@jeffreycao/copilot-api 2.7.1 → 2.7.2

package/dist/{server-BYG25POu.js → server-DbkseGH6.js}

@@ -1,10 +1,10 @@
-import { a as getAnthropicApiKey, c as getConfig, f as isAlphaSearchCodexPriorityEnabled, h as isResponsesApiWebSocketEnabled, i as getAlphaSearchModel, l as getMessageApiWebSearchModel, m as isResponsesApiWebSearchEnabled, o as getClaudeAutoModel, p as isMessagesApiEnabled, s as getClaudeTokenMultiplier, u as getUpstreamTransportConfig, v as reloadConfig, x as PATHS } from "./config-store-CARrUljo.js";
-import { C as resolveMappedModel, S as isGpt56OrAbove, _ as getReasoningEffortForModel, b as isContextManagementEnabledForMessages, d as getOpencodeGoModelRecords, g as getModelResponsesApiCompactThreshold$1, h as getModelMappings, i as listEnabledProviders, l as getOpencodeGoModelConfig, m as getExtraPromptForModel, n as getRawProviderConfig, o as resolveEffectiveProviderType, s as resolveProviderConfigForModel, t as getProviderConfig, u as getOpencodeGoModelIds, v as getSmallModel, w as setModelMappings, x as isContextManagementEnabledForResponses, y as getSmallModelForProvider } from "./config-B2OaAnpL.js";
-import { i as HTTPError, r as isGitHubCopilotEnabled, s as forwardError } from "./github-copilot-provider-DlQZgaCT.js";
-import { $ as state, A as compactAutoContinuePromptStarts, B as CODEX_API_BASE_URL, C as copilotHeaders, D as prepareInteractionHeaders, E as prepareForCompact, F as createAuthMiddleware, G as resolveTraceId$1, H as forwardCodexResponses, I as getConfiguredAdminApiKeys, J as createResponsesSafeStream, K as fetchUpstreamWithLifecycle, N as compactSystemPromptStarts, O as prepareMessageProxyHeaders, P as compactTextOnlyGuard, Q as createWebSocketUrl, S as copilotBaseUrl, T as copilotWebSocketHeaders, U as generateTraceId, V as buildCodexRequestHeaders, W as requestContext, X as getResponsesStreamTerminalDisposition, Y as encodePoolKeyPart, Z as createPooledWebSocketStream, _ as parseUserIdMetadata, b as readGitHubToken, d as generateRequestIdFromPayload, f as getRootSessionId, g as isResponsesStream, h as isNullish, j as compactMessageSections, m as isAsyncIterable, n as invalidateCodexRuntime, p as getUUID, q as createResponsesHttpEventStream, s as setupCodexToken, tt as stripInternalRequestHeaders, u as stopCopilotRefreshLoop, v as getCopilotUsage, x as readGitHubTokenFromEnv } from "./token-Cx07G8-R.js";
+import { a as getAnthropicApiKey, c as getConfig, f as isAlphaSearchCodexPriorityEnabled, h as isResponsesApiWebSocketEnabled, i as getAlphaSearchModel, l as getMessageApiWebSearchModel, m as isResponsesApiWebSearchEnabled, o as getClaudeAutoModel, p as isMessagesApiEnabled, s as getClaudeTokenMultiplier, u as getUpstreamTransportConfig, v as reloadConfig, x as PATHS } from "./config-store-CMgB1mCV.js";
+import { C as isGpt56OrAbove, S as isContextManagementEnabledForResponses, T as setModelMappings, _ as getModelResponsesApiCompactThreshold$1, b as getSmallModelForProvider, d as getOpencodeGoModelIds, f as getOpencodeGoModelRecords, g as getModelMappings, h as getExtraPromptForModel, i as listEnabledProviders, l as getModelsDevModelMaxOutputTokens, n as getRawProviderConfig, o as resolveEffectiveProviderType, s as resolveProviderConfigForModel, t as getProviderConfig, u as getOpencodeGoModelConfig, v as getReasoningEffortForModel, w as resolveMappedModel, x as isContextManagementEnabledForMessages, y as getSmallModel } from "./config-DIDLVXMU.js";
+import { i as HTTPError, r as isGitHubCopilotEnabled, s as forwardError } from "./github-copilot-provider-CLEAHyT-.js";
+import { $ as createResponsesHttpEventStream, A as compactSystemPromptStarts, C as prepareForCompact, D as compactAutoContinuePromptStarts, H as readGitHubToken, I as getXaiAccessToken, J as forwardCodexResponses, K as CODEX_API_BASE_URL, M as createAuthMiddleware, N as getConfiguredAdminApiKeys, O as compactMessageSections, Q as fetchUpstreamWithLifecycle, S as copilotWebSocketHeaders, T as prepareMessageProxyHeaders, U as readGitHubTokenFromEnv, X as requestContext, Y as generateTraceId, Z as resolveTraceId$1, _ as parseUserIdMetadata, at as state, b as copilotHeaders, d as generateRequestIdFromPayload, et as createResponsesSafeStream, f as getRootSessionId, g as isResponsesStream, h as isNullish, it as createWebSocketUrl, j as compactTextOnlyGuard, m as isAsyncIterable, n as invalidateCodexRuntime, nt as getResponsesStreamTerminalDisposition, p as getUUID, q as buildCodexRequestHeaders, rt as createPooledWebSocketStream, s as setupCodexToken, st as stripInternalRequestHeaders, tt as encodePoolKeyPart, u as stopCopilotRefreshLoop, v as getCopilotUsage, w as prepareInteractionHeaders, y as copilotBaseUrl } from "./token-BjzzuPSb.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-BgFcDbNQ.js";
-import { n as isProviderCodexModelVisible, t as getProviderCodexModels } from "./provider-management-z3m7cUkp.js";
-import { a as normalizeSdkModelId, n as stopModelsRefreshLoop, o as toClientModelId, r as findEndpointModel, t as setupCopilotRuntime } from "./copilot-runtime-B7T_EuGD.js";
+import { n as isProviderCodexModelVisible, t as getProviderCodexModels } from "./provider-management-BjU3aJnC.js";
+import { a as normalizeSdkModelId, n as stopModelsRefreshLoop, o as toClientModelId, r as findEndpointModel, t as setupCopilotRuntime } from "./copilot-runtime-CRTYwDWy.js";
 import consola from "consola";
 import { createHash } from "node:crypto";
 import fs, { createWriteStream, readFileSync, rmSync } from "node:fs";
@@ -351,6 +351,13 @@ async function resolveProviderConfig(providerName, getAzureAccessToken = getDefa
 		};
 	}
 	const providerConfig = getProviderConfig(normalizedProviderName);
+	if (providerConfig?.name === "xai" && providerConfig.authType === "oauth2") {
+		const accessToken = await getXaiAccessToken();
+		return accessToken ? {
+			...providerConfig,
+			apiKey: accessToken
+		} : null;
+	}
 	if (providerConfig?.authType !== "azure-entra") return providerConfig;
 	try {
 		return {
@@ -442,6 +449,28 @@ const CODEX_LUNA_PRICING = { tiers: [{
 }] };
 const builtinProviderModelRegistry = new class BuiltinProviderModelRegistry {
 	static catalog = {
+		xai: { "grok-4.7": {
+			contextWindow: 5e5,
+			maxOutputTokens: 5e5,
+			inputModalities: ["text", "image"],
+			reasoningEfforts: [
+				"low",
+				"medium",
+				"high",
+				"xhigh"
+			],
+			defaultReasoningEffort: "high",
+			pricing: { tiers: [{
+				cachedInput: .5,
+				input: 2,
+				output: 6,
+				maxInputTokens: 2e5
+			}, {
+				cachedInput: 1,
+				input: 4,
+				output: 12
+			}] }
+		} },
 		codex: {
 			"codex-auto-review": { pricing: CODEX_LUNA_PRICING },
 			"gpt-reserve": { pricing: CODEX_LUNA_PRICING },
@@ -759,6 +788,20 @@ const builtinProviderModelRegistry = new class BuiltinProviderModelRegistry {
 		return value.trim().toLowerCase();
 	}
 }();
+function getBuiltinProviderModelRecords(provider) {
+	return builtinProviderModelRegistry.getModelIds(provider).map((id) => {
+		const config = builtinProviderModelRegistry.getModelConfig(provider, id);
+		return {
+			id,
+			name: id,
+			object: "model",
+			context_window: config?.contextWindow,
+			max_output_tokens: config?.maxOutputTokens,
+			input_modalities: config?.inputModalities,
+			reasoning_efforts: config?.reasoningEfforts
+		};
+	});
+}
 //#endregion
 //#region src/lib/sqlite.ts
 const MINIMUM_NODE_SQLITE_VERSION = "22.13.0";
@@ -4141,6 +4184,7 @@ function translateToOpenAI(payload, options = {}) {
 		toolContentSupportType: options.toolContentSupportType ?? []
 	};
 	const tools = translateAnthropicToolsToOpenAI(payload.tools);
+	const parallelToolCalls = translateParallelToolCalls(payload.tool_choice);
 	return {
 		model: modelId,
 		messages: translateAnthropicMessagesToOpenAI(payload, modelId, capabilities),
@@ -4154,7 +4198,8 @@ function translateToOpenAI(payload, options = {}) {
 		tool_choice: tools && tools.length > 0 ? translateAnthropicToolChoiceToOpenAI(payload.tool_choice) : void 0,
 		thinking_budget: thinkingBudget,
 		...reasoningEffort ? { reasoning_effort: reasoningEffort } : {},
-		...promptCacheKey ? { prompt_cache_key: promptCacheKey } : {}
+		...promptCacheKey ? { prompt_cache_key: promptCacheKey } : {},
+		...parallelToolCalls === void 0 ? {} : { parallel_tool_calls: parallelToolCalls }
 	};
 }
 function getReasoningEffort(payload, options) {
@@ -4394,6 +4439,10 @@ function translateAnthropicToolChoiceToOpenAI(anthropicToolChoice) {
 		default: return;
 	}
 }
+function translateParallelToolCalls(toolChoice) {
+	if (typeof toolChoice?.disable_parallel_tool_use !== "boolean") return;
+	return !toolChoice.disable_parallel_tool_use;
+}
 function translateToAnthropic(response) {
 	const assistantContentBlocks = [];
 	let stopReason = response.choices[0]?.finish_reason ?? null;
@@ -5427,17 +5476,24 @@ const translateAnthropicMessagesToResponsesPayload = (payload, subagentAgentId)
 		metadata: payload.metadata ? { ...payload.metadata } : null,
 		stream: payload.stream ?? null,
 		store: false,
-		parallel_tool_calls: true,
+		parallel_tool_calls: !payload.tool_choice?.disable_parallel_tool_use,
 		reasoning: {
 			effort: resolveReasoningEffort(payload),
-			summary: "auto",
+			summary: payload.model.includes("grok") ? "concise" : "auto",
 			context: isSupportAllTurns(payload) ? "all_turns" : "auto"
 		},
 		include: ["reasoning.encrypted_content"]
 	};
 	if (translatedTools && translatedTools.length > 0) responsesPayload.tool_choice = toolChoice;
 	if (hasOriginalTools) responsesPayload.prompt_cache_key = promptCacheKey;
-	if (responsesPayload.model.includes("grok")) delete responsesPayload.metadata;
+	if (responsesPayload.model.includes("grok")) {
+		delete responsesPayload.metadata;
+		delete responsesPayload.temperature;
+		delete responsesPayload.top_p;
+		delete responsesPayload.max_output_tokens;
+		if (responsesPayload.instructions) input.unshift(createMessage("system", responsesPayload.instructions));
+		delete responsesPayload.instructions;
+	}
 	return responsesPayload;
 };
 const encodeCompactionCarrierSignature = (compaction) => {
@@ -6552,6 +6608,7 @@ const MESSAGES_TOOL_CALL_TIPS = [
 	"- For long-running commands, keep the returned `session_id` and poll it with `tools.write_stdin` until the command finishes; do not redirect output to a temp file and read it back in a second call.",
 	"- If `functions__exec` returns `aborted`, retry at most 3 times. After 3 failures, terminate immediately and inform the user that `functions__exec` is unavailable."
 ].join("\n");
+const MESSAGES_BATCH_TOOL_CALL_TIPS = ["- Parallel tool calls are disabled for this request. Batch independent commands in one functions__exec call; keep dependent operations sequential.", "const results = await Promise.allSettled([tools.exec_command({cmd: \"git status --short\"})]); for (const result of results) text(JSON.stringify(result.status === \"fulfilled\" ? result.value : {error: String(result.reason)}))"].join("\n");
 const JSON_OUTPUT_CONSTRAINT = "Do not wrap the JSON in markdown code fences and do not add any text outside the JSON object.";
 const buildOutputFormatInstruction = (format) => {
 	if (!format) return null;
@@ -6588,6 +6645,7 @@ function translateResponsesToMessages(payload, options) {
 	const normalized = normalizeResponsesInput(payload.input);
 	const outputFormatInstruction = buildOutputFormatInstruction(payload.text?.format);
 	const { messages, system } = translateInputToAnthropic(normalized.input, registry, payload.instructions, payload.input, options.toolCallTips ?? false);
+	if (options.toolCallTips && payload.parallel_tool_calls === false) appendToolCallTips(system, MESSAGES_BATCH_TOOL_CALL_TIPS);
 	if (normalized.compaction) messages.push({
 		role: "user",
 		content: MESSAGES_COMPACTION_PROMPT
@@ -6603,7 +6661,7 @@ function translateResponsesToMessages(payload, options) {
 	const messagesPayload = {
 		model: options.model,
 		messages,
-		max_tokens: Math.max(1, payload.max_output_tokens ?? 32e3),
+		max_tokens: Math.max(1, payload.max_output_tokens ?? resolveDefaultMaxOutputTokens(options.model)),
 		stream: payload.stream ?? false,
 		temperature: payload.temperature ?? void 0,
 		top_p: payload.top_p ?? void 0,
@@ -6614,6 +6672,10 @@ function translateResponsesToMessages(payload, options) {
 		...payload.service_tier === "auto" || payload.service_tier === "standard_only" ? { service_tier: payload.service_tier } : {},
 		...metadataUserId ? { metadata: { user_id: metadataUserId } } : {}
 	};
+	if (payload.parallel_tool_calls != null && registry.tools.length > 0 && messagesPayload.tool_choice?.type !== "none") messagesPayload.tool_choice = {
+		...messagesPayload.tool_choice ?? { type: "auto" },
+		disable_parallel_tool_use: !payload.parallel_tool_calls
+	};
 	return {
 		compaction: normalized.compaction,
 		messagesPayload,
@@ -6622,6 +6684,17 @@ function translateResponsesToMessages(payload, options) {
 		registry
 	};
 }
+function resolveDefaultMaxOutputTokens(model) {
+	const alias = parseProviderModelAlias(model);
+	const providerConfig = alias ? getRawProviderConfig(alias.provider) : null;
+	const catalogMaxOutputTokens = alias ? getModelsDevModelMaxOutputTokens(providerConfig?.modelsDevProviderId || alias.provider, alias.model) : void 0;
+	const builtinModelConfig = alias ? builtinProviderModelRegistry.getModelConfig(alias.provider, alias.model) : void 0;
+	return (alias && (providerConfig || builtinModelConfig || catalogMaxOutputTokens) ? [
+		providerConfig?.models?.[alias.model]?.maxOutputTokens,
+		catalogMaxOutputTokens,
+		builtinModelConfig?.maxOutputTokens
+	] : [findEndpointModel(model)?.capabilities.limits.max_output_tokens]).find((limit) => typeof limit === "number" && Number.isInteger(limit) && limit > 0) ?? 32e3;
+}
 function translateAnthropicToResponses(response, context) {
 	const finish = translateStopReason(response.stop_reason);
 	const output = context.compaction ? translateCompactionOutput(response) : translateAssistantOutput(response, context.registry);
@@ -6826,16 +6899,16 @@ function translateInputToAnthropic(input, registry, instructions, originalInput,
 		system
 	};
 }
-function appendToolCallTips(system) {
+function appendToolCallTips(system, tips = MESSAGES_TOOL_CALL_TIPS) {
 	const lastSystemBlock = system.at(-1);
 	if (!lastSystemBlock) {
 		system.push({
 			type: "text",
-			text: MESSAGES_TOOL_CALL_TIPS
+			text: tips
 		});
 		return;
 	}
-	lastSystemBlock.text = `${lastSystemBlock.text}\n\n${MESSAGES_TOOL_CALL_TIPS}`;
+	lastSystemBlock.text = `${lastSystemBlock.text}\n\n${tips}`;
 }
 function translateInputItems(input, messages, system, registry) {
 	let userMessageSeen = false;
@@ -10983,13 +11056,6 @@ function getStringField(model, field) {
 	const value = model[field];
 	return typeof value === "string" && value.trim() ? value : void 0;
 }
-function getBuiltinProviderModelRecords(provider) {
-	return builtinProviderModelRegistry.getModelIds(provider).map((id) => ({
-		id,
-		name: id,
-		object: "model"
-	}));
-}
 function getFallbackProviderModelRecords(provider, reason, details = {}) {
 	const fallbackModels = getBuiltinProviderModelRecords(provider);
 	logger$7.warn(`models.provider.fallback_${reason}`, {
@@ -11022,6 +11088,7 @@ function normalizeProviderModels(provider, models) {
 }
 async function getProviderModelRecords(providerConfig, requestHeaders) {
 	if (providerConfig.name === "opencode-go") return getOpencodeGoModelRecords();
+	if (providerConfig.name === "xai" && providerConfig.authType === "oauth2") return getBuiltinProviderModelRecords("xai");
 	try {
 		const response = await forwardProviderModels(providerConfig, requestHeaders);
 		if (!response.ok) return getFallbackProviderModelRecords(providerConfig.name, "non_ok", { statusCode: response.status });
@@ -11128,7 +11195,7 @@ async function getProviderCodexCandidates(provider, requestHeaders) {
 			const id = getStringField(model, "id");
 			return id ? [[id, model]] : [];
 		}));
-		const modelIds = /* @__PURE__ */ new Set([...remoteById.keys(), ...Object.keys(providerConfig.models ?? {})]);
+		const modelIds = /* @__PURE__ */ new Set([...remoteById.keys(), ...providerConfig.name === "xai" && providerConfig.authType === "oauth2" ? [] : Object.keys(providerConfig.models ?? {})]);
 		const candidates = [];
 		for (const modelId of modelIds) {
 			const effectiveType = resolveEffectiveProviderType(providerConfig, modelId);
@@ -11178,7 +11245,7 @@ function createProviderCodexCandidate(providerConfig, modelId, remoteModel, mode
 			"max_context_length",
 			"max_model_len"
 		]) ?? builtinModelConfig?.contextWindow, 256e3),
-		maxOutputTokens: positiveNumber(modelConfig?.maxOutputTokens ?? getFirstPositiveNumber(remoteModel, ["max_output_tokens"]) ?? builtinModelConfig?.maxOutputTokens, 32e3),
+		maxOutputTokens: positiveNumber(modelConfig?.maxOutputTokens ?? getFirstPositiveNumber(remoteModel, ["max_output_tokens"]) ?? getModelsDevModelMaxOutputTokens(providerConfig.modelsDevProviderId || providerConfig.name, modelId) ?? builtinModelConfig?.maxOutputTokens, 32e3),
 		inputModalities: resolveInputModalities(providerConfig.name, configuredModalities, remoteModalities, builtinModalities),
 		reasoningEfforts,
 		defaultReasoningEffort: selectDefaultReasoningEffort(reasoningEfforts, modelConfig?.defaultReasoningEffort ?? builtinModelConfig?.defaultReasoningEffort)
@@ -11362,6 +11429,11 @@ providerModelRoutes.get("/", async (c) => {
 				has_more: false
 			});
 		}
+		if (providerConfig.name === "xai" && providerConfig.authType === "oauth2") return c.json({
+			object: "list",
+			data: getBuiltinProviderModelRecords("xai"),
+			has_more: false
+		});
 		const upstreamResponse = await forwardProviderModels(providerConfig, c.req.raw.headers);
 		logger$4.debug("provider.models.response", {
 			provider,
@@ -12878,4 +12950,4 @@ createServer();
 //#endregion
 export { createServer };
 
-//# sourceMappingURL=server-BYG25POu.js.map
+//# sourceMappingURL=server-DbkseGH6.js.map