@corbat-tech/coco 2.25.14 → 2.27.0

package/dist/index.js

@@ -21,7 +21,7 @@ import { glob } from 'glob';
 import Anthropic from '@anthropic-ai/sdk';
 import { jsonrepair } from 'jsonrepair';
 import OpenAI from 'openai';
-import { GoogleGenerativeAI, FunctionCallingMode } from '@google/generative-ai';
+import { GoogleGenAI, FunctionCallingConfigMode } from '@google/genai';
 import 'events';
 import 'minimatch';
 import { simpleGit } from 'simple-git';
@@ -782,26 +782,58 @@ var init_flow = __esm({
     promisify(execFile);
   }
 });
-async function getADCAccessToken() {
+async function inspectADC() {
   try {
-    const { stdout } = await execAsync2("gcloud auth application-default print-access-token", {
+    const { stdout } = await execAsync2(PRINT_ACCESS_TOKEN_COMMAND, {
       timeout: 1e4
     });
     const accessToken = stdout.trim();
-    if (!accessToken) return null;
+    if (!accessToken) {
+      return {
+        status: "missing",
+        token: null,
+        message: "gcloud ADC is not configured.",
+        suggestion: `Run \`${ADC_LOGIN_COMMAND}\` manually, then retry Coco.`
+      };
+    }
     const expiresAt = Date.now() + 55 * 60 * 1e3;
     return {
-      accessToken,
-      expiresAt
+      status: "ok",
+      token: {
+        accessToken,
+        expiresAt
+      }
     };
   } catch (error) {
     const message = error instanceof Error ? error.message : String(error);
+    if (message.includes("scope is required but not consented")) {
+      return {
+        status: "scope_not_consented",
+        token: null,
+        message: "gcloud ADC exists, but the required Google scope was not consented for this account.",
+        suggestion: `For Vertex AI, rerun \`gcloud auth application-default login\` manually. For Gemini API OAuth, follow Google's OAuth guide with your own OAuth client and run \`gcloud auth application-default login --client-id-file=client_secret.json --scopes='${GEMINI_OAUTH_SCOPES}'\`. Otherwise use a Gemini API key in Coco.`
+      };
+    }
     if (message.includes("not logged in") || message.includes("no application default credentials")) {
-      return null;
+      return {
+        status: "missing",
+        token: null,
+        message: "No application default credentials were found for gcloud.",
+        suggestion: `Run \`${ADC_LOGIN_COMMAND}\` manually, then retry Coco.`
+      };
     }
-    return null;
+    return {
+      status: "error",
+      token: null,
+      message,
+      suggestion: `Try \`${PRINT_ACCESS_TOKEN_COMMAND}\` in your terminal to inspect the local ADC state.`
+    };
   }
 }
+async function getADCAccessToken() {
+  const result = await inspectADC();
+  return result.token;
+}
 async function getCachedADCToken() {
   if (cachedToken && cachedToken.expiresAt && Date.now() < cachedToken.expiresAt) {
     return cachedToken;
@@ -809,10 +841,16 @@ async function getCachedADCToken() {
   cachedToken = await getADCAccessToken();
   return cachedToken;
 }
-var execAsync2, cachedToken;
+var execAsync2, PRINT_ACCESS_TOKEN_COMMAND, ADC_LOGIN_COMMAND, GEMINI_OAUTH_SCOPES, cachedToken;
 var init_gcloud = __esm({
   "src/auth/gcloud.ts"() {
     execAsync2 = promisify(exec);
+    PRINT_ACCESS_TOKEN_COMMAND = "gcloud auth application-default print-access-token";
+    ADC_LOGIN_COMMAND = "gcloud auth application-default login";
+    GEMINI_OAUTH_SCOPES = [
+      "https://www.googleapis.com/auth/cloud-platform",
+      "https://www.googleapis.com/auth/generative-language.retriever"
+    ].join(",");
     cachedToken = null;
   }
 });
@@ -898,6 +936,7 @@ var init_schema = __esm({
         "codex",
         "copilot",
         "gemini",
+        "vertex",
         "kimi",
         "kimi-code",
         "lmstudio",
@@ -914,7 +953,9 @@ var init_schema = __esm({
       model: z.string().default("claude-sonnet-4-6"),
       maxTokens: z.number().min(1).max(2e5).default(8192),
       temperature: z.number().min(0).max(2).default(0),
-      timeout: z.number().min(1e3).default(12e4)
+      timeout: z.number().min(1e3).default(12e4),
+      project: z.string().optional(),
+      location: z.string().optional()
     });
     QualityConfigSchema = z.object({
       minScore: z.number().min(0).max(100).default(85),
@@ -1366,6 +1407,8 @@ function getApiKey(provider) {
       return process.env["OPENAI_API_KEY"];
     case "gemini":
       return process.env["GEMINI_API_KEY"] ?? process.env["GOOGLE_API_KEY"];
+    case "vertex":
+      return void 0;
     case "kimi":
       return process.env["KIMI_API_KEY"] ?? process.env["MOONSHOT_API_KEY"];
     case "kimi-code":
@@ -1414,6 +1457,8 @@ function getBaseUrl(provider) {
       return "https://chatgpt.com/backend-api/codex/responses";
     case "copilot":
       return process.env["COPILOT_BASE_URL"] ?? "https://api.githubcopilot.com";
+    case "vertex":
+      return process.env["VERTEX_BASE_URL"] ?? "https://aiplatform.googleapis.com/v1";
     case "groq":
       return process.env["GROQ_BASE_URL"] ?? "https://api.groq.com/openai/v1";
     case "openrouter":
@@ -1440,6 +1485,8 @@ function getDefaultModel(provider) {
       return process.env["OPENAI_MODEL"] ?? "gpt-5.4-codex";
     case "gemini":
       return process.env["GEMINI_MODEL"] ?? "gemini-3.1-pro-preview";
+    case "vertex":
+      return process.env["VERTEX_MODEL"] ?? "gemini-2.5-pro";
     case "kimi":
       return process.env["KIMI_MODEL"] ?? "kimi-k2.5";
     case "kimi-code":
@@ -1489,6 +1536,7 @@ var init_env = __esm({
       "codex",
       "copilot",
       "gemini",
+      "vertex",
       "kimi",
       "kimi-code",
       "lmstudio",
@@ -16028,19 +16076,14 @@ var CopilotProvider = class extends OpenAIProvider {
 
 // src/providers/gemini.ts
 init_errors();
-init_gcloud();
 var DEFAULT_MODEL5 = "gemini-3.1-pro-preview";
 var CONTEXT_WINDOWS5 = {
-  // Gemini 3.1 series (latest)
   "gemini-3.1-pro-preview": 1e6,
   "gemini-3.1-flash-lite-preview": 1e6,
-  // Gemini 3 series
   "gemini-3-flash-preview": 1e6,
-  // Gemini 2.5 series (production stable)
   "gemini-2.5-pro": 1048576,
   "gemini-2.5-flash": 1048576,
   "gemini-2.5-flash-lite": 1048576,
-  // Legacy
   "gemini-1.5-flash": 1e6,
   "gemini-1.5-pro": 2e6
 };
@@ -16049,129 +16092,54 @@ var GeminiProvider = class {
   name = "Google Gemini";
   client = null;
   config = {};
-  /**
-   * Initialize the provider
-   *
-   * Authentication priority:
-   * 1. API key passed in config (unless it's the ADC marker)
-   * 2. GEMINI_API_KEY environment variable
-   * 3. GOOGLE_API_KEY environment variable
-   * 4. Google Cloud ADC (gcloud auth application-default login)
-   */
   async initialize(config) {
     this.config = config;
-    const isADCMarker = config.apiKey === "__gcloud_adc__";
-    let apiKey = !isADCMarker && config.apiKey ? config.apiKey : process.env["GEMINI_API_KEY"] ?? process.env["GOOGLE_API_KEY"];
-    if (!apiKey || isADCMarker) {
-      try {
-        const adcToken = await getCachedADCToken();
-        if (adcToken) {
-          apiKey = adcToken.accessToken;
-          this.config.useADC = true;
-        }
-      } catch {
-      }
-    }
+    const apiKey = config.apiKey ?? process.env["GEMINI_API_KEY"] ?? process.env["GOOGLE_API_KEY"];
     if (!apiKey) {
       throw new ProviderError(
-        "Gemini API key not provided. Set GEMINI_API_KEY or run: gcloud auth application-default login",
+        "Gemini Developer API key not provided. Set GEMINI_API_KEY or GOOGLE_API_KEY.",
         { provider: this.id }
       );
     }
-    this.client = new GoogleGenerativeAI(apiKey);
+    this.client = new GoogleGenAI({ apiKey });
   }
-  /**
-   * Refresh ADC token if needed and reinitialize client
-   */
-  async refreshADCIfNeeded() {
-    if (!this.config.useADC) return;
-    try {
-      const adcToken = await getCachedADCToken();
-      if (adcToken) {
-        this.client = new GoogleGenerativeAI(adcToken.accessToken);
-      }
-    } catch {
-    }
-  }
-  /**
-   * Send a chat message
-   */
   async chat(messages, options) {
     this.ensureInitialized();
-    await this.refreshADCIfNeeded();
     try {
-      const model = this.client.getGenerativeModel({
-        model: options?.model ?? this.config.model ?? DEFAULT_MODEL5,
-        generationConfig: {
-          maxOutputTokens: options?.maxTokens ?? this.config.maxTokens ?? 8192,
-          temperature: options?.temperature ?? this.config.temperature ?? 0,
-          stopSequences: options?.stopSequences
-        },
-        systemInstruction: this.extractSystem(messages, options?.system)
+      const response = await this.client.models.generateContent({
+        model: this.getModel(options?.model),
+        contents: this.convertContents(messages),
+        config: this.buildConfig(messages, options)
       });
-      const { history, lastMessage } = this.convertMessages(messages);
-      const chat = model.startChat({ history });
-      const result = await chat.sendMessage(lastMessage);
-      return this.parseResponse(result);
+      return this.parseResponse(response, options?.model);
     } catch (error) {
       throw this.handleError(error);
     }
   }
-  /**
-   * Send a chat message with tool use
-   */
   async chatWithTools(messages, options) {
     this.ensureInitialized();
-    await this.refreshADCIfNeeded();
     try {
-      const tools = [
-        {
-          functionDeclarations: this.convertTools(options.tools)
-        }
-      ];
-      const model = this.client.getGenerativeModel({
-        model: options?.model ?? this.config.model ?? DEFAULT_MODEL5,
-        generationConfig: {
-          maxOutputTokens: options?.maxTokens ?? this.config.maxTokens ?? 8192,
-          temperature: options?.temperature ?? this.config.temperature ?? 0
-        },
-        systemInstruction: this.extractSystem(messages, options?.system),
-        tools,
-        toolConfig: {
-          functionCallingConfig: {
-            mode: this.convertToolChoice(options.toolChoice)
-          }
-        }
+      const response = await this.client.models.generateContent({
+        model: this.getModel(options.model),
+        contents: this.convertContents(messages),
+        config: this.buildConfig(messages, options, options.tools, options.toolChoice)
       });
-      const { history, lastMessage } = this.convertMessages(messages);
-      const chat = model.startChat({ history });
-      const result = await chat.sendMessage(lastMessage);
-      return this.parseResponseWithTools(result);
+      return this.parseResponseWithTools(response, options.model);
     } catch (error) {
       throw this.handleError(error);
     }
   }
-  /**
-   * Stream a chat response
-   */
   async *stream(messages, options) {
     this.ensureInitialized();
-    await this.refreshADCIfNeeded();
     try {
-      const model = this.client.getGenerativeModel({
-        model: options?.model ?? this.config.model ?? DEFAULT_MODEL5,
-        generationConfig: {
-          maxOutputTokens: options?.maxTokens ?? this.config.maxTokens ?? 8192,
-          temperature: options?.temperature ?? this.config.temperature ?? 0
-        },
-        systemInstruction: this.extractSystem(messages, options?.system)
+      const stream = await this.client.models.generateContentStream({
+        model: this.getModel(options?.model),
+        contents: this.convertContents(messages),
+        config: this.buildConfig(messages, options)
       });
-      const { history, lastMessage } = this.convertMessages(messages);
-      const chat = model.startChat({ history });
-      const result = await chat.sendMessageStream(lastMessage);
-      let streamStopReason;
-      for await (const chunk of result.stream) {
-        const text = chunk.text();
+      let streamStopReason = "end_turn";
+      for await (const chunk of stream) {
+        const text = chunk.text;
         if (text) {
           yield { type: "text", text };
         }
@@ -16185,116 +16153,76 @@ var GeminiProvider = class {
       throw this.handleError(error);
     }
   }
-  /**
-   * Stream a chat response with tool use
-   */
   async *streamWithTools(messages, options) {
     this.ensureInitialized();
-    await this.refreshADCIfNeeded();
     try {
-      const tools = [
-        {
-          functionDeclarations: this.convertTools(options.tools)
-        }
-      ];
-      const model = this.client.getGenerativeModel({
-        model: options?.model ?? this.config.model ?? DEFAULT_MODEL5,
-        generationConfig: {
-          maxOutputTokens: options?.maxTokens ?? this.config.maxTokens ?? 8192,
-          temperature: options?.temperature ?? this.config.temperature ?? 0
-        },
-        systemInstruction: this.extractSystem(messages, options?.system),
-        tools,
-        toolConfig: {
-          functionCallingConfig: {
-            mode: this.convertToolChoice(options.toolChoice)
-          }
-        }
+      const stream = await this.client.models.generateContentStream({
+        model: this.getModel(options.model),
+        contents: this.convertContents(messages),
+        config: this.buildConfig(messages, options, options.tools, options.toolChoice)
       });
-      const { history, lastMessage } = this.convertMessages(messages);
-      const chat = model.startChat({ history });
-      const result = await chat.sendMessageStream(lastMessage);
-      let streamStopReason;
-      let streamToolCallCounter = 0;
-      for await (const chunk of result.stream) {
-        const text = chunk.text();
+      let streamStopReason = "end_turn";
+      let fallbackToolCounter = 0;
+      const emittedToolIds = /* @__PURE__ */ new Set();
+      for await (const chunk of stream) {
+        const text = chunk.text;
         if (text) {
           yield { type: "text", text };
         }
+        const functionCalls = this.extractFunctionCalls(chunk);
+        for (const functionCall of functionCalls) {
+          const toolCallId = functionCall.id ?? `gemini_call_${++fallbackToolCounter}`;
+          if (emittedToolIds.has(toolCallId)) continue;
+          emittedToolIds.add(toolCallId);
+          const toolCall = {
+            id: toolCallId,
+            name: functionCall.name ?? "unknown_function",
+            input: functionCall.args ?? {}
+          };
+          yield {
+            type: "tool_use_start",
+            toolCall: {
+              id: toolCall.id,
+              name: toolCall.name
+            }
+          };
+          yield {
+            type: "tool_use_end",
+            toolCall
+          };
+        }
         const finishReason = chunk.candidates?.[0]?.finishReason;
-        if (finishReason) {
+        if (functionCalls.length > 0) {
+          streamStopReason = "tool_use";
+        } else if (finishReason) {
           streamStopReason = this.mapFinishReason(finishReason);
         }
-        const candidate = chunk.candidates?.[0];
-        if (candidate?.content?.parts) {
-          for (const part of candidate.content.parts) {
-            if ("functionCall" in part && part.functionCall) {
-              const funcCall = part.functionCall;
-              streamToolCallCounter++;
-              const toolCall = {
-                id: `gemini_call_${streamToolCallCounter}`,
-                name: funcCall.name,
-                input: funcCall.args ?? {}
-              };
-              yield {
-                type: "tool_use_start",
-                toolCall: {
-                  id: toolCall.id,
-                  name: toolCall.name
-                }
-              };
-              yield {
-                type: "tool_use_end",
-                toolCall
-              };
-            }
-          }
-        }
       }
       yield { type: "done", stopReason: streamStopReason };
     } catch (error) {
       throw this.handleError(error);
     }
   }
-  /**
-   * Count tokens (approximate)
-   *
-   * Gemini uses a SentencePiece tokenizer. The average ratio varies:
-   * - English text: ~4 characters per token
-   * - Code: ~3.2 characters per token
-   * - Mixed content: ~3.5 characters per token
-   *
-   * Using 3.5 as the default provides a better estimate for typical
-   * coding agent workloads which mix code and natural language.
-   */
   countTokens(text) {
     if (!text) return 0;
     return Math.ceil(text.length / 3.5);
   }
-  /**
-   * Get context window size
-   */
   getContextWindow() {
     const model = this.config.model ?? DEFAULT_MODEL5;
     return CONTEXT_WINDOWS5[model] ?? 1e6;
   }
-  /**
-   * Check if provider is available
-   */
   async isAvailable() {
     if (!this.client) return false;
     try {
-      const modelName = this.config.model ?? DEFAULT_MODEL5;
-      const model = this.client.getGenerativeModel({ model: modelName });
-      await model.generateContent("hi");
+      await this.client.models.generateContent({
+        model: this.getModel(),
+        contents: "hi"
+      });
       return true;
     } catch {
       return false;
     }
   }
-  /**
-   * Ensure client is initialized
-   */
   ensureInitialized() {
     if (!this.client) {
       throw new ProviderError("Provider not initialized. Call initialize() first.", {
@@ -16302,13 +16230,24 @@ var GeminiProvider = class {
       });
     }
   }
-  /**
-   * Extract system prompt from messages array or options.
-   *
-   * convertMessages() skips system-role messages ("handled via systemInstruction"),
-   * but all callers forgot to also pass it via options.system. This helper bridges
-   * that gap — mirrors the same fix applied to AnthropicProvider.
-   */
+  getModel(model) {
+    return model ?? this.config.model ?? DEFAULT_MODEL5;
+  }
+  buildConfig(messages, options, tools, toolChoice) {
+    const config = {
+      maxOutputTokens: options?.maxTokens ?? this.config.maxTokens ?? 8192,
+      temperature: options?.temperature ?? this.config.temperature ?? 0,
+      stopSequences: options?.stopSequences,
+      systemInstruction: this.extractSystem(messages, options?.system)
+    };
+    if (tools && tools.length > 0) {
+      config.tools = [{ functionDeclarations: this.convertTools(tools) }];
+      config.toolConfig = {
+        functionCallingConfig: this.convertToolChoice(toolChoice)
+      };
+    }
+    return config;
+  }
   extractSystem(messages, optionsSystem) {
     if (optionsSystem !== void 0) return optionsSystem;
     const systemMsg = messages.find((m) => m.role === "system");
@@ -16317,55 +16256,36 @@ var GeminiProvider = class {
     const text = systemMsg.content.filter((b) => b.type === "text").map((b) => b.text).join("");
     return text || void 0;
   }
-  /**
-   * Convert messages to Gemini format
-   */
-  convertMessages(messages) {
+  convertContents(messages) {
     const toolNameByUseId = this.buildToolUseNameMap(messages);
     const conversation = messages.filter((m) => m.role !== "system");
-    const history = [];
-    let lastUserMessage = "";
-    for (let i = 0; i < conversation.length; i++) {
-      const msg = conversation[i];
-      const isLastMessage = i === conversation.length - 1;
+    const contents = [];
+    for (const msg of conversation) {
       if (msg.role === "user") {
         if (Array.isArray(msg.content) && msg.content[0]?.type === "tool_result") {
-          const functionResponses = [];
+          const parts = [];
           for (const block of msg.content) {
             if (block.type === "tool_result") {
               const toolResult = block;
-              functionResponses.push({
+              parts.push({
                 functionResponse: {
-                  // Gemini expects the function name in functionResponse.name.
-                  // Recover it from prior assistant tool_use blocks when possible.
+                  id: toolResult.tool_use_id,
                   name: toolNameByUseId.get(toolResult.tool_use_id) ?? toolResult.tool_use_id,
                   response: { result: toolResult.content }
                 }
               });
             }
           }
-          history.push({ role: "function", parts: functionResponses });
-          if (isLastMessage) {
-            lastUserMessage = "";
-          }
+          contents.push({ role: "user", parts });
         } else {
-          const parts = this.convertContent(msg.content);
-          if (isLastMessage) {
-            lastUserMessage = parts;
-          } else {
-            history.push({ role: "user", parts });
-          }
+          contents.push({ role: "user", parts: this.convertContent(msg.content) });
         }
       } else if (msg.role === "assistant") {
-        const parts = this.convertContent(msg.content);
-        history.push({ role: "model", parts });
+        contents.push({ role: "model", parts: this.convertContent(msg.content) });
       }
     }
-    return { history, lastMessage: lastUserMessage };
+    return contents.length > 0 ? contents : [{ role: "user", parts: [{ text: "" }] }];
   }
-  /**
-   * Build a map from tool_use IDs to function names from assistant history.
-   */
   buildToolUseNameMap(messages) {
     const map = /* @__PURE__ */ new Map();
     for (const msg of messages) {
@@ -16378,9 +16298,6 @@ var GeminiProvider = class {
     }
     return map;
   }
-  /**
-   * Convert content to Gemini parts
-   */
   convertContent(content) {
     if (typeof content === "string") {
       return [{ text: content }];
@@ -16390,27 +16307,26 @@ var GeminiProvider = class {
       if (block.type === "text") {
         parts.push({ text: block.text });
       } else if (block.type === "image") {
-        const imgBlock = block;
+        const image = block;
         parts.push({
           inlineData: {
-            data: imgBlock.source.data,
-            mimeType: imgBlock.source.media_type
+            data: image.source.data,
+            mimeType: image.source.media_type
           }
         });
       } else if (block.type === "tool_use") {
+        const toolUse = block;
         parts.push({
           functionCall: {
-            name: block.name,
-            args: block.input
+            id: toolUse.id,
+            name: toolUse.name,
+            args: toolUse.input
           }
         });
       }
     }
     return parts.length > 0 ? parts : [{ text: "" }];
   }
-  /**
-   * Convert tools to Gemini format
-   */
   convertTools(tools) {
     return tools.map((tool) => ({
       name: tool.name,
@@ -16418,72 +16334,58 @@ var GeminiProvider = class {
       parameters: tool.input_schema
     }));
   }
-  /**
-   * Convert tool choice to Gemini format
-   */
   convertToolChoice(choice) {
-    if (!choice || choice === "auto") return FunctionCallingMode.AUTO;
-    if (choice === "any") return FunctionCallingMode.ANY;
-    return FunctionCallingMode.AUTO;
+    if (!choice || choice === "auto") {
+      return { mode: FunctionCallingConfigMode.AUTO };
+    }
+    if (choice === "any") {
+      return { mode: FunctionCallingConfigMode.ANY };
+    }
+    return {
+      mode: FunctionCallingConfigMode.ANY,
+      allowedFunctionNames: [choice.name]
+    };
   }
-  /**
-   * Parse response from Gemini
-   */
-  parseResponse(result) {
-    const response = result.response;
-    const text = response.text();
+  extractFunctionCalls(response) {
+    if (response.functionCalls && response.functionCalls.length > 0) {
+      return response.functionCalls;
+    }
+    const candidate = response.candidates?.[0];
+    const parts = candidate?.content?.parts ?? [];
+    return parts.filter((part) => !!part.functionCall).map((part) => part.functionCall).filter(Boolean);
+  }
+  parseResponse(response, model) {
     const usage = response.usageMetadata;
     return {
       id: `gemini-${Date.now()}`,
-      content: text,
+      content: response.text ?? "",
       stopReason: this.mapFinishReason(response.candidates?.[0]?.finishReason),
       usage: {
         inputTokens: usage?.promptTokenCount ?? 0,
         outputTokens: usage?.candidatesTokenCount ?? 0
       },
-      model: this.config.model ?? DEFAULT_MODEL5
+      model: this.getModel(model)
     };
   }
-  /**
-   * Parse response with tool calls from Gemini
-   */
-  parseResponseWithTools(result) {
-    const response = result.response;
-    const candidate = response.candidates?.[0];
+  parseResponseWithTools(response, model) {
     const usage = response.usageMetadata;
-    let textContent = "";
-    const toolCalls = [];
-    if (candidate?.content?.parts) {
-      let toolIndex = 0;
-      for (const part of candidate.content.parts) {
-        if ("text" in part && part.text) {
-          textContent += part.text;
-        }
-        if ("functionCall" in part && part.functionCall) {
-          toolIndex++;
-          toolCalls.push({
-            id: `gemini_call_${toolIndex}`,
-            name: part.functionCall.name,
-            input: part.functionCall.args ?? {}
-          });
-        }
-      }
-    }
+    const toolCalls = this.extractFunctionCalls(response).map((functionCall, index) => ({
+      id: functionCall.id ?? `gemini_call_${index + 1}`,
+      name: functionCall.name ?? "unknown_function",
+      input: functionCall.args ?? {}
+    }));
     return {
       id: `gemini-${Date.now()}`,
-      content: textContent,
-      stopReason: toolCalls.length > 0 ? "tool_use" : this.mapFinishReason(candidate?.finishReason),
+      content: response.text ?? "",
+      stopReason: toolCalls.length > 0 ? "tool_use" : this.mapFinishReason(response.candidates?.[0]?.finishReason),
       usage: {
         inputTokens: usage?.promptTokenCount ?? 0,
         outputTokens: usage?.candidatesTokenCount ?? 0
       },
-      model: this.config.model ?? DEFAULT_MODEL5,
+      model: this.getModel(model),
       toolCalls
     };
   }
-  /**
-   * Map finish reason to our format
-   */
   mapFinishReason(reason) {
     switch (reason) {
       case "STOP":
@@ -16498,9 +16400,6 @@ var GeminiProvider = class {
         return "end_turn";
     }
   }
-  /**
-   * Handle API errors
-   */
   handleError(error) {
     const message = error instanceof Error ? error.message : String(error);
     const msg = message.toLowerCase();
@@ -16519,6 +16418,422 @@ var GeminiProvider = class {
   }
 };
 
+// src/providers/vertex.ts
+init_errors();
+init_gcloud();
+var DEFAULT_MODEL6 = "gemini-2.5-pro";
+var DEFAULT_BASE_URL = "https://aiplatform.googleapis.com/v1";
+var DEFAULT_LOCATION = "global";
+var CONTEXT_WINDOWS6 = {
+  "gemini-2.5-pro": 1048576,
+  "gemini-2.5-flash": 1048576,
+  "gemini-2.5-flash-lite": 1048576,
+  "gemini-2.0-flash-001": 1048576,
+  "gemini-2.0-flash-lite-001": 1048576
+};
+var VertexProvider = class {
+  id = "vertex";
+  name = "Google Vertex AI Gemini";
+  config = {};
+  project = "";
+  location = DEFAULT_LOCATION;
+  retryConfig = DEFAULT_RETRY_CONFIG;
+  async initialize(config) {
+    this.config = config;
+    this.project = config.project ?? process.env["VERTEX_PROJECT"] ?? process.env["GOOGLE_CLOUD_PROJECT"] ?? process.env["GCLOUD_PROJECT"] ?? "";
+    this.location = config.location ?? process.env["VERTEX_LOCATION"] ?? process.env["GOOGLE_CLOUD_LOCATION"] ?? DEFAULT_LOCATION;
+    if (!this.project.trim()) {
+      throw new ProviderError(
+        "Vertex AI project not configured. Set provider.project, VERTEX_PROJECT, or GOOGLE_CLOUD_PROJECT.",
+        { provider: this.id }
+      );
+    }
+    const token = await getCachedADCToken();
+    if (!token) {
+      throw new ProviderError(
+        "Vertex AI ADC is not configured. Run `gcloud auth application-default login` manually, then retry.",
+        { provider: this.id }
+      );
+    }
+  }
+  async chat(messages, options) {
+    this.ensureInitialized();
+    return withRetry(async () => {
+      const response = await this.generateContent(messages, options);
+      return this.parseResponse(response, options?.model);
+    }, this.retryConfig);
+  }
+  async chatWithTools(messages, options) {
+    this.ensureInitialized();
+    return withRetry(async () => {
+      const response = await this.generateContent(
+        messages,
+        options,
+        options.tools,
+        options.toolChoice
+      );
+      return this.parseResponseWithTools(response, options.model);
+    }, this.retryConfig);
+  }
+  async *stream(messages, options) {
+    this.ensureInitialized();
+    const stream = await this.streamGenerateContent(messages, options);
+    let stopReason = "end_turn";
+    for await (const chunk of stream) {
+      const candidate = chunk.candidates?.[0];
+      const parts = candidate?.content?.parts ?? [];
+      for (const part of parts) {
+        if (part.text) {
+          yield { type: "text", text: part.text };
+        }
+      }
+      stopReason = this.mapFinishReason(candidate?.finishReason);
+    }
+    yield { type: "done", stopReason };
+  }
+  async *streamWithTools(messages, options) {
+    this.ensureInitialized();
+    const stream = await this.streamGenerateContent(
+      messages,
+      options,
+      options.tools,
+      options.toolChoice
+    );
+    let stopReason = "end_turn";
+    let streamToolCallCounter = 0;
+    for await (const chunk of stream) {
+      const candidate = chunk.candidates?.[0];
+      const parts = candidate?.content?.parts ?? [];
+      for (const part of parts) {
+        if (part.text) {
+          yield { type: "text", text: part.text };
+        }
+        if (part.functionCall) {
+          streamToolCallCounter++;
+          yield {
+            type: "tool_use_start",
+            toolCall: {
+              id: `vertex_call_${streamToolCallCounter}`,
+              name: part.functionCall.name,
+              input: part.functionCall.args ?? {}
+            }
+          };
+          yield {
+            type: "tool_use_end",
+            toolCall: {
+              id: `vertex_call_${streamToolCallCounter}`,
+              name: part.functionCall.name,
+              input: part.functionCall.args ?? {}
+            }
+          };
+        }
+      }
+      stopReason = parts.some((part) => part.functionCall) ? "tool_use" : this.mapFinishReason(candidate?.finishReason);
+    }
+    yield { type: "done", stopReason };
+  }
+  countTokens(text) {
+    return Math.ceil(text.length / 4);
+  }
+  getContextWindow() {
+    return CONTEXT_WINDOWS6[this.config.model ?? DEFAULT_MODEL6] ?? 1048576;
+  }
+  async isAvailable() {
+    try {
+      await this.generateContent([{ role: "user", content: "hi" }], { maxTokens: 8 });
+      return true;
+    } catch {
+      return false;
+    }
+  }
+  ensureInitialized() {
+    if (!this.project) {
+      throw new ProviderError("Provider not initialized. Call initialize() first.", {
+        provider: this.id
+      });
+    }
+  }
+  getModel(model) {
+    return model ?? this.config.model ?? DEFAULT_MODEL6;
+  }
+  getResolvedBaseUrl() {
+    if (this.config.baseUrl && this.config.baseUrl.trim()) {
+      return this.config.baseUrl;
+    }
+    if (this.location === DEFAULT_LOCATION) {
+      return DEFAULT_BASE_URL;
+    }
+    return `https://${encodeURIComponent(this.location)}-aiplatform.googleapis.com/v1`;
+  }
+  buildEndpoint(model, stream = false) {
+    const action = stream ? "streamGenerateContent?alt=sse" : "generateContent";
+    return `${this.getResolvedBaseUrl()}/projects/${encodeURIComponent(this.project)}/locations/${encodeURIComponent(this.location)}/publishers/google/models/${encodeURIComponent(this.getModel(model))}:${action}`;
+  }
+  async getHeaders() {
+    const token = await getCachedADCToken();
+    if (!token) {
+      throw new ProviderError(
+        "Vertex AI ADC token is unavailable. Re-authenticate with gcloud and retry.",
+        { provider: this.id }
+      );
+    }
+    return {
+      "Content-Type": "application/json",
+      Authorization: `Bearer ${token.accessToken}`,
+      "x-goog-user-project": this.project
+    };
+  }
+  extractSystem(messages, optionsSystem) {
+    if (optionsSystem !== void 0) return optionsSystem;
+    const systemMsg = messages.find((m) => m.role === "system");
+    if (!systemMsg) return void 0;
+    if (typeof systemMsg.content === "string") return systemMsg.content;
+    const text = systemMsg.content.filter((b) => b.type === "text").map((b) => b.text).join("");
+    return text || void 0;
+  }
+  buildToolUseNameMap(messages) {
+    const map = /* @__PURE__ */ new Map();
+    for (const msg of messages) {
+      if (msg.role !== "assistant" || !Array.isArray(msg.content)) continue;
+      for (const block of msg.content) {
+        if (block.type === "tool_use") {
+          map.set(block.id, block.name);
+        }
+      }
+    }
+    return map;
+  }
+  convertContents(messages) {
+    const toolNameByUseId = this.buildToolUseNameMap(messages);
+    const conversation = messages.filter((m) => m.role !== "system");
+    const contents = [];
+    for (let i = 0; i < conversation.length; i++) {
+      const msg = conversation[i];
+      if (msg.role === "user") {
+        if (Array.isArray(msg.content) && msg.content[0]?.type === "tool_result") {
+          const functionResponses = [];
+          for (const block of msg.content) {
+            if (block.type === "tool_result") {
+              const toolResult = block;
+              functionResponses.push({
+                functionResponse: {
+                  name: toolNameByUseId.get(toolResult.tool_use_id) ?? toolResult.tool_use_id,
+                  response: { result: toolResult.content }
+                }
+              });
+            }
+          }
+          contents.push({ role: "user", parts: functionResponses });
+        } else {
+          contents.push({ role: "user", parts: this.convertContent(msg.content) });
+        }
+      } else if (msg.role === "assistant") {
+        contents.push({ role: "model", parts: this.convertContent(msg.content) });
+      }
+    }
+    return contents.length > 0 ? contents : [{ role: "user", parts: [{ text: "" }] }];
+  }
+  convertContent(content) {
+    if (typeof content === "string") return [{ text: content }];
+    const parts = [];
+    for (const block of content) {
+      if (block.type === "text") {
+        parts.push({ text: block.text });
+      } else if (block.type === "image") {
+        const image = block;
+        parts.push({
+          inlineData: {
+            data: image.source.data,
+            mimeType: image.source.media_type
+          }
+        });
+      } else if (block.type === "tool_use") {
+        const toolUse = block;
+        parts.push({
+          functionCall: {
+            name: toolUse.name,
+            args: toolUse.input
+          }
+        });
+      }
+    }
+    return parts.length > 0 ? parts : [{ text: "" }];
+  }
+  convertTools(tools) {
+    return [
+      {
+        functionDeclarations: tools.map((tool) => ({
+          name: tool.name,
+          description: tool.description,
+          parameters: tool.input_schema
+        }))
+      }
+    ];
+  }
+  convertToolChoice(choice) {
+    if (!choice || choice === "auto") {
+      return { functionCallingConfig: { mode: "AUTO" } };
+    }
+    if (choice === "any") {
+      return { functionCallingConfig: { mode: "ANY" } };
+    }
+    return { functionCallingConfig: { mode: "ANY", allowedFunctionNames: [choice.name] } };
+  }
+  buildRequestBody(messages, options, tools, toolChoice) {
+    const body = {
+      contents: this.convertContents(messages),
+      generationConfig: {
+        maxOutputTokens: options?.maxTokens ?? this.config.maxTokens ?? 8192,
+        temperature: options?.temperature ?? this.config.temperature ?? 0,
+        stopSequences: options?.stopSequences
+      }
+    };
+    const systemInstruction = this.extractSystem(messages, options?.system);
+    if (systemInstruction) {
+      body["systemInstruction"] = {
+        parts: [{ text: systemInstruction }]
+      };
+    }
+    if (tools && tools.length > 0) {
+      body["tools"] = this.convertTools(tools);
+      const convertedChoice = this.convertToolChoice(toolChoice);
+      if (convertedChoice) {
+        body["toolConfig"] = convertedChoice;
+      }
+    }
+    return body;
+  }
+  async generateContent(messages, options, tools, toolChoice) {
+    const response = await fetch(this.buildEndpoint(options?.model), {
+      method: "POST",
+      headers: await this.getHeaders(),
+      body: JSON.stringify(this.buildRequestBody(messages, options, tools, toolChoice)),
+      signal: options?.signal
+    });
+    if (!response.ok) {
+      throw await this.buildHttpError(response);
+    }
+    const data = await response.json();
+    if (data.error?.message) {
+      throw new ProviderError(data.error.message, {
+        provider: this.id,
+        statusCode: data.error.code
+      });
+    }
+    return data;
+  }
+  async *streamGenerateContent(messages, options, tools, toolChoice) {
+    const response = await fetch(this.buildEndpoint(options?.model, true), {
+      method: "POST",
+      headers: await this.getHeaders(),
+      body: JSON.stringify(this.buildRequestBody(messages, options, tools, toolChoice)),
+      signal: options?.signal
+    });
+    if (!response.ok) {
+      throw await this.buildHttpError(response);
+    }
+    if (!response.body) {
+      throw new ProviderError("Vertex AI streaming response body is empty.", {
+        provider: this.id
+      });
+    }
+    const reader = response.body.getReader();
+    const decoder = new TextDecoder();
+    let buffer = "";
+    while (true) {
+      const { value, done } = await reader.read();
+      if (done) break;
+      buffer += decoder.decode(value, { stream: true });
+      while (true) {
+        const eventBoundary = buffer.indexOf("\n\n");
+        if (eventBoundary === -1) break;
+        const rawEvent = buffer.slice(0, eventBoundary);
+        buffer = buffer.slice(eventBoundary + 2);
+        const dataLines = rawEvent.split("\n").filter((line) => line.startsWith("data:")).map((line) => line.slice(5).trim()).filter(Boolean);
+        for (const line of dataLines) {
+          if (line === "[DONE]") return;
+          yield JSON.parse(line);
+        }
+      }
+    }
+    const trailing = buffer.trim();
+    if (trailing.startsWith("data:")) {
+      const line = trailing.slice(5).trim();
+      if (line && line !== "[DONE]") {
+        yield JSON.parse(line);
+      }
+    }
+  }
+  parseResponse(response, model) {
+    const candidate = response.candidates?.[0];
+    const text = (candidate?.content?.parts ?? []).filter((part) => part.text).map((part) => part.text).join("");
+    return {
+      id: `vertex-${Date.now()}`,
+      content: text,
+      stopReason: this.mapFinishReason(candidate?.finishReason),
+      usage: {
+        inputTokens: response.usageMetadata?.promptTokenCount ?? 0,
+        outputTokens: response.usageMetadata?.candidatesTokenCount ?? 0
+      },
+      model: this.getModel(model)
+    };
+  }
+  parseResponseWithTools(response, model) {
+    const candidate = response.candidates?.[0];
+    const parts = candidate?.content?.parts ?? [];
+    const toolCalls = [];
+    let textContent = "";
+    let toolIndex = 0;
+    for (const part of parts) {
+      if (part.text) {
+        textContent += part.text;
+      }
+      if (part.functionCall) {
+        toolIndex++;
+        toolCalls.push({
+          id: `vertex_call_${toolIndex}`,
+          name: part.functionCall.name,
+          input: part.functionCall.args ?? {}
+        });
+      }
+    }
+    return {
+      id: `vertex-${Date.now()}`,
+      content: textContent,
+      stopReason: toolCalls.length > 0 ? "tool_use" : this.mapFinishReason(candidate?.finishReason),
+      usage: {
+        inputTokens: response.usageMetadata?.promptTokenCount ?? 0,
+        outputTokens: response.usageMetadata?.candidatesTokenCount ?? 0
+      },
+      model: this.getModel(model),
+      toolCalls
+    };
+  }
+  mapFinishReason(reason) {
+    switch (reason) {
+      case "STOP":
+        return "end_turn";
+      case "MAX_TOKENS":
+        return "max_tokens";
+      case "SAFETY":
+      case "RECITATION":
+      case "OTHER":
+        return "stop_sequence";
+      default:
+        return "end_turn";
+    }
+  }
+  async buildHttpError(response) {
+    const body = await response.text();
+    const retryable = response.status === 429 || response.status >= 500;
+    return new ProviderError(`Vertex AI error: ${response.status} - ${body}`, {
+      provider: this.id,
+      statusCode: response.status,
+      retryable
+    });
+  }
+};
+
 // src/providers/circuit-breaker.ts
 init_errors();
 var DEFAULT_CIRCUIT_BREAKER_CONFIG = {
@@ -16820,6 +17135,11 @@ function normalizeProviderModel(model) {
   const trimmed = model.trim();
   return trimmed.length > 0 ? trimmed : void 0;
 }
+function normalizeOptional(value) {
+  if (typeof value !== "string") return void 0;
+  const trimmed = value.trim();
+  return trimmed.length > 0 ? trimmed : void 0;
+}
 async function createProvider(type, config = {}) {
   let provider;
   const mergedConfig = {
@@ -16828,7 +17148,11 @@ async function createProvider(type, config = {}) {
     model: normalizeProviderModel(config.model) ?? getDefaultModel(type),
     maxTokens: config.maxTokens,
     temperature: config.temperature,
-    timeout: config.timeout
+    timeout: config.timeout,
+    project: normalizeOptional(config.project) ?? (type === "vertex" ? normalizeOptional(
+      process.env["VERTEX_PROJECT"] ?? process.env["GOOGLE_CLOUD_PROJECT"] ?? process.env["GCLOUD_PROJECT"]
+    ) : void 0),
+    location: normalizeOptional(config.location) ?? (type === "vertex" ? normalizeOptional(process.env["VERTEX_LOCATION"] ?? process.env["GOOGLE_CLOUD_LOCATION"]) : void 0)
   };
   switch (type) {
     case "anthropic":
@@ -16846,6 +17170,9 @@ async function createProvider(type, config = {}) {
     case "gemini":
       provider = new GeminiProvider();
       break;
+    case "vertex":
+      provider = new VertexProvider();
+      break;
     case "kimi":
       provider = createKimiProvider(mergedConfig);
       break;
@@ -24902,7 +25229,7 @@ Examples:
         description = response.choices[0]?.message?.content ?? "No description generated";
       } else if (selectedProvider === "gemini") {
         model = "gemini-2.0-flash";
-        const { GoogleGenerativeAI: GoogleGenerativeAI2 } = await import('@google/generative-ai');
+        const { GoogleGenAI: GoogleGenAI2 } = await import('@google/genai');
         const apiKey = process.env.GOOGLE_API_KEY ?? process.env.GEMINI_API_KEY;
         if (!apiKey) {
           throw new ToolError(
@@ -24910,18 +25237,25 @@ Examples:
             { tool: "read_image" }
           );
         }
-        const genAI = new GoogleGenerativeAI2(apiKey);
-        const genModel = genAI.getGenerativeModel({ model });
-        const result = await genModel.generateContent([
-          effectivePrompt,
-          {
-            inlineData: {
-              data: base64,
-              mimeType
+        const genAI = new GoogleGenAI2({ apiKey });
+        const result = await genAI.models.generateContent({
+          model,
+          contents: [
+            {
+              role: "user",
+              parts: [
+                { text: effectivePrompt },
+                {
+                  inlineData: {
+                    data: base64,
+                    mimeType
+                  }
+                }
+              ]
             }
-          }
-        ]);
-        description = result.response.text() ?? "No description generated";
+          ]
+        });
+        description = result.text ?? "No description generated";
       } else {
         throw new ToolError(`Unsupported provider: ${selectedProvider}`, {
           tool: "read_image"
@@ -24933,7 +25267,7 @@ Examples:
         const pkgMap = {
           anthropic: "@anthropic-ai/sdk",
           openai: "openai",
-          gemini: "@google/generative-ai"
+          gemini: "@google/genai"
         };
         const pkg = pkgMap[selectedProvider] ?? selectedProvider;
         throw new ToolError(`Provider SDK not installed. Run: pnpm add ${pkg}`, {