@aigne/gemini 0.14.16 → 1.74.0-beta

package/lib/cjs/gemini-chat-model.d.ts

@@ -1,117 +0,0 @@
-import { type AgentInvokeOptions, type AgentProcessResult, ChatModel, type ChatModelInput, type ChatModelInputOptions, type ChatModelOptions, type ChatModelOutput } from "@aigne/core";
-import { type PromiseOrValue } from "@aigne/core/utils/type-utils.js";
-import { GoogleGenAI, type GoogleGenAIOptions, ThinkingLevel } from "@google/genai";
-export interface GeminiChatModelOptions extends ChatModelOptions {
-    /**
-     * API key for Gemini API
-     *
-     * If not provided, will look for GEMINI_API_KEY or GOOGLE_API_KEY in environment variables
-     */
-    apiKey?: string;
-    /**
-     * Optional client options for the Gemini SDK
-     */
-    clientOptions?: Partial<GoogleGenAIOptions>;
-}
-/**
- * Implementation of the ChatModel interface for Google's Gemini API
- *
- * @example
- * Here's how to create and use a Gemini chat model:
- * {@includeCode ../test/gemini-chat-model.test.ts#example-gemini-chat-model}
- *
- * @example
- * Here's an example with streaming response:
- * {@includeCode ../test/gemini-chat-model.test.ts#example-gemini-chat-model-streaming}
- */
-export declare class GeminiChatModel extends ChatModel {
-    options?: GeminiChatModelOptions | undefined;
-    constructor(options?: GeminiChatModelOptions | undefined);
-    protected apiKeyEnvName: string;
-    protected _googleClient?: GoogleGenAI;
-    get googleClient(): GoogleGenAI;
-    get credential(): {
-        apiKey: string | undefined;
-        model: string;
-    };
-    get modelOptions(): Partial<{
-        [x: string]: unknown;
-        model?: string | {
-            $get: string;
-        } | undefined;
-        temperature?: number | {
-            $get: string;
-        } | undefined;
-        topP?: number | {
-            $get: string;
-        } | undefined;
-        frequencyPenalty?: number | {
-            $get: string;
-        } | undefined;
-        presencePenalty?: number | {
-            $get: string;
-        } | undefined;
-        parallelToolCalls?: boolean | {
-            $get: string;
-        } | undefined;
-        modalities?: import("@aigne/core").Modality[] | {
-            $get: string;
-        } | undefined;
-        preferInputFileType?: "url" | "file" | {
-            $get: string;
-        } | undefined;
-        reasoningEffort?: number | "minimal" | "low" | "medium" | "high" | {
-            $get: string;
-        } | undefined;
-        cacheConfig?: import("@aigne/core").CacheConfig | {
-            $get: string;
-        } | undefined;
-    }> | undefined;
-    countTokens(input: ChatModelInput): Promise<number>;
-    private contentUnionToContent;
-    process(input: ChatModelInput, options: AgentInvokeOptions): PromiseOrValue<AgentProcessResult<ChatModelOutput>>;
-    protected thinkingBudgetModelMap: ({
-        pattern: RegExp;
-        support: boolean;
-        type?: undefined;
-        min?: undefined;
-        max?: undefined;
-    } | {
-        pattern: RegExp;
-        support: boolean;
-        type: string;
-        min?: undefined;
-        max?: undefined;
-    } | {
-        pattern: RegExp;
-        support: boolean;
-        min: number;
-        max: number;
-        type?: undefined;
-    })[];
-    protected thinkingBudgetLevelMap: {
-        high: number;
-        medium: number;
-        low: number;
-        minimal: number;
-    };
-    protected thinkingLevelMap: {
-        high: ThinkingLevel;
-        medium: ThinkingLevel;
-        low: ThinkingLevel;
-        minimal: ThinkingLevel;
-    };
-    protected getThinkingBudget(model: string, effort: ChatModelInputOptions["reasoningEffort"]): {
-        support: boolean;
-        budget?: number;
-        level?: ThinkingLevel;
-    };
-    private getParameters;
-    private processInput;
-    private buildConfig;
-    private buildTools;
-    private buildVideoContentParts;
-    private buildContents;
-    private contentToParts;
-    private ensureMessagesHasUserMessage;
-}

package/lib/cjs/gemini-chat-model.js

@@ -1,564 +0,0 @@
-"use strict";
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.GeminiChatModel = void 0;
-const core_1 = require("@aigne/core");
-const logger_js_1 = require("@aigne/core/utils/logger.js");
-const model_utils_js_1 = require("@aigne/core/utils/model-utils.js");
-const type_utils_js_1 = require("@aigne/core/utils/type-utils.js");
-const index_js_1 = require("@aigne/platform-helpers/nodejs/index.js");
-const uuid_1 = require("@aigne/uuid");
-const genai_1 = require("@google/genai");
-const yaml_1 = require("yaml");
-const zod_1 = require("zod");
-const zod_to_json_schema_1 = require("zod-to-json-schema");
-const GEMINI_DEFAULT_CHAT_MODEL = "gemini-2.0-flash";
-const OUTPUT_FUNCTION_NAME = "output";
-const NEED_UPLOAD_MAX_FILE_SIZE_MB = 20;
-/**
- * Implementation of the ChatModel interface for Google's Gemini API
- *
- * @example
- * Here's how to create and use a Gemini chat model:
- * {@includeCode ../test/gemini-chat-model.test.ts#example-gemini-chat-model}
- *
- * @example
- * Here's an example with streaming response:
- * {@includeCode ../test/gemini-chat-model.test.ts#example-gemini-chat-model-streaming}
- */
-class GeminiChatModel extends core_1.ChatModel {
-    options;
-    constructor(options) {
-        super({
-            ...options,
-            model: options?.model || GEMINI_DEFAULT_CHAT_MODEL,
-        });
-        this.options = options;
-    }
-    apiKeyEnvName = "GEMINI_API_KEY";
-    _googleClient;
-    get googleClient() {
-        if (this._googleClient)
-            return this._googleClient;
-        const { apiKey } = this.credential;
-        if (!apiKey)
-            throw new Error(`${this.name} requires an API key. Please provide it via \`options.apiKey\`, or set the \`${this.apiKeyEnvName}\` environment variable`);
-        this._googleClient ??= new genai_1.GoogleGenAI({
-            apiKey,
-            ...this.options?.clientOptions,
-        });
-        return this._googleClient;
-    }
-    get credential() {
-        const apiKey = this.options?.apiKey ||
-            process.env[this.apiKeyEnvName] ||
-            process.env.GEMINI_API_KEY ||
-            process.env.GOOGLE_API_KEY;
-        return {
-            apiKey,
-            model: this.options?.model || GEMINI_DEFAULT_CHAT_MODEL,
-        };
-    }
-    get modelOptions() {
-        return this.options?.modelOptions;
-    }
-    async countTokens(input) {
-        const { model, ...request } = await this.getParameters(input);
-        const contents = [];
-        const { systemInstruction, tools } = request.config ?? {};
-        if (systemInstruction)
-            contents.push(this.contentUnionToContent(systemInstruction));
-        if (tools?.length)
-            contents.push({ role: "system", parts: [{ text: JSON.stringify(tools) }] });
-        contents.push(...[request.contents].flat().map(this.contentUnionToContent));
-        const tokens = (await this.googleClient.models.countTokens({
-            model,
-            contents,
-        })).totalTokens;
-        if (!(0, type_utils_js_1.isNil)(tokens))
-            return tokens;
-        return super.countTokens(input);
-    }
-    contentUnionToContent(content) {
-        if (typeof content === "object" && "parts" in content) {
-            return { role: "system", parts: content.parts };
-        }
-        else if (typeof content === "string") {
-            return { role: "system", parts: [{ text: content }] };
-        }
-        else if (Array.isArray(content)) {
-            return {
-                role: "system",
-                parts: content.map((i) => (typeof i === "string" ? { text: i } : i)),
-            };
-        }
-        else {
-            return { role: "system", parts: [content] };
-        }
-    }
-    process(input, options) {
-        return this.processInput(input, options);
-    }
-    // References: https://ai.google.dev/gemini-api/docs/thinking#set-budget
-    thinkingBudgetModelMap = [
-        // 注意：gemini-2.5-flash-image-preview 模型并不支持 thinking。see: https://github.com/CherryHQ/cherry-studio/issues/9614
-        {
-            pattern: /gemini-2.5-flash-image-preview/,
-            support: false,
-        },
-        {
-            pattern: /gemini-3(?!.*-image-)/,
-            support: true,
-            type: "level",
-        },
-        {
-            pattern: /gemini-2.5-pro/,
-            support: true,
-            min: 128,
-            max: 32768,
-        },
-        {
-            pattern: /gemini-2.5-flash/,
-            support: true,
-            min: 0,
-            max: 24576,
-        },
-        {
-            pattern: /2.5-flash-lite/,
-            support: true,
-            min: 512,
-            max: 24576,
-        },
-        {
-            pattern: /.*/,
-            support: false,
-        },
-    ];
-    thinkingBudgetLevelMap = {
-        high: 100000, // use 100k for high, finally capped by model max
-        medium: 10000,
-        low: 5000,
-        minimal: 200,
-    };
-    thinkingLevelMap = {
-        high: genai_1.ThinkingLevel.HIGH,
-        medium: genai_1.ThinkingLevel.HIGH,
-        low: genai_1.ThinkingLevel.LOW,
-        minimal: genai_1.ThinkingLevel.LOW,
-    };
-    getThinkingBudget(model, effort) {
-        const m = this.thinkingBudgetModelMap.find((i) => i.pattern.test(model));
-        if (!m?.support)
-            return { support: false };
-        if (m.type === "level") {
-            let level = genai_1.ThinkingLevel.THINKING_LEVEL_UNSPECIFIED;
-            if (typeof effort === "string") {
-                level = this.thinkingLevelMap[effort];
-            }
-            else if (typeof effort === "number") {
-                level =
-                    effort >= this.thinkingBudgetLevelMap["medium"] ? genai_1.ThinkingLevel.HIGH : genai_1.ThinkingLevel.LOW;
-            }
-            return { support: true, level };
-        }
-        let budget = typeof effort === "string" ? this.thinkingBudgetLevelMap[effort] || undefined : effort;
-        if (typeof budget === "undefined")
-            return { support: true };
-        if (typeof m.min === "number")
-            budget = Math.max(m.min, budget);
-        if (typeof m.max === "number")
-            budget = Math.min(m.max, budget);
-        return { support: true, budget };
-    }
-    async getParameters(input) {
-        const { modelOptions = {} } = input;
-        const model = modelOptions.model || this.credential.model;
-        const { contents, config } = await this.buildContents(input);
-        const thinkingBudget = this.getThinkingBudget(model, modelOptions.reasoningEffort);
-        const parameters = {
-            model,
-            contents,
-            config: {
-                thinkingConfig: thinkingBudget.support
-                    ? {
-                        includeThoughts: true,
-                        thinkingBudget: thinkingBudget.budget,
-                        thinkingLevel: thinkingBudget.level,
-                    }
-                    : undefined,
-                responseModalities: modelOptions.modalities,
-                temperature: modelOptions.temperature,
-                topP: modelOptions.topP,
-                frequencyPenalty: modelOptions.frequencyPenalty,
-                presencePenalty: modelOptions.presencePenalty,
-                ...config,
-                ...(await this.buildConfig(input)),
-            },
-        };
-        return parameters;
-    }
-    async *processInput(input, options) {
-        const parameters = await this.getParameters(input);
-        const response = await this.googleClient.models.generateContentStream(parameters);
-        let usage = {
-            inputTokens: 0,
-            outputTokens: 0,
-        };
-        let responseModel;
-        const files = [];
-        const toolCalls = [];
-        let text = "";
-        let json;
-        for await (const chunk of response) {
-            if (!responseModel && chunk.modelVersion) {
-                responseModel = chunk.modelVersion;
-                yield { delta: { json: { model: responseModel } } };
-            }
-            for (const { content } of chunk.candidates ?? []) {
-                if (content?.parts) {
-                    for (const part of content.parts) {
-                        if (part.text) {
-                            if (part.thought) {
-                                yield { delta: { text: { thoughts: part.text } } };
-                            }
-                            else {
-                                text += part.text;
-                                if (input.responseFormat?.type !== "json_schema") {
-                                    yield { delta: { text: { text: part.text } } };
-                                }
-                            }
-                        }
-                        if (part.inlineData?.data) {
-                            files.push({
-                                type: "file",
-                                data: part.inlineData.data,
-                                filename: part.inlineData.displayName,
-                                mimeType: part.inlineData.mimeType,
-                            });
-                        }
-                        if (part.functionCall?.name) {
-                            if (part.functionCall.name === OUTPUT_FUNCTION_NAME) {
-                                json = part.functionCall.args;
-                            }
-                            else {
-                                const toolCall = {
-                                    id: part.functionCall.id || (0, uuid_1.v7)(),
-                                    type: "function",
-                                    function: {
-                                        name: part.functionCall.name,
-                                        arguments: part.functionCall.args || {},
-                                    },
-                                };
-                                // Preserve thought_signature for 3.x models
-                                if (part.thoughtSignature && parameters.model.includes("gemini-3")) {
-                                    toolCall.metadata = {
-                                        thoughtSignature: part.thoughtSignature,
-                                    };
-                                }
-                                toolCalls.push(toolCall);
-                            }
-                        }
-                    }
-                }
-            }
-            if (chunk.usageMetadata) {
-                if (chunk.usageMetadata.promptTokenCount)
-                    usage.inputTokens = chunk.usageMetadata.promptTokenCount;
-                if (chunk.usageMetadata.candidatesTokenCount || chunk.usageMetadata.thoughtsTokenCount)
-                    usage.outputTokens =
-                        (chunk.usageMetadata.candidatesTokenCount || 0) +
-                            (chunk.usageMetadata.thoughtsTokenCount || 0);
-                // Parse cache statistics if available
-                if (chunk.usageMetadata.cachedContentTokenCount) {
-                    usage.cacheReadInputTokens = chunk.usageMetadata.cachedContentTokenCount;
-                }
-            }
-        }
-        if (toolCalls.length) {
-            yield { delta: { json: { toolCalls } } };
-        }
-        if (input.responseFormat?.type === "json_schema") {
-            if (json) {
-                yield { delta: { json: { json } } };
-            }
-            else if (text) {
-                yield { delta: { json: { json: (0, core_1.safeParseJSON)(text) } } };
-            }
-            else if (!toolCalls.length) {
-                throw new core_1.StructuredOutputError("No JSON response from the model");
-            }
-        }
-        else if (!toolCalls.length) {
-            // NOTE: gemini-2.5-pro sometimes returns an empty response,
-            // so we check here and retry with structured output mode (empty responses occur less frequently with tool calls)
-            if (!text && !files.length) {
-                logger_js_1.logger.warn("Empty response from Gemini, retrying with structured output mode");
-                try {
-                    const outputSchema = zod_1.z.object({
-                        output: zod_1.z.string().describe("The final answer from the model"),
-                    });
-                    const response = await this.process({
-                        ...input,
-                        responseFormat: {
-                            type: "json_schema",
-                            jsonSchema: {
-                                name: "output",
-                                schema: (0, zod_to_json_schema_1.zodToJsonSchema)(outputSchema),
-                            },
-                        },
-                    }, options);
-                    const result = await (0, core_1.agentProcessResultToObject)(response);
-                    // Merge retry usage with the original usage
-                    usage = (0, model_utils_js_1.mergeUsage)(usage, result.usage);
-                    // Return the tool calls if retry has tool calls
-                    if (result.toolCalls?.length) {
-                        toolCalls.push(...result.toolCalls);
-                        yield { delta: { json: { toolCalls } } };
-                    }
-                    // Return the text from structured output of retry
-                    else {
-                        if (!result.json)
-                            throw new Error("Retrying with structured output mode got no json response");
-                        const parsed = outputSchema.safeParse(result.json);
-                        if (!parsed.success)
-                            throw new Error("Retrying with structured output mode got invalid json response");
-                        text = parsed.data.output;
-                        yield { delta: { text: { text } } };
-                        logger_js_1.logger.warn("Empty response from Gemini, retried with structured output mode successfully");
-                    }
-                }
-                catch (error) {
-                    logger_js_1.logger.error("Empty response from Gemini, retrying with structured output mode failed", error);
-                    throw new core_1.StructuredOutputError("No response from the model");
-                }
-            }
-        }
-        yield {
-            delta: {
-                json: {
-                    usage,
-                    files: files.length ? files : undefined,
-                    modelOptions: {
-                        reasoningEffort: parameters.config?.thinkingConfig?.thinkingLevel ||
-                            parameters.config?.thinkingConfig?.thinkingBudget,
-                    },
-                },
-            },
-        };
-    }
-    async buildConfig(input) {
-        const config = {};
-        const { tools, toolConfig } = await this.buildTools(input);
-        config.tools = tools;
-        config.toolConfig = toolConfig;
-        if (input.responseFormat?.type === "json_schema") {
-            if (config.tools?.length) {
-                config.tools.push({
-                    functionDeclarations: [
-                        {
-                            name: OUTPUT_FUNCTION_NAME,
-                            description: "Output the final response",
-                            parametersJsonSchema: input.responseFormat.jsonSchema.schema,
-                        },
-                    ],
-                });
-                config.toolConfig = {
-                    ...config.toolConfig,
-                    functionCallingConfig: { mode: genai_1.FunctionCallingConfigMode.ANY },
-                };
-            }
-            else {
-                config.responseJsonSchema = input.responseFormat.jsonSchema.schema;
-                config.responseMimeType = "application/json";
-            }
-        }
-        return config;
-    }
-    async buildTools(input) {
-        const tools = [];
-        for (const tool of input.tools ?? []) {
-            tools.push({
-                functionDeclarations: [
-                    {
-                        name: tool.function.name,
-                        description: tool.function.description,
-                        parametersJsonSchema: tool.function.parameters,
-                    },
-                ],
-            });
-        }
-        const functionCallingConfig = !input.toolChoice
-            ? undefined
-            : input.toolChoice === "auto"
-                ? { mode: genai_1.FunctionCallingConfigMode.AUTO }
-                : input.toolChoice === "none"
-                    ? { mode: genai_1.FunctionCallingConfigMode.NONE }
-                    : input.toolChoice === "required"
-                        ? { mode: genai_1.FunctionCallingConfigMode.ANY }
-                        : {
-                            mode: genai_1.FunctionCallingConfigMode.ANY,
-                            allowedFunctionNames: [input.toolChoice.function.name],
-                        };
-        return { tools, toolConfig: { functionCallingConfig } };
-    }
-    async buildVideoContentParts(media) {
-        const { path: filePath, mimeType: fileMimeType } = await this.transformFileType("local", media);
-        if (filePath) {
-            const stats = await index_js_1.nodejs.fs.stat(filePath);
-            const fileSizeInBytes = stats.size;
-            const fileSizeMB = fileSizeInBytes / (1024 * 1024);
-            if (fileSizeMB > NEED_UPLOAD_MAX_FILE_SIZE_MB) {
-                const uploadedFile = await this.googleClient.files.upload({
-                    file: filePath,
-                    config: { mimeType: fileMimeType },
-                });
-                let file = uploadedFile;
-                while (file.state === "PROCESSING") {
-                    await new Promise((resolve) => setTimeout(resolve, 1000));
-                    if (file.name) {
-                        file = await this.googleClient.files.get({ name: file.name });
-                    }
-                }
-                if (file.state !== "ACTIVE") {
-                    throw new Error(`File ${file.name} failed to process: ${file.state}`);
-                }
-                if (file.uri && file.mimeType) {
-                    const result = (0, genai_1.createUserContent)([(0, genai_1.createPartFromUri)(file.uri, file.mimeType), ""]);
-                    const part = result.parts?.find((x) => x.fileData);
-                    if (part) {
-                        await index_js_1.nodejs.fs.rm(filePath);
-                        return part;
-                    }
-                }
-            }
-        }
-    }
-    async buildContents(input) {
-        const result = {
-            contents: [],
-        };
-        const systemParts = [];
-        result.contents = (await Promise.all(input.messages.map(async (msg) => {
-            if (msg.role === "system") {
-                if (typeof msg.content === "string") {
-                    systemParts.push({ text: msg.content });
-                }
-                else if (Array.isArray(msg.content)) {
-                    systemParts.push(...msg.content.map((item) => {
-                        if (item.type === "text")
-                            return { text: item.text };
-                        throw new Error(`Unsupported content type: ${item.type}`);
-                    }));
-                }
-                return;
-            }
-            const content = {
-                role: msg.role === "agent" ? "model" : msg.role === "user" ? "user" : undefined,
-            };
-            if (msg.toolCalls) {
-                content.parts = msg.toolCalls.map((call) => {
-                    const part = {
-                        functionCall: {
-                            id: call.id,
-                            name: call.function.name,
-                            args: call.function.arguments,
-                        },
-                    };
-                    // Restore thought_signature for 3.x models
-                    if (call.metadata?.thoughtSignature) {
-                        part.thoughtSignature = call.metadata.thoughtSignature;
-                    }
-                    return part;
-                });
-            }
-            else if (msg.toolCallId) {
-                const call = input.messages
-                    .flatMap((i) => i.toolCalls)
-                    .find((c) => c?.id === msg.toolCallId);
-                if (!call)
-                    throw new Error(`Tool call not found: ${msg.toolCallId}`);
-                if (!msg.content)
-                    throw new Error("Tool call must have content");
-                // parse tool result as a record
-                let toolResult;
-                {
-                    let text;
-                    if (typeof msg.content === "string")
-                        text = msg.content;
-                    else if (msg.content?.length === 1) {
-                        const first = msg.content[0];
-                        if (first?.type === "text")
-                            text = first.text;
-                    }
-                    if (text) {
-                        try {
-                            const obj = (0, yaml_1.parse)(text);
-                            if ((0, type_utils_js_1.isRecord)(obj))
-                                toolResult = obj;
-                        }
-                        catch {
-                            // ignore
-                        }
-                        if (!toolResult)
-                            toolResult = { result: text };
-                    }
-                }
-                const functionResponse = {
-                    id: msg.toolCallId,
-                    name: call.function.name,
-                };
-                if (toolResult) {
-                    functionResponse.response = toolResult;
-                }
-                else {
-                    functionResponse.parts = await this.contentToParts(msg.content);
-                }
-                content.parts = [{ functionResponse }];
-            }
-            else if (msg.content) {
-                content.parts = await this.contentToParts(msg.content);
-            }
-            return content;
-        }))).filter(type_utils_js_1.isNonNullable);
-        this.ensureMessagesHasUserMessage(systemParts, result.contents);
-        if (systemParts.length) {
-            result.config ??= {};
-            result.config.systemInstruction = systemParts;
-        }
-        return result;
-    }
-    async contentToParts(content) {
-        if (typeof content === "string")
-            return [{ text: content }];
-        return Promise.all(content.map(async (item) => {
-            switch (item.type) {
-                case "text":
-                    return { text: item.text };
-                case "url":
-                    return { fileData: { fileUri: item.url, mimeType: item.mimeType } };
-                case "file": {
-                    const part = await this.buildVideoContentParts(item);
-                    if (part)
-                        return part;
-                    return { inlineData: { data: item.data, mimeType: item.mimeType } };
-                }
-                case "local":
-                    throw new Error(`Unsupported local file: ${item.path}, it should be converted to base64 at ChatModel`);
-            }
-        }));
-    }
-    ensureMessagesHasUserMessage(systems, contents) {
-        // no messages but system messages
-        if (!contents.length && systems.length) {
-            const system = systems.pop();
-            if (system)
-                contents.push({ role: "user", parts: [system] });
-        }
-        // first message is from model
-        if (contents[0]?.role === "model") {
-            const system = systems.pop();
-            if (system)
-                contents.unshift({ role: "user", parts: [system] });
-        }
-    }
-}
-exports.GeminiChatModel = GeminiChatModel;