@mariozechner/pi-ai 0.31.1 → 0.32.0

package/dist/providers/google-vertex.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"google-vertex.d.ts","sourceRoot":"","sources":["../../src/providers/google-vertex.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAKX,cAAc,EACd,aAAa,EAIb,MAAM,aAAa,CAAC;AAGrB,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAGlE,MAAM,WAAW,mBAAoB,SAAQ,aAAa;IACzD,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,KAAK,CAAC;IACrC,QAAQ,CAAC,EAAE;QACV,OAAO,EAAE,OAAO,CAAC;QACjB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,KAAK,CAAC,EAAE,mBAAmB,CAAC;KAC5B,CAAC;IACF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;CAClB;AAeD,eAAO,MAAM,kBAAkB,EAAE,cAAc,CAAC,eAAe,CAqN9D,CAAC","sourcesContent":["import {\n\ttype GenerateContentConfig,\n\ttype GenerateContentParameters,\n\tGoogleGenAI,\n\ttype ThinkingConfig,\n\tThinkingLevel,\n} from \"@google/genai\";\nimport { calculateCost } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingContent,\n\tToolCall,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport type { GoogleThinkingLevel } from \"./google-gemini-cli.js\";\nimport { convertMessages, convertTools, mapStopReason, mapToolChoice } from \"./google-shared.js\";\n\nexport interface GoogleVertexOptions extends StreamOptions {\n\ttoolChoice?: \"auto\" | \"none\" | \"any\";\n\tthinking?: {\n\t\tenabled: boolean;\n\t\tbudgetTokens?: number; // -1 for dynamic, 0 to disable\n\t\tlevel?: GoogleThinkingLevel;\n\t};\n\tproject?: string;\n\tlocation?: string;\n}\n\nconst API_VERSION = \"v1\";\n\nconst THINKING_LEVEL_MAP: Record<GoogleThinkingLevel, ThinkingLevel> = {\n\tTHINKING_LEVEL_UNSPECIFIED: ThinkingLevel.THINKING_LEVEL_UNSPECIFIED,\n\tMINIMAL: ThinkingLevel.MINIMAL,\n\tLOW: ThinkingLevel.LOW,\n\tMEDIUM: ThinkingLevel.MEDIUM,\n\tHIGH: ThinkingLevel.HIGH,\n};\n\n// Counter for generating unique tool call IDs\nlet toolCallCounter = 0;\n\nexport const streamGoogleVertex: StreamFunction<\"google-vertex\"> = (\n\tmodel: Model<\"google-vertex\">,\n\tcontext: Context,\n\toptions?: GoogleVertexOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"google-vertex\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\tconst project = resolveProject(options);\n\t\t\tconst location = resolveLocation(options);\n\t\t\tconst client = createClient(model, project, location);\n\t\t\tconst params = buildParams(model, context, options);\n\t\t\tconst googleStream = await client.models.generateContentStream(params);\n\n\t\t\tstream.push({ type: \"start\", partial: output });\n\t\t\tlet currentBlock: TextContent | ThinkingContent | null = null;\n\t\t\tconst blocks = output.content;\n\t\t\tconst blockIndex = () => blocks.length - 1;\n\t\t\tfor await (const chunk of googleStream) {\n\t\t\t\tconst candidate = chunk.candidates?.[0];\n\t\t\t\tif (candidate?.content?.parts) {\n\t\t\t\t\tfor (const part of candidate.content.parts) {\n\t\t\t\t\t\tif (part.text !== undefined) {\n\t\t\t\t\t\t\tconst isThinking = part.thought === true;\n\t\t\t\t\t\t\tif (\n\t\t\t\t\t\t\t\t!currentBlock ||\n\t\t\t\t\t\t\t\t(isThinking && currentBlock.type !== \"thinking\") ||\n\t\t\t\t\t\t\t\t(!isThinking && currentBlock.type !== \"text\")\n\t\t\t\t\t\t\t) {\n\t\t\t\t\t\t\t\tif (currentBlock) {\n\t\t\t\t\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blocks.length - 1,\n\t\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tif (isThinking) {\n\t\t\t\t\t\t\t\t\tcurrentBlock = { type: \"thinking\", thinking: \"\", thinkingSignature: undefined };\n\t\t\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\t\t\tstream.push({ type: \"thinking_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\tcurrentBlock = { type: \"text\", text: \"\" };\n\t\t\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (currentBlock.type === \"thinking\") {\n\t\t\t\t\t\t\t\tcurrentBlock.thinking += part.text;\n\t\t\t\t\t\t\t\tcurrentBlock.thinkingSignature = part.thoughtSignature;\n\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\tdelta: part.text,\n\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\tcurrentBlock.text += part.text;\n\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\tdelta: part.text,\n\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tif (part.functionCall) {\n\t\t\t\t\t\t\tif (currentBlock) {\n\t\t\t\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tcurrentBlock = null;\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\tconst providedId = part.functionCall.id;\n\t\t\t\t\t\t\tconst needsNewId =\n\t\t\t\t\t\t\t\t!providedId || output.content.some((b) => b.type === \"toolCall\" && b.id === providedId);\n\t\t\t\t\t\t\tconst toolCallId = needsNewId\n\t\t\t\t\t\t\t\t? `${part.functionCall.name}_${Date.now()}_${++toolCallCounter}`\n\t\t\t\t\t\t\t\t: providedId;\n\n\t\t\t\t\t\t\tconst toolCall: ToolCall = {\n\t\t\t\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\t\t\t\tid: toolCallId,\n\t\t\t\t\t\t\t\tname: part.functionCall.name || \"\",\n\t\t\t\t\t\t\t\targuments: part.functionCall.args as Record<string, any>,\n\t\t\t\t\t\t\t\t...(part.thoughtSignature && { thoughtSignature: part.thoughtSignature }),\n\t\t\t\t\t\t\t};\n\n\t\t\t\t\t\t\toutput.content.push(toolCall);\n\t\t\t\t\t\t\tstream.push({ type: \"toolcall_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\tdelta: JSON.stringify(toolCall.arguments),\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\tstream.push({ type: \"toolcall_end\", contentIndex: blockIndex(), toolCall, partial: output });\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tif (candidate?.finishReason) {\n\t\t\t\t\toutput.stopReason = mapStopReason(candidate.finishReason);\n\t\t\t\t\tif (output.content.some((b) => b.type === \"toolCall\")) {\n\t\t\t\t\t\toutput.stopReason = \"toolUse\";\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tif (chunk.usageMetadata) {\n\t\t\t\t\toutput.usage = {\n\t\t\t\t\t\tinput: chunk.usageMetadata.promptTokenCount || 0,\n\t\t\t\t\t\toutput:\n\t\t\t\t\t\t\t(chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),\n\t\t\t\t\t\tcacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,\n\t\t\t\t\t\tcacheWrite: 0,\n\t\t\t\t\t\ttotalTokens: chunk.usageMetadata.totalTokenCount || 0,\n\t\t\t\t\t\tcost: {\n\t\t\t\t\t\t\tinput: 0,\n\t\t\t\t\t\t\toutput: 0,\n\t\t\t\t\t\t\tcacheRead: 0,\n\t\t\t\t\t\t\tcacheWrite: 0,\n\t\t\t\t\t\t\ttotal: 0,\n\t\t\t\t\t\t},\n\t\t\t\t\t};\n\t\t\t\t\tcalculateCost(model, output.usage);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (currentBlock) {\n\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t} else {\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\t// Remove internal index property used during streaming\n\t\t\tfor (const block of output.content) {\n\t\t\t\tif (\"index\" in block) {\n\t\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t}\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nfunction createClient(model: Model<\"google-vertex\">, project: string, location: string): GoogleGenAI {\n\tconst httpOptions: { headers?: Record<string, string> } = {};\n\n\tif (model.headers) {\n\t\thttpOptions.headers = { ...model.headers };\n\t}\n\n\tconst hasHttpOptions = Object.values(httpOptions).some(Boolean);\n\n\treturn new GoogleGenAI({\n\t\tvertexai: true,\n\t\tproject,\n\t\tlocation,\n\t\tapiVersion: API_VERSION,\n\t\thttpOptions: hasHttpOptions ? httpOptions : undefined,\n\t});\n}\n\nfunction resolveProject(options?: GoogleVertexOptions): string {\n\tconst project = options?.project || process.env.GOOGLE_CLOUD_PROJECT || process.env.GCLOUD_PROJECT;\n\tif (!project) {\n\t\tthrow new Error(\n\t\t\t\"Vertex AI requires a project ID. Set GOOGLE_CLOUD_PROJECT/GCLOUD_PROJECT or pass project in options.\",\n\t\t);\n\t}\n\treturn project;\n}\n\nfunction resolveLocation(options?: GoogleVertexOptions): string {\n\tconst location = options?.location || process.env.GOOGLE_CLOUD_LOCATION;\n\tif (!location) {\n\t\tthrow new Error(\"Vertex AI requires a location. Set GOOGLE_CLOUD_LOCATION or pass location in options.\");\n\t}\n\treturn location;\n}\n\nfunction buildParams(\n\tmodel: Model<\"google-vertex\">,\n\tcontext: Context,\n\toptions: GoogleVertexOptions = {},\n): GenerateContentParameters {\n\tconst contents = convertMessages(model, context);\n\n\tconst generationConfig: GenerateContentConfig = {};\n\tif (options.temperature !== undefined) {\n\t\tgenerationConfig.temperature = options.temperature;\n\t}\n\tif (options.maxTokens !== undefined) {\n\t\tgenerationConfig.maxOutputTokens = options.maxTokens;\n\t}\n\n\tconst config: GenerateContentConfig = {\n\t\t...(Object.keys(generationConfig).length > 0 && generationConfig),\n\t\t...(context.systemPrompt && { systemInstruction: sanitizeSurrogates(context.systemPrompt) }),\n\t\t...(context.tools && context.tools.length > 0 && { tools: convertTools(context.tools) }),\n\t};\n\n\tif (context.tools && context.tools.length > 0 && options.toolChoice) {\n\t\tconfig.toolConfig = {\n\t\t\tfunctionCallingConfig: {\n\t\t\t\tmode: mapToolChoice(options.toolChoice),\n\t\t\t},\n\t\t};\n\t} else {\n\t\tconfig.toolConfig = undefined;\n\t}\n\n\tif (options.thinking?.enabled && model.reasoning) {\n\t\tconst thinkingConfig: ThinkingConfig = { includeThoughts: true };\n\t\tif (options.thinking.level !== undefined) {\n\t\t\tthinkingConfig.thinkingLevel = THINKING_LEVEL_MAP[options.thinking.level];\n\t\t} else if (options.thinking.budgetTokens !== undefined) {\n\t\t\tthinkingConfig.thinkingBudget = options.thinking.budgetTokens;\n\t\t}\n\t\tconfig.thinkingConfig = thinkingConfig;\n\t}\n\n\tif (options.signal) {\n\t\tif (options.signal.aborted) {\n\t\t\tthrow new Error(\"Request aborted\");\n\t\t}\n\t\tconfig.abortSignal = options.signal;\n\t}\n\n\tconst params: GenerateContentParameters = {\n\t\tmodel: model.id,\n\t\tcontents,\n\t\tconfig,\n\t};\n\n\treturn params;\n}\n"]}

package/dist/providers/google-vertex.js

@@ -0,0 +1,289 @@
+import { GoogleGenAI, ThinkingLevel, } from "@google/genai";
+import { calculateCost } from "../models.js";
+import { AssistantMessageEventStream } from "../utils/event-stream.js";
+import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
+import { convertMessages, convertTools, mapStopReason, mapToolChoice } from "./google-shared.js";
+const API_VERSION = "v1";
+const THINKING_LEVEL_MAP = {
+    THINKING_LEVEL_UNSPECIFIED: ThinkingLevel.THINKING_LEVEL_UNSPECIFIED,
+    MINIMAL: ThinkingLevel.MINIMAL,
+    LOW: ThinkingLevel.LOW,
+    MEDIUM: ThinkingLevel.MEDIUM,
+    HIGH: ThinkingLevel.HIGH,
+};
+// Counter for generating unique tool call IDs
+let toolCallCounter = 0;
+export const streamGoogleVertex = (model, context, options) => {
+    const stream = new AssistantMessageEventStream();
+    (async () => {
+        const output = {
+            role: "assistant",
+            content: [],
+            api: "google-vertex",
+            provider: model.provider,
+            model: model.id,
+            usage: {
+                input: 0,
+                output: 0,
+                cacheRead: 0,
+                cacheWrite: 0,
+                totalTokens: 0,
+                cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+            },
+            stopReason: "stop",
+            timestamp: Date.now(),
+        };
+        try {
+            const project = resolveProject(options);
+            const location = resolveLocation(options);
+            const client = createClient(model, project, location);
+            const params = buildParams(model, context, options);
+            const googleStream = await client.models.generateContentStream(params);
+            stream.push({ type: "start", partial: output });
+            let currentBlock = null;
+            const blocks = output.content;
+            const blockIndex = () => blocks.length - 1;
+            for await (const chunk of googleStream) {
+                const candidate = chunk.candidates?.[0];
+                if (candidate?.content?.parts) {
+                    for (const part of candidate.content.parts) {
+                        if (part.text !== undefined) {
+                            const isThinking = part.thought === true;
+                            if (!currentBlock ||
+                                (isThinking && currentBlock.type !== "thinking") ||
+                                (!isThinking && currentBlock.type !== "text")) {
+                                if (currentBlock) {
+                                    if (currentBlock.type === "text") {
+                                        stream.push({
+                                            type: "text_end",
+                                            contentIndex: blocks.length - 1,
+                                            content: currentBlock.text,
+                                            partial: output,
+                                        });
+                                    }
+                                    else {
+                                        stream.push({
+                                            type: "thinking_end",
+                                            contentIndex: blockIndex(),
+                                            content: currentBlock.thinking,
+                                            partial: output,
+                                        });
+                                    }
+                                }
+                                if (isThinking) {
+                                    currentBlock = { type: "thinking", thinking: "", thinkingSignature: undefined };
+                                    output.content.push(currentBlock);
+                                    stream.push({ type: "thinking_start", contentIndex: blockIndex(), partial: output });
+                                }
+                                else {
+                                    currentBlock = { type: "text", text: "" };
+                                    output.content.push(currentBlock);
+                                    stream.push({ type: "text_start", contentIndex: blockIndex(), partial: output });
+                                }
+                            }
+                            if (currentBlock.type === "thinking") {
+                                currentBlock.thinking += part.text;
+                                currentBlock.thinkingSignature = part.thoughtSignature;
+                                stream.push({
+                                    type: "thinking_delta",
+                                    contentIndex: blockIndex(),
+                                    delta: part.text,
+                                    partial: output,
+                                });
+                            }
+                            else {
+                                currentBlock.text += part.text;
+                                stream.push({
+                                    type: "text_delta",
+                                    contentIndex: blockIndex(),
+                                    delta: part.text,
+                                    partial: output,
+                                });
+                            }
+                        }
+                        if (part.functionCall) {
+                            if (currentBlock) {
+                                if (currentBlock.type === "text") {
+                                    stream.push({
+                                        type: "text_end",
+                                        contentIndex: blockIndex(),
+                                        content: currentBlock.text,
+                                        partial: output,
+                                    });
+                                }
+                                else {
+                                    stream.push({
+                                        type: "thinking_end",
+                                        contentIndex: blockIndex(),
+                                        content: currentBlock.thinking,
+                                        partial: output,
+                                    });
+                                }
+                                currentBlock = null;
+                            }
+                            const providedId = part.functionCall.id;
+                            const needsNewId = !providedId || output.content.some((b) => b.type === "toolCall" && b.id === providedId);
+                            const toolCallId = needsNewId
+                                ? `${part.functionCall.name}_${Date.now()}_${++toolCallCounter}`
+                                : providedId;
+                            const toolCall = {
+                                type: "toolCall",
+                                id: toolCallId,
+                                name: part.functionCall.name || "",
+                                arguments: part.functionCall.args,
+                                ...(part.thoughtSignature && { thoughtSignature: part.thoughtSignature }),
+                            };
+                            output.content.push(toolCall);
+                            stream.push({ type: "toolcall_start", contentIndex: blockIndex(), partial: output });
+                            stream.push({
+                                type: "toolcall_delta",
+                                contentIndex: blockIndex(),
+                                delta: JSON.stringify(toolCall.arguments),
+                                partial: output,
+                            });
+                            stream.push({ type: "toolcall_end", contentIndex: blockIndex(), toolCall, partial: output });
+                        }
+                    }
+                }
+                if (candidate?.finishReason) {
+                    output.stopReason = mapStopReason(candidate.finishReason);
+                    if (output.content.some((b) => b.type === "toolCall")) {
+                        output.stopReason = "toolUse";
+                    }
+                }
+                if (chunk.usageMetadata) {
+                    output.usage = {
+                        input: chunk.usageMetadata.promptTokenCount || 0,
+                        output: (chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),
+                        cacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,
+                        cacheWrite: 0,
+                        totalTokens: chunk.usageMetadata.totalTokenCount || 0,
+                        cost: {
+                            input: 0,
+                            output: 0,
+                            cacheRead: 0,
+                            cacheWrite: 0,
+                            total: 0,
+                        },
+                    };
+                    calculateCost(model, output.usage);
+                }
+            }
+            if (currentBlock) {
+                if (currentBlock.type === "text") {
+                    stream.push({
+                        type: "text_end",
+                        contentIndex: blockIndex(),
+                        content: currentBlock.text,
+                        partial: output,
+                    });
+                }
+                else {
+                    stream.push({
+                        type: "thinking_end",
+                        contentIndex: blockIndex(),
+                        content: currentBlock.thinking,
+                        partial: output,
+                    });
+                }
+            }
+            if (options?.signal?.aborted) {
+                throw new Error("Request was aborted");
+            }
+            if (output.stopReason === "aborted" || output.stopReason === "error") {
+                throw new Error("An unknown error occurred");
+            }
+            stream.push({ type: "done", reason: output.stopReason, message: output });
+            stream.end();
+        }
+        catch (error) {
+            // Remove internal index property used during streaming
+            for (const block of output.content) {
+                if ("index" in block) {
+                    delete block.index;
+                }
+            }
+            output.stopReason = options?.signal?.aborted ? "aborted" : "error";
+            output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
+            stream.push({ type: "error", reason: output.stopReason, error: output });
+            stream.end();
+        }
+    })();
+    return stream;
+};
+function createClient(model, project, location) {
+    const httpOptions = {};
+    if (model.headers) {
+        httpOptions.headers = { ...model.headers };
+    }
+    const hasHttpOptions = Object.values(httpOptions).some(Boolean);
+    return new GoogleGenAI({
+        vertexai: true,
+        project,
+        location,
+        apiVersion: API_VERSION,
+        httpOptions: hasHttpOptions ? httpOptions : undefined,
+    });
+}
+function resolveProject(options) {
+    const project = options?.project || process.env.GOOGLE_CLOUD_PROJECT || process.env.GCLOUD_PROJECT;
+    if (!project) {
+        throw new Error("Vertex AI requires a project ID. Set GOOGLE_CLOUD_PROJECT/GCLOUD_PROJECT or pass project in options.");
+    }
+    return project;
+}
+function resolveLocation(options) {
+    const location = options?.location || process.env.GOOGLE_CLOUD_LOCATION;
+    if (!location) {
+        throw new Error("Vertex AI requires a location. Set GOOGLE_CLOUD_LOCATION or pass location in options.");
+    }
+    return location;
+}
+function buildParams(model, context, options = {}) {
+    const contents = convertMessages(model, context);
+    const generationConfig = {};
+    if (options.temperature !== undefined) {
+        generationConfig.temperature = options.temperature;
+    }
+    if (options.maxTokens !== undefined) {
+        generationConfig.maxOutputTokens = options.maxTokens;
+    }
+    const config = {
+        ...(Object.keys(generationConfig).length > 0 && generationConfig),
+        ...(context.systemPrompt && { systemInstruction: sanitizeSurrogates(context.systemPrompt) }),
+        ...(context.tools && context.tools.length > 0 && { tools: convertTools(context.tools) }),
+    };
+    if (context.tools && context.tools.length > 0 && options.toolChoice) {
+        config.toolConfig = {
+            functionCallingConfig: {
+                mode: mapToolChoice(options.toolChoice),
+            },
+        };
+    }
+    else {
+        config.toolConfig = undefined;
+    }
+    if (options.thinking?.enabled && model.reasoning) {
+        const thinkingConfig = { includeThoughts: true };
+        if (options.thinking.level !== undefined) {
+            thinkingConfig.thinkingLevel = THINKING_LEVEL_MAP[options.thinking.level];
+        }
+        else if (options.thinking.budgetTokens !== undefined) {
+            thinkingConfig.thinkingBudget = options.thinking.budgetTokens;
+        }
+        config.thinkingConfig = thinkingConfig;
+    }
+    if (options.signal) {
+        if (options.signal.aborted) {
+            throw new Error("Request aborted");
+        }
+        config.abortSignal = options.signal;
+    }
+    const params = {
+        model: model.id,
+        contents,
+        config,
+    };
+    return params;
+}
+//# sourceMappingURL=google-vertex.js.map

package/dist/providers/google-vertex.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"google-vertex.js","sourceRoot":"","sources":["../../src/providers/google-vertex.ts"],"names":[],"mappings":"AAAA,OAAO,EAGN,WAAW,EAEX,aAAa,GACb,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAY7C,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AAElE,OAAO,EAAE,eAAe,EAAE,YAAY,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAajG,MAAM,WAAW,GAAG,IAAI,CAAC;AAEzB,MAAM,kBAAkB,GAA+C;IACtE,0BAA0B,EAAE,aAAa,CAAC,0BAA0B;IACpE,OAAO,EAAE,aAAa,CAAC,OAAO;IAC9B,GAAG,EAAE,aAAa,CAAC,GAAG;IACtB,MAAM,EAAE,aAAa,CAAC,MAAM;IAC5B,IAAI,EAAE,aAAa,CAAC,IAAI;CACxB,CAAC;AAEF,8CAA8C;AAC9C,IAAI,eAAe,GAAG,CAAC,CAAC;AAExB,MAAM,CAAC,MAAM,kBAAkB,GAAoC,CAClE,KAA6B,EAC7B,OAAgB,EAChB,OAA6B,EACC,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IAEjD,CAAC,KAAK,IAAI,EAAE,CAAC;QACZ,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,eAAsB;YAC3B,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,MAAM;YAClB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,MAAM,OAAO,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;YACxC,MAAM,QAAQ,GAAG,eAAe,CAAC,OAAO,CAAC,CAAC;YAC1C,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC;YACtD,MAAM,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YACpD,MAAM,YAAY,GAAG,MAAM,MAAM,CAAC,MAAM,CAAC,qBAAqB,CAAC,MAAM,CAAC,CAAC;YAEvE,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAChD,IAAI,YAAY,GAAyC,IAAI,CAAC;YAC9D,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC;YAC9B,MAAM,UAAU,GAAG,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC;YAC3C,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,YAAY,EAAE,CAAC;gBACxC,MAAM,SAAS,GAAG,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,CAAC;gBACxC,IAAI,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;oBAC/B,KAAK,MAAM,IAAI,IAAI,SAAS,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;wBAC5C,IAAI,IAAI,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;4BAC7B,MAAM,UAAU,GAAG,IAAI,CAAC,OAAO,KAAK,IAAI,CAAC;4BACzC,IACC,CAAC,YAAY;gCACb,CAAC,UAAU,IAAI,YAAY,CAAC,IAAI,KAAK,UAAU,CAAC;gCAChD,CAAC,CAAC,UAAU,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,CAAC,EAC5C,CAAC;gCACF,IAAI,YAAY,EAAE,CAAC;oCAClB,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;wCAClC,MAAM,CAAC,IAAI,CAAC;4CACX,IAAI,EAAE,UAAU;4CAChB,YAAY,EAAE,MAAM,CAAC,MAAM,GAAG,CAAC;4CAC/B,OAAO,EAAE,YAAY,CAAC,IAAI;4CAC1B,OAAO,EAAE,MAAM;yCACf,CAAC,CAAC;oCACJ,CAAC;yCAAM,CAAC;wCACP,MAAM,CAAC,IAAI,CAAC;4CACX,IAAI,EAAE,cAAc;4CACpB,YAAY,EAAE,UAAU,EAAE;4CAC1B,OAAO,EAAE,YAAY,CAAC,QAAQ;4CAC9B,OAAO,EAAE,MAAM;yCACf,CAAC,CAAC;oCACJ,CAAC;gCACF,CAAC;gCACD,IAAI,UAAU,EAAE,CAAC;oCAChB,YAAY,GAAG,EAAE,IAAI,EAAE,UAAU,EAAE,QAAQ,EAAE,EAAE,EAAE,iBAAiB,EAAE,SAAS,EAAE,CAAC;oCAChF,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;oCAClC,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;gCACtF,CAAC;qCAAM,CAAC;oCACP,YAAY,GAAG,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC;oCAC1C,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;oCAClC,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;gCAClF,CAAC;4BACF,CAAC;4BACD,IAAI,YAAY,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;gCACtC,YAAY,CAAC,QAAQ,IAAI,IAAI,CAAC,IAAI,CAAC;gCACnC,YAAY,CAAC,iBAAiB,GAAG,IAAI,CAAC,gBAAgB,CAAC;gCACvD,MAAM,CAAC,IAAI,CAAC;oCACX,IAAI,EAAE,gBAAgB;oCACtB,YAAY,EAAE,UAAU,EAAE;oCAC1B,KAAK,EAAE,IAAI,CAAC,IAAI;oCAChB,OAAO,EAAE,MAAM;iCACf,CAAC,CAAC;4BACJ,CAAC;iCAAM,CAAC;gCACP,YAAY,CAAC,IAAI,IAAI,IAAI,CAAC,IAAI,CAAC;gCAC/B,MAAM,CAAC,IAAI,CAAC;oCACX,IAAI,EAAE,YAAY;oCAClB,YAAY,EAAE,UAAU,EAAE;oCAC1B,KAAK,EAAE,IAAI,CAAC,IAAI;oCAChB,OAAO,EAAE,MAAM;iCACf,CAAC,CAAC;4BACJ,CAAC;wBACF,CAAC;wBAED,IAAI,IAAI,CAAC,YAAY,EAAE,CAAC;4BACvB,IAAI,YAAY,EAAE,CAAC;gCAClB,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;oCAClC,MAAM,CAAC,IAAI,CAAC;wCACX,IAAI,EAAE,UAAU;wCAChB,YAAY,EAAE,UAAU,EAAE;wCAC1B,OAAO,EAAE,YAAY,CAAC,IAAI;wCAC1B,OAAO,EAAE,MAAM;qCACf,CAAC,CAAC;gCACJ,CAAC;qCAAM,CAAC;oCACP,MAAM,CAAC,IAAI,CAAC;wCACX,IAAI,EAAE,cAAc;wCACpB,YAAY,EAAE,UAAU,EAAE;wCAC1B,OAAO,EAAE,YAAY,CAAC,QAAQ;wCAC9B,OAAO,EAAE,MAAM;qCACf,CAAC,CAAC;gCACJ,CAAC;gCACD,YAAY,GAAG,IAAI,CAAC;4BACrB,CAAC;4BAED,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,CAAC,EAAE,CAAC;4BACxC,MAAM,UAAU,GACf,CAAC,UAAU,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,IAAI,CAAC,CAAC,EAAE,KAAK,UAAU,CAAC,CAAC;4BACzF,MAAM,UAAU,GAAG,UAAU;gCAC5B,CAAC,CAAC,GAAG,IAAI,CAAC,YAAY,CAAC,IAAI,IAAI,IAAI,CAAC,GAAG,EAAE,IAAI,EAAE,eAAe,EAAE;gCAChE,CAAC,CAAC,UAAU,CAAC;4BAEd,MAAM,QAAQ,GAAa;gCAC1B,IAAI,EAAE,UAAU;gCAChB,EAAE,EAAE,UAAU;gCACd,IAAI,EAAE,IAAI,CAAC,YAAY,CAAC,IAAI,IAAI,EAAE;gCAClC,SAAS,EAAE,IAAI,CAAC,YAAY,CAAC,IAA2B;gCACxD,GAAG,CAAC,IAAI,CAAC,gBAAgB,IAAI,EAAE,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,EAAE,CAAC;6BACzE,CAAC;4BAEF,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;4BAC9B,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;4BACrF,MAAM,CAAC,IAAI,CAAC;gCACX,IAAI,EAAE,gBAAgB;gCACtB,YAAY,EAAE,UAAU,EAAE;gCAC1B,KAAK,EAAE,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC;gCACzC,OAAO,EAAE,MAAM;6BACf,CAAC,CAAC;4BACH,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,cAAc,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;wBAC9F,CAAC;oBACF,CAAC;gBACF,CAAC;gBAED,IAAI,SAAS,EAAE,YAAY,EAAE,CAAC;oBAC7B,MAAM,CAAC,UAAU,GAAG,aAAa,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC;oBAC1D,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,EAAE,CAAC;wBACvD,MAAM,CAAC,UAAU,GAAG,SAAS,CAAC;oBAC/B,CAAC;gBACF,CAAC;gBAED,IAAI,KAAK,CAAC,aAAa,EAAE,CAAC;oBACzB,MAAM,CAAC,KAAK,GAAG;wBACd,KAAK,EAAE,KAAK,CAAC,aAAa,CAAC,gBAAgB,IAAI,CAAC;wBAChD,MAAM,EACL,CAAC,KAAK,CAAC,aAAa,CAAC,oBAAoB,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,aAAa,CAAC,kBAAkB,IAAI,CAAC,CAAC;wBAChG,SAAS,EAAE,KAAK,CAAC,aAAa,CAAC,uBAAuB,IAAI,CAAC;wBAC3D,UAAU,EAAE,CAAC;wBACb,WAAW,EAAE,KAAK,CAAC,aAAa,CAAC,eAAe,IAAI,CAAC;wBACrD,IAAI,EAAE;4BACL,KAAK,EAAE,CAAC;4BACR,MAAM,EAAE,CAAC;4BACT,SAAS,EAAE,CAAC;4BACZ,UAAU,EAAE,CAAC;4BACb,KAAK,EAAE,CAAC;yBACR;qBACD,CAAC;oBACF,aAAa,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;gBACpC,CAAC;YACF,CAAC;YAED,IAAI,YAAY,EAAE,CAAC;gBAClB,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;oBAClC,MAAM,CAAC,IAAI,CAAC;wBACX,IAAI,EAAE,UAAU;wBAChB,YAAY,EAAE,UAAU,EAAE;wBAC1B,OAAO,EAAE,YAAY,CAAC,IAAI;wBAC1B,OAAO,EAAE,MAAM;qBACf,CAAC,CAAC;gBACJ,CAAC;qBAAM,CAAC;oBACP,MAAM,CAAC,IAAI,CAAC;wBACX,IAAI,EAAE,cAAc;wBACpB,YAAY,EAAE,UAAU,EAAE;wBAC1B,OAAO,EAAE,YAAY,CAAC,QAAQ;wBAC9B,OAAO,EAAE,MAAM;qBACf,CAAC,CAAC;gBACJ,CAAC;YACF,CAAC;YAED,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,CAAC;YAC9C,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,uDAAuD;YACvD,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;gBACpC,IAAI,OAAO,IAAI,KAAK,EAAE,CAAC;oBACtB,OAAQ,KAA4B,CAAC,KAAK,CAAC;gBAC5C,CAAC;YACF,CAAC;YACD,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,CAAC,YAAY,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YACrF,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IAAA,CACD,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AAAA,CACd,CAAC;AAEF,SAAS,YAAY,CAAC,KAA6B,EAAE,OAAe,EAAE,QAAgB,EAAe;IACpG,MAAM,WAAW,GAAyC,EAAE,CAAC;IAE7D,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,WAAW,CAAC,OAAO,GAAG,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IAC5C,CAAC;IAED,MAAM,cAAc,GAAG,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAEhE,OAAO,IAAI,WAAW,CAAC;QACtB,QAAQ,EAAE,IAAI;QACd,OAAO;QACP,QAAQ;QACR,UAAU,EAAE,WAAW;QACvB,WAAW,EAAE,cAAc,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS;KACrD,CAAC,CAAC;AAAA,CACH;AAED,SAAS,cAAc,CAAC,OAA6B,EAAU;IAC9D,MAAM,OAAO,GAAG,OAAO,EAAE,OAAO,IAAI,OAAO,CAAC,GAAG,CAAC,oBAAoB,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC;IACnG,IAAI,CAAC,OAAO,EAAE,CAAC;QACd,MAAM,IAAI,KAAK,CACd,sGAAsG,CACtG,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AAAA,CACf;AAED,SAAS,eAAe,CAAC,OAA6B,EAAU;IAC/D,MAAM,QAAQ,GAAG,OAAO,EAAE,QAAQ,IAAI,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC;IACxE,IAAI,CAAC,QAAQ,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,uFAAuF,CAAC,CAAC;IAC1G,CAAC;IACD,OAAO,QAAQ,CAAC;AAAA,CAChB;AAED,SAAS,WAAW,CACnB,KAA6B,EAC7B,OAAgB,EAChB,OAAO,GAAwB,EAAE,EACL;IAC5B,MAAM,QAAQ,GAAG,eAAe,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IAEjD,MAAM,gBAAgB,GAA0B,EAAE,CAAC;IACnD,IAAI,OAAO,CAAC,WAAW,KAAK,SAAS,EAAE,CAAC;QACvC,gBAAgB,CAAC,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC;IACpD,CAAC;IACD,IAAI,OAAO,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;QACrC,gBAAgB,CAAC,eAAe,GAAG,OAAO,CAAC,SAAS,CAAC;IACtD,CAAC;IAED,MAAM,MAAM,GAA0B;QACrC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,MAAM,GAAG,CAAC,IAAI,gBAAgB,CAAC;QACjE,GAAG,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,iBAAiB,EAAE,kBAAkB,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,CAAC;QAC5F,GAAG,CAAC,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE,YAAY,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;KACxF,CAAC;IAEF,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;QACrE,MAAM,CAAC,UAAU,GAAG;YACnB,qBAAqB,EAAE;gBACtB,IAAI,EAAE,aAAa,CAAC,OAAO,CAAC,UAAU,CAAC;aACvC;SACD,CAAC;IACH,CAAC;SAAM,CAAC;QACP,MAAM,CAAC,UAAU,GAAG,SAAS,CAAC;IAC/B,CAAC;IAED,IAAI,OAAO,CAAC,QAAQ,EAAE,OAAO,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QAClD,MAAM,cAAc,GAAmB,EAAE,eAAe,EAAE,IAAI,EAAE,CAAC;QACjE,IAAI,OAAO,CAAC,QAAQ,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;YAC1C,cAAc,CAAC,aAAa,GAAG,kBAAkB,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QAC3E,CAAC;aAAM,IAAI,OAAO,CAAC,QAAQ,CAAC,YAAY,KAAK,SAAS,EAAE,CAAC;YACxD,cAAc,CAAC,cAAc,GAAG,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAC;QAC/D,CAAC;QACD,MAAM,CAAC,cAAc,GAAG,cAAc,CAAC;IACxC,CAAC;IAED,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;QACpB,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YAC5B,MAAM,IAAI,KAAK,CAAC,iBAAiB,CAAC,CAAC;QACpC,CAAC;QACD,MAAM,CAAC,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC;IACrC,CAAC;IAED,MAAM,MAAM,GAA8B;QACzC,KAAK,EAAE,KAAK,CAAC,EAAE;QACf,QAAQ;QACR,MAAM;KACN,CAAC;IAEF,OAAO,MAAM,CAAC;AAAA,CACd","sourcesContent":["import {\n\ttype GenerateContentConfig,\n\ttype GenerateContentParameters,\n\tGoogleGenAI,\n\ttype ThinkingConfig,\n\tThinkingLevel,\n} from \"@google/genai\";\nimport { calculateCost } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingContent,\n\tToolCall,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport type { GoogleThinkingLevel } from \"./google-gemini-cli.js\";\nimport { convertMessages, convertTools, mapStopReason, mapToolChoice } from \"./google-shared.js\";\n\nexport interface GoogleVertexOptions extends StreamOptions {\n\ttoolChoice?: \"auto\" | \"none\" | \"any\";\n\tthinking?: {\n\t\tenabled: boolean;\n\t\tbudgetTokens?: number; // -1 for dynamic, 0 to disable\n\t\tlevel?: GoogleThinkingLevel;\n\t};\n\tproject?: string;\n\tlocation?: string;\n}\n\nconst API_VERSION = \"v1\";\n\nconst THINKING_LEVEL_MAP: Record<GoogleThinkingLevel, ThinkingLevel> = {\n\tTHINKING_LEVEL_UNSPECIFIED: ThinkingLevel.THINKING_LEVEL_UNSPECIFIED,\n\tMINIMAL: ThinkingLevel.MINIMAL,\n\tLOW: ThinkingLevel.LOW,\n\tMEDIUM: ThinkingLevel.MEDIUM,\n\tHIGH: ThinkingLevel.HIGH,\n};\n\n// Counter for generating unique tool call IDs\nlet toolCallCounter = 0;\n\nexport const streamGoogleVertex: StreamFunction<\"google-vertex\"> = (\n\tmodel: Model<\"google-vertex\">,\n\tcontext: Context,\n\toptions?: GoogleVertexOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"google-vertex\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\tconst project = resolveProject(options);\n\t\t\tconst location = resolveLocation(options);\n\t\t\tconst client = createClient(model, project, location);\n\t\t\tconst params = buildParams(model, context, options);\n\t\t\tconst googleStream = await client.models.generateContentStream(params);\n\n\t\t\tstream.push({ type: \"start\", partial: output });\n\t\t\tlet currentBlock: TextContent | ThinkingContent | null = null;\n\t\t\tconst blocks = output.content;\n\t\t\tconst blockIndex = () => blocks.length - 1;\n\t\t\tfor await (const chunk of googleStream) {\n\t\t\t\tconst candidate = chunk.candidates?.[0];\n\t\t\t\tif (candidate?.content?.parts) {\n\t\t\t\t\tfor (const part of candidate.content.parts) {\n\t\t\t\t\t\tif (part.text !== undefined) {\n\t\t\t\t\t\t\tconst isThinking = part.thought === true;\n\t\t\t\t\t\t\tif (\n\t\t\t\t\t\t\t\t!currentBlock ||\n\t\t\t\t\t\t\t\t(isThinking && currentBlock.type !== \"thinking\") ||\n\t\t\t\t\t\t\t\t(!isThinking && currentBlock.type !== \"text\")\n\t\t\t\t\t\t\t) {\n\t\t\t\t\t\t\t\tif (currentBlock) {\n\t\t\t\t\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blocks.length - 1,\n\t\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tif (isThinking) {\n\t\t\t\t\t\t\t\t\tcurrentBlock = { type: \"thinking\", thinking: \"\", thinkingSignature: undefined };\n\t\t\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\t\t\tstream.push({ type: \"thinking_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\tcurrentBlock = { type: \"text\", text: \"\" };\n\t\t\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (currentBlock.type === \"thinking\") {\n\t\t\t\t\t\t\t\tcurrentBlock.thinking += part.text;\n\t\t\t\t\t\t\t\tcurrentBlock.thinkingSignature = part.thoughtSignature;\n\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\tdelta: part.text,\n\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\tcurrentBlock.text += part.text;\n\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\tdelta: part.text,\n\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tif (part.functionCall) {\n\t\t\t\t\t\t\tif (currentBlock) {\n\t\t\t\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tcurrentBlock = null;\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\tconst providedId = part.functionCall.id;\n\t\t\t\t\t\t\tconst needsNewId =\n\t\t\t\t\t\t\t\t!providedId || output.content.some((b) => b.type === \"toolCall\" && b.id === providedId);\n\t\t\t\t\t\t\tconst toolCallId = needsNewId\n\t\t\t\t\t\t\t\t? `${part.functionCall.name}_${Date.now()}_${++toolCallCounter}`\n\t\t\t\t\t\t\t\t: providedId;\n\n\t\t\t\t\t\t\tconst toolCall: ToolCall = {\n\t\t\t\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\t\t\t\tid: toolCallId,\n\t\t\t\t\t\t\t\tname: part.functionCall.name || \"\",\n\t\t\t\t\t\t\t\targuments: part.functionCall.args as Record<string, any>,\n\t\t\t\t\t\t\t\t...(part.thoughtSignature && { thoughtSignature: part.thoughtSignature }),\n\t\t\t\t\t\t\t};\n\n\t\t\t\t\t\t\toutput.content.push(toolCall);\n\t\t\t\t\t\t\tstream.push({ type: \"toolcall_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\tdelta: JSON.stringify(toolCall.arguments),\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\tstream.push({ type: \"toolcall_end\", contentIndex: blockIndex(), toolCall, partial: output });\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tif (candidate?.finishReason) {\n\t\t\t\t\toutput.stopReason = mapStopReason(candidate.finishReason);\n\t\t\t\t\tif (output.content.some((b) => b.type === \"toolCall\")) {\n\t\t\t\t\t\toutput.stopReason = \"toolUse\";\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tif (chunk.usageMetadata) {\n\t\t\t\t\toutput.usage = {\n\t\t\t\t\t\tinput: chunk.usageMetadata.promptTokenCount || 0,\n\t\t\t\t\t\toutput:\n\t\t\t\t\t\t\t(chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),\n\t\t\t\t\t\tcacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,\n\t\t\t\t\t\tcacheWrite: 0,\n\t\t\t\t\t\ttotalTokens: chunk.usageMetadata.totalTokenCount || 0,\n\t\t\t\t\t\tcost: {\n\t\t\t\t\t\t\tinput: 0,\n\t\t\t\t\t\t\toutput: 0,\n\t\t\t\t\t\t\tcacheRead: 0,\n\t\t\t\t\t\t\tcacheWrite: 0,\n\t\t\t\t\t\t\ttotal: 0,\n\t\t\t\t\t\t},\n\t\t\t\t\t};\n\t\t\t\t\tcalculateCost(model, output.usage);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (currentBlock) {\n\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t} else {\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\t// Remove internal index property used during streaming\n\t\t\tfor (const block of output.content) {\n\t\t\t\tif (\"index\" in block) {\n\t\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t}\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nfunction createClient(model: Model<\"google-vertex\">, project: string, location: string): GoogleGenAI {\n\tconst httpOptions: { headers?: Record<string, string> } = {};\n\n\tif (model.headers) {\n\t\thttpOptions.headers = { ...model.headers };\n\t}\n\n\tconst hasHttpOptions = Object.values(httpOptions).some(Boolean);\n\n\treturn new GoogleGenAI({\n\t\tvertexai: true,\n\t\tproject,\n\t\tlocation,\n\t\tapiVersion: API_VERSION,\n\t\thttpOptions: hasHttpOptions ? httpOptions : undefined,\n\t});\n}\n\nfunction resolveProject(options?: GoogleVertexOptions): string {\n\tconst project = options?.project || process.env.GOOGLE_CLOUD_PROJECT || process.env.GCLOUD_PROJECT;\n\tif (!project) {\n\t\tthrow new Error(\n\t\t\t\"Vertex AI requires a project ID. Set GOOGLE_CLOUD_PROJECT/GCLOUD_PROJECT or pass project in options.\",\n\t\t);\n\t}\n\treturn project;\n}\n\nfunction resolveLocation(options?: GoogleVertexOptions): string {\n\tconst location = options?.location || process.env.GOOGLE_CLOUD_LOCATION;\n\tif (!location) {\n\t\tthrow new Error(\"Vertex AI requires a location. Set GOOGLE_CLOUD_LOCATION or pass location in options.\");\n\t}\n\treturn location;\n}\n\nfunction buildParams(\n\tmodel: Model<\"google-vertex\">,\n\tcontext: Context,\n\toptions: GoogleVertexOptions = {},\n): GenerateContentParameters {\n\tconst contents = convertMessages(model, context);\n\n\tconst generationConfig: GenerateContentConfig = {};\n\tif (options.temperature !== undefined) {\n\t\tgenerationConfig.temperature = options.temperature;\n\t}\n\tif (options.maxTokens !== undefined) {\n\t\tgenerationConfig.maxOutputTokens = options.maxTokens;\n\t}\n\n\tconst config: GenerateContentConfig = {\n\t\t...(Object.keys(generationConfig).length > 0 && generationConfig),\n\t\t...(context.systemPrompt && { systemInstruction: sanitizeSurrogates(context.systemPrompt) }),\n\t\t...(context.tools && context.tools.length > 0 && { tools: convertTools(context.tools) }),\n\t};\n\n\tif (context.tools && context.tools.length > 0 && options.toolChoice) {\n\t\tconfig.toolConfig = {\n\t\t\tfunctionCallingConfig: {\n\t\t\t\tmode: mapToolChoice(options.toolChoice),\n\t\t\t},\n\t\t};\n\t} else {\n\t\tconfig.toolConfig = undefined;\n\t}\n\n\tif (options.thinking?.enabled && model.reasoning) {\n\t\tconst thinkingConfig: ThinkingConfig = { includeThoughts: true };\n\t\tif (options.thinking.level !== undefined) {\n\t\t\tthinkingConfig.thinkingLevel = THINKING_LEVEL_MAP[options.thinking.level];\n\t\t} else if (options.thinking.budgetTokens !== undefined) {\n\t\t\tthinkingConfig.thinkingBudget = options.thinking.budgetTokens;\n\t\t}\n\t\tconfig.thinkingConfig = thinkingConfig;\n\t}\n\n\tif (options.signal) {\n\t\tif (options.signal.aborted) {\n\t\t\tthrow new Error(\"Request aborted\");\n\t\t}\n\t\tconfig.abortSignal = options.signal;\n\t}\n\n\tconst params: GenerateContentParameters = {\n\t\tmodel: model.id,\n\t\tcontents,\n\t\tconfig,\n\t};\n\n\treturn params;\n}\n"]}

package/dist/stream.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"stream.d.ts","sourceRoot":"","sources":["../src/stream.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EACX,GAAG,EACH,gBAAgB,EAChB,2BAA2B,EAC3B,OAAO,EACP,aAAa,EACb,KAAK,EACL,aAAa,EAEb,mBAAmB,EACnB,MAAM,YAAY,CAAC;AAEpB;;;;GAIG;AACH,wBAAgB,YAAY,CAAC,QAAQ,EAAE,aAAa,GAAG,MAAM,GAAG,SAAS,CAAC;AAC1E,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAAC;AA2BnE,wBAAgB,MAAM,CAAC,IAAI,SAAS,GAAG,EACtC,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,aAAa,CAAC,IAAI,CAAC,GAC3B,2BAA2B,CAkC7B;AAED,wBAAsB,QAAQ,CAAC,IAAI,SAAS,GAAG,EAC9C,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,aAAa,CAAC,IAAI,CAAC,GAC3B,OAAO,CAAC,gBAAgB,CAAC,CAG3B;AAED,wBAAgB,YAAY,CAAC,IAAI,SAAS,GAAG,EAC5C,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,mBAAmB,GAC3B,2BAA2B,CAQ7B;AAED,wBAAsB,cAAc,CAAC,IAAI,SAAS,GAAG,EACpD,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,mBAAmB,GAC3B,OAAO,CAAC,gBAAgB,CAAC,CAG3B","sourcesContent":["import { supportsXhigh } from \"./models.js\";\nimport { type AnthropicOptions, streamAnthropic } from \"./providers/anthropic.js\";\nimport { type GoogleOptions, streamGoogle } from \"./providers/google.js\";\nimport {\n\ttype GoogleGeminiCliOptions,\n\ttype GoogleThinkingLevel,\n\tstreamGoogleGeminiCli,\n} from \"./providers/google-gemini-cli.js\";\nimport { type OpenAICompletionsOptions, streamOpenAICompletions } from \"./providers/openai-completions.js\";\nimport { type OpenAIResponsesOptions, streamOpenAIResponses } from \"./providers/openai-responses.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tAssistantMessageEventStream,\n\tContext,\n\tKnownProvider,\n\tModel,\n\tOptionsForApi,\n\tReasoningEffort,\n\tSimpleStreamOptions,\n} from \"./types.js\";\n\n/**\n * Get API key for provider from known environment variables, e.g. OPENAI_API_KEY.\n *\n * Will not return API keys for providers that require OAuth tokens.\n */\nexport function getEnvApiKey(provider: KnownProvider): string | undefined;\nexport function getEnvApiKey(provider: string): string | undefined;\nexport function getEnvApiKey(provider: any): string | undefined {\n\t// Fall back to environment variables\n\tif (provider === \"github-copilot\") {\n\t\treturn process.env.COPILOT_GITHUB_TOKEN || process.env.GH_TOKEN || process.env.GITHUB_TOKEN;\n\t}\n\n\t// ANTHROPIC_OAUTH_TOKEN takes precedence over ANTHROPIC_API_KEY\n\tif (provider === \"anthropic\") {\n\t\treturn process.env.ANTHROPIC_OAUTH_TOKEN || process.env.ANTHROPIC_API_KEY;\n\t}\n\n\tconst envMap: Record<string, string> = {\n\t\topenai: \"OPENAI_API_KEY\",\n\t\tgoogle: \"GEMINI_API_KEY\",\n\t\tgroq: \"GROQ_API_KEY\",\n\t\tcerebras: \"CEREBRAS_API_KEY\",\n\t\txai: \"XAI_API_KEY\",\n\t\topenrouter: \"OPENROUTER_API_KEY\",\n\t\tzai: \"ZAI_API_KEY\",\n\t\tmistral: \"MISTRAL_API_KEY\",\n\t};\n\n\tconst envVar = envMap[provider];\n\treturn envVar ? process.env[envVar] : undefined;\n}\n\nexport function stream<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: OptionsForApi<TApi>,\n): AssistantMessageEventStream {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\tconst providerOptions = { ...options, apiKey };\n\n\tconst api: Api = model.api;\n\tswitch (api) {\n\t\tcase \"anthropic-messages\":\n\t\t\treturn streamAnthropic(model as Model<\"anthropic-messages\">, context, providerOptions);\n\n\t\tcase \"openai-completions\":\n\t\t\treturn streamOpenAICompletions(model as Model<\"openai-completions\">, context, providerOptions as any);\n\n\t\tcase \"openai-responses\":\n\t\t\treturn streamOpenAIResponses(model as Model<\"openai-responses\">, context, providerOptions as any);\n\n\t\tcase \"google-generative-ai\":\n\t\t\treturn streamGoogle(model as Model<\"google-generative-ai\">, context, providerOptions);\n\n\t\tcase \"google-gemini-cli\":\n\t\t\treturn streamGoogleGeminiCli(\n\t\t\t\tmodel as Model<\"google-gemini-cli\">,\n\t\t\t\tcontext,\n\t\t\t\tproviderOptions as GoogleGeminiCliOptions,\n\t\t\t);\n\n\t\tdefault: {\n\t\t\t// This should never be reached if all Api cases are handled\n\t\t\tconst _exhaustive: never = api;\n\t\t\tthrow new Error(`Unhandled API: ${_exhaustive}`);\n\t\t}\n\t}\n}\n\nexport async function complete<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: OptionsForApi<TApi>,\n): Promise<AssistantMessage> {\n\tconst s = stream(model, context, options);\n\treturn s.result();\n}\n\nexport function streamSimple<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst providerOptions = mapOptionsForApi(model, options, apiKey);\n\treturn stream(model, context, providerOptions);\n}\n\nexport async function completeSimple<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): Promise<AssistantMessage> {\n\tconst s = streamSimple(model, context, options);\n\treturn s.result();\n}\n\nfunction mapOptionsForApi<TApi extends Api>(\n\tmodel: Model<TApi>,\n\toptions?: SimpleStreamOptions,\n\tapiKey?: string,\n): OptionsForApi<TApi> {\n\tconst base = {\n\t\ttemperature: options?.temperature,\n\t\tmaxTokens: options?.maxTokens || Math.min(model.maxTokens, 32000),\n\t\tsignal: options?.signal,\n\t\tapiKey: apiKey || options?.apiKey,\n\t};\n\n\t// Helper to clamp xhigh to high for providers that don't support it\n\tconst clampReasoning = (effort: ReasoningEffort | undefined) => (effort === \"xhigh\" ? \"high\" : effort);\n\n\tswitch (model.api) {\n\t\tcase \"anthropic-messages\": {\n\t\t\t// Explicitly disable thinking when reasoning is not specified\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinkingEnabled: false } satisfies AnthropicOptions;\n\t\t\t}\n\n\t\t\tconst anthropicBudgets = {\n\t\t\t\tminimal: 1024,\n\t\t\t\tlow: 2048,\n\t\t\t\tmedium: 8192,\n\t\t\t\thigh: 16384,\n\t\t\t};\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tthinkingEnabled: true,\n\t\t\t\tthinkingBudgetTokens: anthropicBudgets[clampReasoning(options.reasoning)!],\n\t\t\t} satisfies AnthropicOptions;\n\t\t}\n\n\t\tcase \"openai-completions\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\treasoningEffort: supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning),\n\t\t\t} satisfies OpenAICompletionsOptions;\n\n\t\tcase \"openai-responses\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\treasoningEffort: supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning),\n\t\t\t} satisfies OpenAIResponsesOptions;\n\n\t\tcase \"google-generative-ai\": {\n\t\t\t// Explicitly disable thinking when reasoning is not specified\n\t\t\t// This is needed because Gemini has \"dynamic thinking\" enabled by default\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinking: { enabled: false } } satisfies GoogleOptions;\n\t\t\t}\n\n\t\t\tconst googleModel = model as Model<\"google-generative-ai\">;\n\t\t\tconst effort = clampReasoning(options.reasoning)!;\n\n\t\t\t// Gemini 3 models use thinkingLevel exclusively instead of thinkingBudget.\n\t\t\t// https://ai.google.dev/gemini-api/docs/thinking#set-budget\n\t\t\tif (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel)) {\n\t\t\t\treturn {\n\t\t\t\t\t...base,\n\t\t\t\t\tthinking: {\n\t\t\t\t\t\tenabled: true,\n\t\t\t\t\t\tlevel: getGemini3ThinkingLevel(effort, googleModel),\n\t\t\t\t\t},\n\t\t\t\t} satisfies GoogleOptions;\n\t\t\t}\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tthinking: {\n\t\t\t\t\tenabled: true,\n\t\t\t\t\tbudgetTokens: getGoogleBudget(googleModel, effort),\n\t\t\t\t},\n\t\t\t} satisfies GoogleOptions;\n\t\t}\n\n\t\tcase \"google-gemini-cli\": {\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinking: { enabled: false } } satisfies GoogleGeminiCliOptions;\n\t\t\t}\n\n\t\t\tconst effort = clampReasoning(options.reasoning)!;\n\n\t\t\t// Gemini 3 models use thinkingLevel instead of thinkingBudget\n\t\t\tif (model.id.includes(\"3-pro\") || model.id.includes(\"3-flash\")) {\n\t\t\t\treturn {\n\t\t\t\t\t...base,\n\t\t\t\t\tthinking: {\n\t\t\t\t\t\tenabled: true,\n\t\t\t\t\t\tlevel: getGeminiCliThinkingLevel(effort, model.id),\n\t\t\t\t\t},\n\t\t\t\t} satisfies GoogleGeminiCliOptions;\n\t\t\t}\n\n\t\t\t// Gemini 2.x models use thinkingBudget\n\t\t\tconst budgets: Record<ClampedReasoningEffort, number> = {\n\t\t\t\tminimal: 1024,\n\t\t\t\tlow: 2048,\n\t\t\t\tmedium: 8192,\n\t\t\t\thigh: 16384,\n\t\t\t};\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tthinking: {\n\t\t\t\t\tenabled: true,\n\t\t\t\t\tbudgetTokens: budgets[effort],\n\t\t\t\t},\n\t\t\t} satisfies GoogleGeminiCliOptions;\n\t\t}\n\n\t\tdefault: {\n\t\t\t// Exhaustiveness check\n\t\t\tconst _exhaustive: never = model.api;\n\t\t\tthrow new Error(`Unhandled API in mapOptionsForApi: ${_exhaustive}`);\n\t\t}\n\t}\n}\n\ntype ClampedReasoningEffort = Exclude<ReasoningEffort, \"xhigh\">;\n\nfunction isGemini3ProModel(model: Model<\"google-generative-ai\">): boolean {\n\t// Covers gemini-3-pro, gemini-3-pro-preview, and possible other prefixed ids in the future\n\treturn model.id.includes(\"3-pro\");\n}\n\nfunction isGemini3FlashModel(model: Model<\"google-generative-ai\">): boolean {\n\t// Covers gemini-3-flash, gemini-3-flash-preview, and possible other prefixed ids in the future\n\treturn model.id.includes(\"3-flash\");\n}\n\nfunction getGemini3ThinkingLevel(\n\teffort: ClampedReasoningEffort,\n\tmodel: Model<\"google-generative-ai\">,\n): GoogleThinkingLevel {\n\tif (isGemini3ProModel(model)) {\n\t\t// Gemini 3 Pro only supports LOW/HIGH (for now)\n\t\tswitch (effort) {\n\t\t\tcase \"minimal\":\n\t\t\tcase \"low\":\n\t\t\t\treturn \"LOW\";\n\t\t\tcase \"medium\":\n\t\t\tcase \"high\":\n\t\t\t\treturn \"HIGH\";\n\t\t}\n\t}\n\t// Gemini 3 Flash supports all four levels\n\tswitch (effort) {\n\t\tcase \"minimal\":\n\t\t\treturn \"MINIMAL\";\n\t\tcase \"low\":\n\t\t\treturn \"LOW\";\n\t\tcase \"medium\":\n\t\t\treturn \"MEDIUM\";\n\t\tcase \"high\":\n\t\t\treturn \"HIGH\";\n\t}\n}\n\nfunction getGeminiCliThinkingLevel(effort: ClampedReasoningEffort, modelId: string): GoogleThinkingLevel {\n\tif (modelId.includes(\"3-pro\")) {\n\t\t// Gemini 3 Pro only supports LOW/HIGH (for now)\n\t\tswitch (effort) {\n\t\t\tcase \"minimal\":\n\t\t\tcase \"low\":\n\t\t\t\treturn \"LOW\";\n\t\t\tcase \"medium\":\n\t\t\tcase \"high\":\n\t\t\t\treturn \"HIGH\";\n\t\t}\n\t}\n\t// Gemini 3 Flash supports all four levels\n\tswitch (effort) {\n\t\tcase \"minimal\":\n\t\t\treturn \"MINIMAL\";\n\t\tcase \"low\":\n\t\t\treturn \"LOW\";\n\t\tcase \"medium\":\n\t\t\treturn \"MEDIUM\";\n\t\tcase \"high\":\n\t\t\treturn \"HIGH\";\n\t}\n}\n\nfunction getGoogleBudget(model: Model<\"google-generative-ai\">, effort: ClampedReasoningEffort): number {\n\t// See https://ai.google.dev/gemini-api/docs/thinking#set-budget\n\tif (model.id.includes(\"2.5-pro\")) {\n\t\tconst budgets: Record<ClampedReasoningEffort, number> = {\n\t\t\tminimal: 128,\n\t\t\tlow: 2048,\n\t\t\tmedium: 8192,\n\t\t\thigh: 32768,\n\t\t};\n\t\treturn budgets[effort];\n\t}\n\n\tif (model.id.includes(\"2.5-flash\")) {\n\t\t// Covers 2.5-flash-lite as well\n\t\tconst budgets: Record<ClampedReasoningEffort, number> = {\n\t\t\tminimal: 128,\n\t\t\tlow: 2048,\n\t\t\tmedium: 8192,\n\t\t\thigh: 24576,\n\t\t};\n\t\treturn budgets[effort];\n\t}\n\n\t// Unknown model - use dynamic\n\treturn -1;\n}\n"]}
+{"version":3,"file":"stream.d.ts","sourceRoot":"","sources":["../src/stream.ts"],"names":[],"mappings":"AAWA,OAAO,KAAK,EACX,GAAG,EACH,gBAAgB,EAChB,2BAA2B,EAC3B,OAAO,EACP,aAAa,EACb,KAAK,EACL,aAAa,EAEb,mBAAmB,EACnB,MAAM,YAAY,CAAC;AAEpB;;;;GAIG;AACH,wBAAgB,YAAY,CAAC,QAAQ,EAAE,aAAa,GAAG,MAAM,GAAG,SAAS,CAAC;AAC1E,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAAC;AAmCnE,wBAAgB,MAAM,CAAC,IAAI,SAAS,GAAG,EACtC,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,aAAa,CAAC,IAAI,CAAC,GAC3B,2BAA2B,CAqC7B;AAED,wBAAsB,QAAQ,CAAC,IAAI,SAAS,GAAG,EAC9C,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,aAAa,CAAC,IAAI,CAAC,GAC3B,OAAO,CAAC,gBAAgB,CAAC,CAG3B;AAED,wBAAgB,YAAY,CAAC,IAAI,SAAS,GAAG,EAC5C,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,mBAAmB,GAC3B,2BAA2B,CAQ7B;AAED,wBAAsB,cAAc,CAAC,IAAI,SAAS,GAAG,EACpD,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,mBAAmB,GAC3B,OAAO,CAAC,gBAAgB,CAAC,CAG3B","sourcesContent":["import { supportsXhigh } from \"./models.js\";\nimport { type AnthropicOptions, streamAnthropic } from \"./providers/anthropic.js\";\nimport { type GoogleOptions, streamGoogle } from \"./providers/google.js\";\nimport {\n\ttype GoogleGeminiCliOptions,\n\ttype GoogleThinkingLevel,\n\tstreamGoogleGeminiCli,\n} from \"./providers/google-gemini-cli.js\";\nimport { type GoogleVertexOptions, streamGoogleVertex } from \"./providers/google-vertex.js\";\nimport { type OpenAICompletionsOptions, streamOpenAICompletions } from \"./providers/openai-completions.js\";\nimport { type OpenAIResponsesOptions, streamOpenAIResponses } from \"./providers/openai-responses.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tAssistantMessageEventStream,\n\tContext,\n\tKnownProvider,\n\tModel,\n\tOptionsForApi,\n\tReasoningEffort,\n\tSimpleStreamOptions,\n} from \"./types.js\";\n\n/**\n * Get API key for provider from known environment variables, e.g. OPENAI_API_KEY.\n *\n * Will not return API keys for providers that require OAuth tokens.\n */\nexport function getEnvApiKey(provider: KnownProvider): string | undefined;\nexport function getEnvApiKey(provider: string): string | undefined;\nexport function getEnvApiKey(provider: any): string | undefined {\n\t// Fall back to environment variables\n\tif (provider === \"github-copilot\") {\n\t\treturn process.env.COPILOT_GITHUB_TOKEN || process.env.GH_TOKEN || process.env.GITHUB_TOKEN;\n\t}\n\n\t// ANTHROPIC_OAUTH_TOKEN takes precedence over ANTHROPIC_API_KEY\n\tif (provider === \"anthropic\") {\n\t\treturn process.env.ANTHROPIC_OAUTH_TOKEN || process.env.ANTHROPIC_API_KEY;\n\t}\n\n\t// Vertex AI doesn't use API keys.\n\t// It relies on Google Cloud auth: `gcloud auth application-default login`.\n\t// @google/genai library picks up and manages the auth automatically.\n\t// Return a dummy value to maintain consistency.\n\tif (provider === \"google-vertex\") {\n\t\treturn \"vertex-ai-authenticated\";\n\t}\n\n\tconst envMap: Record<string, string> = {\n\t\topenai: \"OPENAI_API_KEY\",\n\t\tgoogle: \"GEMINI_API_KEY\",\n\t\tgroq: \"GROQ_API_KEY\",\n\t\tcerebras: \"CEREBRAS_API_KEY\",\n\t\txai: \"XAI_API_KEY\",\n\t\topenrouter: \"OPENROUTER_API_KEY\",\n\t\tzai: \"ZAI_API_KEY\",\n\t\tmistral: \"MISTRAL_API_KEY\",\n\t};\n\n\tconst envVar = envMap[provider];\n\treturn envVar ? process.env[envVar] : undefined;\n}\n\nexport function stream<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: OptionsForApi<TApi>,\n): AssistantMessageEventStream {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\tconst providerOptions = { ...options, apiKey };\n\n\tconst api: Api = model.api;\n\tswitch (api) {\n\t\tcase \"anthropic-messages\":\n\t\t\treturn streamAnthropic(model as Model<\"anthropic-messages\">, context, providerOptions);\n\n\t\tcase \"openai-completions\":\n\t\t\treturn streamOpenAICompletions(model as Model<\"openai-completions\">, context, providerOptions as any);\n\n\t\tcase \"openai-responses\":\n\t\t\treturn streamOpenAIResponses(model as Model<\"openai-responses\">, context, providerOptions as any);\n\n\t\tcase \"google-generative-ai\":\n\t\t\treturn streamGoogle(model as Model<\"google-generative-ai\">, context, providerOptions);\n\n\t\tcase \"google-gemini-cli\":\n\t\t\treturn streamGoogleGeminiCli(\n\t\t\t\tmodel as Model<\"google-gemini-cli\">,\n\t\t\t\tcontext,\n\t\t\t\tproviderOptions as GoogleGeminiCliOptions,\n\t\t\t);\n\n\t\tcase \"google-vertex\":\n\t\t\treturn streamGoogleVertex(model as Model<\"google-vertex\">, context, providerOptions as GoogleVertexOptions);\n\n\t\tdefault: {\n\t\t\t// This should never be reached if all Api cases are handled\n\t\t\tconst _exhaustive: never = api;\n\t\t\tthrow new Error(`Unhandled API: ${_exhaustive}`);\n\t\t}\n\t}\n}\n\nexport async function complete<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: OptionsForApi<TApi>,\n): Promise<AssistantMessage> {\n\tconst s = stream(model, context, options);\n\treturn s.result();\n}\n\nexport function streamSimple<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst providerOptions = mapOptionsForApi(model, options, apiKey);\n\treturn stream(model, context, providerOptions);\n}\n\nexport async function completeSimple<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): Promise<AssistantMessage> {\n\tconst s = streamSimple(model, context, options);\n\treturn s.result();\n}\n\nfunction mapOptionsForApi<TApi extends Api>(\n\tmodel: Model<TApi>,\n\toptions?: SimpleStreamOptions,\n\tapiKey?: string,\n): OptionsForApi<TApi> {\n\tconst base = {\n\t\ttemperature: options?.temperature,\n\t\tmaxTokens: options?.maxTokens || Math.min(model.maxTokens, 32000),\n\t\tsignal: options?.signal,\n\t\tapiKey: apiKey || options?.apiKey,\n\t};\n\n\t// Helper to clamp xhigh to high for providers that don't support it\n\tconst clampReasoning = (effort: ReasoningEffort | undefined) => (effort === \"xhigh\" ? \"high\" : effort);\n\n\tswitch (model.api) {\n\t\tcase \"anthropic-messages\": {\n\t\t\t// Explicitly disable thinking when reasoning is not specified\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinkingEnabled: false } satisfies AnthropicOptions;\n\t\t\t}\n\n\t\t\t// Claude requires max_tokens > thinking.budget_tokens\n\t\t\t// So we need to ensure maxTokens accounts for both thinking and output\n\t\t\tconst anthropicBudgets = {\n\t\t\t\tminimal: 1024,\n\t\t\t\tlow: 2048,\n\t\t\t\tmedium: 8192,\n\t\t\t\thigh: 16384,\n\t\t\t};\n\n\t\t\tconst minOutputTokens = 1024;\n\t\t\tlet thinkingBudget = anthropicBudgets[clampReasoning(options.reasoning)!];\n\t\t\t// Caller's maxTokens is the desired output; add thinking budget on top, capped at model limit\n\t\t\tconst maxTokens = Math.min((base.maxTokens || 0) + thinkingBudget, model.maxTokens);\n\n\t\t\t// If not enough room for thinking + output, reduce thinking budget\n\t\t\tif (maxTokens <= thinkingBudget) {\n\t\t\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t\t\t}\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tmaxTokens,\n\t\t\t\tthinkingEnabled: true,\n\t\t\t\tthinkingBudgetTokens: thinkingBudget,\n\t\t\t} satisfies AnthropicOptions;\n\t\t}\n\n\t\tcase \"openai-completions\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\treasoningEffort: supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning),\n\t\t\t} satisfies OpenAICompletionsOptions;\n\n\t\tcase \"openai-responses\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\treasoningEffort: supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning),\n\t\t\t} satisfies OpenAIResponsesOptions;\n\n\t\tcase \"google-generative-ai\": {\n\t\t\t// Explicitly disable thinking when reasoning is not specified\n\t\t\t// This is needed because Gemini has \"dynamic thinking\" enabled by default\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinking: { enabled: false } } satisfies GoogleOptions;\n\t\t\t}\n\n\t\t\tconst googleModel = model as Model<\"google-generative-ai\">;\n\t\t\tconst effort = clampReasoning(options.reasoning)!;\n\n\t\t\t// Gemini 3 models use thinkingLevel exclusively instead of thinkingBudget.\n\t\t\t// https://ai.google.dev/gemini-api/docs/thinking#set-budget\n\t\t\tif (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel)) {\n\t\t\t\treturn {\n\t\t\t\t\t...base,\n\t\t\t\t\tthinking: {\n\t\t\t\t\t\tenabled: true,\n\t\t\t\t\t\tlevel: getGemini3ThinkingLevel(effort, googleModel),\n\t\t\t\t\t},\n\t\t\t\t} satisfies GoogleOptions;\n\t\t\t}\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tthinking: {\n\t\t\t\t\tenabled: true,\n\t\t\t\t\tbudgetTokens: getGoogleBudget(googleModel, effort),\n\t\t\t\t},\n\t\t\t} satisfies GoogleOptions;\n\t\t}\n\n\t\tcase \"google-gemini-cli\": {\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinking: { enabled: false } } satisfies GoogleGeminiCliOptions;\n\t\t\t}\n\n\t\t\tconst effort = clampReasoning(options.reasoning)!;\n\n\t\t\t// Gemini 3 models use thinkingLevel instead of thinkingBudget\n\t\t\tif (model.id.includes(\"3-pro\") || model.id.includes(\"3-flash\")) {\n\t\t\t\treturn {\n\t\t\t\t\t...base,\n\t\t\t\t\tthinking: {\n\t\t\t\t\t\tenabled: true,\n\t\t\t\t\t\tlevel: getGeminiCliThinkingLevel(effort, model.id),\n\t\t\t\t\t},\n\t\t\t\t} satisfies GoogleGeminiCliOptions;\n\t\t\t}\n\n\t\t\t// Models using thinkingBudget (Gemini 2.x, Claude via Antigravity)\n\t\t\t// Claude requires max_tokens > thinking.budget_tokens\n\t\t\t// So we need to ensure maxTokens accounts for both thinking and output\n\t\t\tconst budgets: Record<ClampedReasoningEffort, number> = {\n\t\t\t\tminimal: 1024,\n\t\t\t\tlow: 2048,\n\t\t\t\tmedium: 8192,\n\t\t\t\thigh: 16384,\n\t\t\t};\n\n\t\t\tconst minOutputTokens = 1024;\n\t\t\tlet thinkingBudget = budgets[effort];\n\t\t\t// Caller's maxTokens is the desired output; add thinking budget on top, capped at model limit\n\t\t\tconst maxTokens = Math.min((base.maxTokens || 0) + thinkingBudget, model.maxTokens);\n\n\t\t\t// If not enough room for thinking + output, reduce thinking budget\n\t\t\tif (maxTokens <= thinkingBudget) {\n\t\t\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t\t\t}\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tmaxTokens,\n\t\t\t\tthinking: {\n\t\t\t\t\tenabled: true,\n\t\t\t\t\tbudgetTokens: thinkingBudget,\n\t\t\t\t},\n\t\t\t} satisfies GoogleGeminiCliOptions;\n\t\t}\n\n\t\tcase \"google-vertex\": {\n\t\t\t// Explicitly disable thinking when reasoning is not specified\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinking: { enabled: false } } satisfies GoogleVertexOptions;\n\t\t\t}\n\n\t\t\tconst vertexModel = model as Model<\"google-vertex\">;\n\t\t\tconst effort = clampReasoning(options.reasoning)!;\n\t\t\tconst geminiModel = vertexModel as unknown as Model<\"google-generative-ai\">;\n\n\t\t\tif (isGemini3ProModel(geminiModel) || isGemini3FlashModel(geminiModel)) {\n\t\t\t\treturn {\n\t\t\t\t\t...base,\n\t\t\t\t\tthinking: {\n\t\t\t\t\t\tenabled: true,\n\t\t\t\t\t\tlevel: getGemini3ThinkingLevel(effort, geminiModel),\n\t\t\t\t\t},\n\t\t\t\t} satisfies GoogleVertexOptions;\n\t\t\t}\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tthinking: {\n\t\t\t\t\tenabled: true,\n\t\t\t\t\tbudgetTokens: getGoogleBudget(geminiModel, effort),\n\t\t\t\t},\n\t\t\t} satisfies GoogleVertexOptions;\n\t\t}\n\n\t\tdefault: {\n\t\t\t// Exhaustiveness check\n\t\t\tconst _exhaustive: never = model.api;\n\t\t\tthrow new Error(`Unhandled API in mapOptionsForApi: ${_exhaustive}`);\n\t\t}\n\t}\n}\n\ntype ClampedReasoningEffort = Exclude<ReasoningEffort, \"xhigh\">;\n\nfunction isGemini3ProModel(model: Model<\"google-generative-ai\">): boolean {\n\t// Covers gemini-3-pro, gemini-3-pro-preview, and possible other prefixed ids in the future\n\treturn model.id.includes(\"3-pro\");\n}\n\nfunction isGemini3FlashModel(model: Model<\"google-generative-ai\">): boolean {\n\t// Covers gemini-3-flash, gemini-3-flash-preview, and possible other prefixed ids in the future\n\treturn model.id.includes(\"3-flash\");\n}\n\nfunction getGemini3ThinkingLevel(\n\teffort: ClampedReasoningEffort,\n\tmodel: Model<\"google-generative-ai\">,\n): GoogleThinkingLevel {\n\tif (isGemini3ProModel(model)) {\n\t\t// Gemini 3 Pro only supports LOW/HIGH (for now)\n\t\tswitch (effort) {\n\t\t\tcase \"minimal\":\n\t\t\tcase \"low\":\n\t\t\t\treturn \"LOW\";\n\t\t\tcase \"medium\":\n\t\t\tcase \"high\":\n\t\t\t\treturn \"HIGH\";\n\t\t}\n\t}\n\t// Gemini 3 Flash supports all four levels\n\tswitch (effort) {\n\t\tcase \"minimal\":\n\t\t\treturn \"MINIMAL\";\n\t\tcase \"low\":\n\t\t\treturn \"LOW\";\n\t\tcase \"medium\":\n\t\t\treturn \"MEDIUM\";\n\t\tcase \"high\":\n\t\t\treturn \"HIGH\";\n\t}\n}\n\nfunction getGeminiCliThinkingLevel(effort: ClampedReasoningEffort, modelId: string): GoogleThinkingLevel {\n\tif (modelId.includes(\"3-pro\")) {\n\t\t// Gemini 3 Pro only supports LOW/HIGH (for now)\n\t\tswitch (effort) {\n\t\t\tcase \"minimal\":\n\t\t\tcase \"low\":\n\t\t\t\treturn \"LOW\";\n\t\t\tcase \"medium\":\n\t\t\tcase \"high\":\n\t\t\t\treturn \"HIGH\";\n\t\t}\n\t}\n\t// Gemini 3 Flash supports all four levels\n\tswitch (effort) {\n\t\tcase \"minimal\":\n\t\t\treturn \"MINIMAL\";\n\t\tcase \"low\":\n\t\t\treturn \"LOW\";\n\t\tcase \"medium\":\n\t\t\treturn \"MEDIUM\";\n\t\tcase \"high\":\n\t\t\treturn \"HIGH\";\n\t}\n}\n\nfunction getGoogleBudget(model: Model<\"google-generative-ai\">, effort: ClampedReasoningEffort): number {\n\t// See https://ai.google.dev/gemini-api/docs/thinking#set-budget\n\tif (model.id.includes(\"2.5-pro\")) {\n\t\tconst budgets: Record<ClampedReasoningEffort, number> = {\n\t\t\tminimal: 128,\n\t\t\tlow: 2048,\n\t\t\tmedium: 8192,\n\t\t\thigh: 32768,\n\t\t};\n\t\treturn budgets[effort];\n\t}\n\n\tif (model.id.includes(\"2.5-flash\")) {\n\t\t// Covers 2.5-flash-lite as well\n\t\tconst budgets: Record<ClampedReasoningEffort, number> = {\n\t\t\tminimal: 128,\n\t\t\tlow: 2048,\n\t\t\tmedium: 8192,\n\t\t\thigh: 24576,\n\t\t};\n\t\treturn budgets[effort];\n\t}\n\n\t// Unknown model - use dynamic\n\treturn -1;\n}\n"]}

package/dist/stream.js

@@ -2,6 +2,7 @@ import { supportsXhigh } from "./models.js";
 import { streamAnthropic } from "./providers/anthropic.js";
 import { streamGoogle } from "./providers/google.js";
 import { streamGoogleGeminiCli, } from "./providers/google-gemini-cli.js";
+import { streamGoogleVertex } from "./providers/google-vertex.js";
 import { streamOpenAICompletions } from "./providers/openai-completions.js";
 import { streamOpenAIResponses } from "./providers/openai-responses.js";
 export function getEnvApiKey(provider) {
@@ -13,6 +14,13 @@ export function getEnvApiKey(provider) {
     if (provider === "anthropic") {
         return process.env.ANTHROPIC_OAUTH_TOKEN || process.env.ANTHROPIC_API_KEY;
     }
+    // Vertex AI doesn't use API keys.
+    // It relies on Google Cloud auth: `gcloud auth application-default login`.
+    // @google/genai library picks up and manages the auth automatically.
+    // Return a dummy value to maintain consistency.
+    if (provider === "google-vertex") {
+        return "vertex-ai-authenticated";
+    }
     const envMap = {
         openai: "OPENAI_API_KEY",
         google: "GEMINI_API_KEY",
@@ -44,6 +52,8 @@ export function stream(model, context, options) {
             return streamGoogle(model, context, providerOptions);
         case "google-gemini-cli":
             return streamGoogleGeminiCli(model, context, providerOptions);
+        case "google-vertex":
+            return streamGoogleVertex(model, context, providerOptions);
         default: {
             // This should never be reached if all Api cases are handled
             const _exhaustive = api;
@@ -82,16 +92,27 @@ function mapOptionsForApi(model, options, apiKey) {
             if (!options?.reasoning) {
                 return { ...base, thinkingEnabled: false };
             }
+            // Claude requires max_tokens > thinking.budget_tokens
+            // So we need to ensure maxTokens accounts for both thinking and output
             const anthropicBudgets = {
                 minimal: 1024,
                 low: 2048,
                 medium: 8192,
                 high: 16384,
             };
+            const minOutputTokens = 1024;
+            let thinkingBudget = anthropicBudgets[clampReasoning(options.reasoning)];
+            // Caller's maxTokens is the desired output; add thinking budget on top, capped at model limit
+            const maxTokens = Math.min((base.maxTokens || 0) + thinkingBudget, model.maxTokens);
+            // If not enough room for thinking + output, reduce thinking budget
+            if (maxTokens <= thinkingBudget) {
+                thinkingBudget = Math.max(0, maxTokens - minOutputTokens);
+            }
             return {
                 ...base,
+                maxTokens,
                 thinkingEnabled: true,
-                thinkingBudgetTokens: anthropicBudgets[clampReasoning(options.reasoning)],
+                thinkingBudgetTokens: thinkingBudget,
             };
         }
         case "openai-completions":
@@ -146,18 +167,54 @@ function mapOptionsForApi(model, options, apiKey) {
                     },
                 };
             }
-            // Gemini 2.x models use thinkingBudget
+            // Models using thinkingBudget (Gemini 2.x, Claude via Antigravity)
+            // Claude requires max_tokens > thinking.budget_tokens
+            // So we need to ensure maxTokens accounts for both thinking and output
             const budgets = {
                 minimal: 1024,
                 low: 2048,
                 medium: 8192,
                 high: 16384,
             };
+            const minOutputTokens = 1024;
+            let thinkingBudget = budgets[effort];
+            // Caller's maxTokens is the desired output; add thinking budget on top, capped at model limit
+            const maxTokens = Math.min((base.maxTokens || 0) + thinkingBudget, model.maxTokens);
+            // If not enough room for thinking + output, reduce thinking budget
+            if (maxTokens <= thinkingBudget) {
+                thinkingBudget = Math.max(0, maxTokens - minOutputTokens);
+            }
+            return {
+                ...base,
+                maxTokens,
+                thinking: {
+                    enabled: true,
+                    budgetTokens: thinkingBudget,
+                },
+            };
+        }
+        case "google-vertex": {
+            // Explicitly disable thinking when reasoning is not specified
+            if (!options?.reasoning) {
+                return { ...base, thinking: { enabled: false } };
+            }
+            const vertexModel = model;
+            const effort = clampReasoning(options.reasoning);
+            const geminiModel = vertexModel;
+            if (isGemini3ProModel(geminiModel) || isGemini3FlashModel(geminiModel)) {
+                return {
+                    ...base,
+                    thinking: {
+                        enabled: true,
+                        level: getGemini3ThinkingLevel(effort, geminiModel),
+                    },
+                };
+            }
             return {
                 ...base,
                 thinking: {
                     enabled: true,
-                    budgetTokens: budgets[effort],
+                    budgetTokens: getGoogleBudget(geminiModel, effort),
                 },
             };
         }