startsWith("gpt");
 	return {
 		slug: `${providerConfig.name}/${modelId}`,
 		catalogSlug: modelId,
-		catalogMatchRequired: effectiveType === "openai-responses",
+		catalogMatchRequired: requiresCatalogMatch,
 		providerName: providerConfig.name,
 		displayName: `${displayName} (${providerConfig.name})`,
 		description: `${displayName} through the ${providerConfig.name} ${adapterName} adapter.`,
@@ -8167,7 +8225,7 @@ function createProviderCodexCandidate(providerConfig, modelId, remoteModel, mode
 		maxOutputTokens: positiveNumber(modelConfig?.maxOutputTokens ?? getFirstPositiveNumber(remoteModel, ["max_output_tokens"]) ?? builtinModelConfig?.maxOutputTokens, 32e3),
 		inputModalities: resolveInputModalities(providerConfig.name, configuredModalities, remoteModalities, builtinModalities),
 		reasoningEfforts,
-		defaultReasoningEffort: selectDefaultReasoningEffort(reasoningEfforts, modelConfig?.defaultReasoningEffort)
+		defaultReasoningEffort: selectDefaultReasoningEffort(reasoningEfforts, modelConfig?.defaultReasoningEffort ?? builtinModelConfig?.defaultReasoningEffort)
 	};
 }
 function normalizeRemoteReasoningEfforts(model) {
@@ -9937,7 +9995,8 @@ async function handleProviderResponsesForProvider(c, options) {
 		type: "invalid_request_error"
 	} }, 400);
 	const effectiveType = resolveEffectiveProviderType(providerConfig, payload.model);
-	if (effectiveType === "anthropic" || effectiveType === "openai-compatible") return await handleResponsesViaMessages(c, {
+	normalizeProviderResponsesReasoningEffort(payload, providerConfig);
+	if (shouldFallbackToMessages$1(c, payload.model, effectiveType)) return await handleResponsesViaMessages(c, {
 		payload,
 		publicModel: options.publicModel ?? payload.model,
 		targetModel: `${provider}/${payload.model}`
@@ -9978,6 +10037,11 @@ async function handleProviderResponsesForProvider(c, options) {
 	recordUsage(normalizeResponsesUsage((await upstreamResponse.clone().json()).usage));
 	return createProviderProxyResponse(upstreamResponse);
 }
+const shouldFallbackToMessages$1 = (c, modelId, effectiveType) => {
+	if (effectiveType === "anthropic" || effectiveType === "openai-compatible") return true;
+	if (isCodexUserAgent(c.req.header("user-agent"))) return !modelId.startsWith("gpt");
+	return false;
+};
 const createProviderResponsesUsageRecorder = (payload, provider, modelConfig, pricingCurrency) => {
 	const sessionAffinity = requestContext.getStore()?.sessionAffinity?.trim() || null;
 	return createProviderTokenUsageRecorder({
@@ -10433,4 +10497,4 @@ server.route("/:provider/images", providerImageRoutes);
 //#endregion
 export { server };
 
-//# sourceMappingURL=server-s6RzK7Oo.js.map
+//# sourceMappingURL=server-tlEQmO0Y.js.map