@aigne/openai 0.11.3 → 0.12.0

package/CHANGELOG.md

@@ -1,5 +1,43 @@
 # Changelog
 
+## [0.12.0](https://github.com/AIGNE-io/aigne-framework/compare/openai-v0.11.5...openai-v0.12.0) (2025-08-20)
+
+
+### Features
+
+* add ImageModel/ImageAgent support ([#383](https://github.com/AIGNE-io/aigne-framework/issues/383)) ([96a2093](https://github.com/AIGNE-io/aigne-framework/commit/96a209368d91d98f47db6de1e404640368a86fa8))
+
+
+### Dependencies
+
+* The following workspace dependencies were updated
+  * dependencies
+    * @aigne/core bumped to 1.53.0
+  * devDependencies
+    * @aigne/test-utils bumped to 0.5.32
+
+## [0.11.5](https://github.com/AIGNE-io/aigne-framework/compare/openai-v0.11.4...openai-v0.11.5) (2025-08-20)
+
+
+### Dependencies
+
+* The following workspace dependencies were updated
+  * dependencies
+    * @aigne/core bumped to 1.52.0
+  * devDependencies
+    * @aigne/test-utils bumped to 0.5.31
+
+## [0.11.4](https://github.com/AIGNE-io/aigne-framework/compare/openai-v0.11.3...openai-v0.11.4) (2025-08-18)
+
+
+### Dependencies
+
+* The following workspace dependencies were updated
+  * dependencies
+    * @aigne/core bumped to 1.51.0
+  * devDependencies
+    * @aigne/test-utils bumped to 0.5.30
+
 ## [0.11.3](https://github.com/AIGNE-io/aigne-framework/compare/openai-v0.11.2...openai-v0.11.3) (2025-08-16)
 
 

package/lib/cjs/index.d.ts

@@ -1 +1,2 @@
 export * from "./openai-chat-model.js";
+export * from "./openai-image-model.js";

package/lib/cjs/index.js

@@ -15,3 +15,4 @@ var __exportStar = (this && this.__exportStar) || function(m, exports) {
 };
 Object.defineProperty(exports, "__esModule", { value: true });
 __exportStar(require("./openai-chat-model.js"), exports);
+__exportStar(require("./openai-image-model.js"), exports);

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

@@ -1,6 +1,6 @@
 import { type AgentProcessResult, ChatModel, type ChatModelInput, type ChatModelInputMessage, type ChatModelInputTool, type ChatModelOptions, type ChatModelOutput } from "@aigne/core";
 import { type PromiseOrValue } from "@aigne/core/utils/type-utils.js";
-import OpenAI, { type ClientOptions } from "openai";
+import type { ClientOptions, OpenAI } from "openai";
 import type { ChatCompletionMessageParam, ChatCompletionTool } from "openai/resources";
 import { z } from "zod";
 export interface OpenAIChatModelCapabilities {
@@ -131,12 +131,12 @@ export declare class OpenAIChatModel extends ChatModel {
     protected supportsToolsEmptyParameters: boolean;
     protected supportsToolStreaming: boolean;
     protected supportsTemperature: boolean;
-    client(): Promise<OpenAI>;
-    getCredential(): Promise<{
+    get client(): OpenAI;
+    get credential(): {
         url: string | undefined;
         apiKey: string | undefined;
         model: string;
-    }>;
+    };
     get modelOptions(): ChatModelOptions | undefined;
     /**
      * Process the input and generate a response

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

@@ -1,7 +1,4 @@
 "use strict";
-var __importDefault = (this && this.__importDefault) || function (mod) {
-    return (mod && mod.__esModule) ? mod : { "default": mod };
-};
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.OpenAIChatModel = exports.openAIChatModelOptionsSchema = void 0;
 exports.contentsFromInputMessages = contentsFromInputMessages;
@@ -14,9 +11,9 @@ const prompts_js_1 = require("@aigne/core/utils/prompts.js");
 const stream_utils_js_1 = require("@aigne/core/utils/stream-utils.js");
 const type_utils_js_1 = require("@aigne/core/utils/type-utils.js");
 const ajv_1 = require("ajv");
-const openai_1 = __importDefault(require("openai"));
 const uuid_1 = require("uuid");
 const zod_1 = require("zod");
+const openai_js_1 = require("./openai.js");
 const CHAT_MODEL_OPENAI_DEFAULT_MODEL = "gpt-4o-mini";
 const OPENAI_CHAT_MODEL_CAPABILITIES = {
     "o4-mini": { supportsParallelToolCalls: false, supportsTemperature: false },
@@ -81,18 +78,18 @@ class OpenAIChatModel extends core_1.ChatModel {
     supportsToolsEmptyParameters = true;
     supportsToolStreaming = true;
     supportsTemperature = true;
-    async client() {
-        const { apiKey, url } = await this.getCredential();
+    get client() {
+        const { apiKey, url } = 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._client ??= new CustomOpenAI({
+        this._client ??= new openai_js_1.CustomOpenAI({
             baseURL: url,
             apiKey,
             ...this.options?.clientOptions,
         });
         return this._client;
     }
-    async getCredential() {
+    get credential() {
         return {
             url: this.options?.baseURL || process.env.OPENAI_BASE_URL,
             apiKey: this.options?.apiKey || process.env[this.apiKeyEnvName] || this.apiKeyDefault,
@@ -113,7 +110,7 @@ class OpenAIChatModel extends core_1.ChatModel {
     ajv = new ajv_1.Ajv();
     async _process(input) {
         const messages = await this.getRunMessages(input);
-        const { model } = await this.getCredential();
+        const model = input.modelOptions?.model || this.credential.model;
         const body = {
             model,
             temperature: this.supportsTemperature
@@ -134,8 +131,7 @@ class OpenAIChatModel extends core_1.ChatModel {
             return await this.requestStructuredOutput(body, input.responseFormat);
         }
         const { jsonMode, responseFormat } = await this.getRunResponseFormat(input);
-        const client = await this.client();
-        const stream = (await client.chat.completions.create({
+        const stream = (await this.client.chat.completions.create({
             ...body,
             tools: toolsFromInputTools(input.tools, {
                 addTypeToEmptyParameters: !this.supportsToolsEmptyParameters,
@@ -214,8 +210,7 @@ class OpenAIChatModel extends core_1.ChatModel {
         const { jsonMode, responseFormat: resolvedResponseFormat } = await this.getRunResponseFormat({
             responseFormat,
         });
-        const client = await this.client();
-        const res = (await client.chat.completions.create({
+        const res = (await this.client.chat.completions.create({
             ...body,
             response_format: resolvedResponseFormat,
         }));
@@ -428,13 +423,4 @@ function handleCompleteToolCall(toolCalls, call) {
         args: call.function?.arguments || "",
     });
 }
-// Use a custom OpenAI client to handle API errors for better error messages
-class CustomOpenAI extends openai_1.default {
-    makeStatusError(status, error, message, headers) {
-        if (!("error" in error) || typeof error.error !== "string") {
-            message = JSON.stringify(error);
-        }
-        return super.makeStatusError(status, error, message, headers);
-    }
-}
 // safeParseJSON is now imported from @aigne/core

package/lib/cjs/openai-image-model.d.ts

@@ -0,0 +1,55 @@
+import { ImageModel, type ImageModelInput, type ImageModelOptions, type ImageModelOutput } from "@aigne/core";
+import { type Camelize } from "@aigne/core/utils/camelize.js";
+import type OpenAI from "openai";
+import type { ClientOptions } from "openai";
+export interface OpenAIImageModelInput extends ImageModelInput, Camelize<Omit<OpenAI.ImageGenerateParams, "prompt" | "model" | "n" | "response_format">> {
+}
+export interface OpenAIImageModelOutput extends ImageModelOutput {
+}
+export interface OpenAIImageModelOptions extends ImageModelOptions<OpenAIImageModelInput, OpenAIImageModelOutput> {
+    /**
+     * API key for OpenAI API
+     *
+     * If not provided, will look for OPENAI_API_KEY in environment variables
+     */
+    apiKey?: string;
+    /**
+     * Base URL for OpenAI API
+     *
+     * Useful for proxies or alternate endpoints
+     */
+    baseURL?: string;
+    /**
+     * OpenAI model to use
+     *
+     * Defaults to 'dall-e-2'
+     */
+    model?: string;
+    /**
+     * Additional model options to control behavior
+     */
+    modelOptions?: Omit<Partial<OpenAIImageModelInput>, "model">;
+    /**
+     * Client options for OpenAI API
+     */
+    clientOptions?: Partial<ClientOptions>;
+}
+export declare class OpenAIImageModel extends ImageModel<OpenAIImageModelInput, OpenAIImageModelOutput> {
+    options?: OpenAIImageModelOptions | undefined;
+    constructor(options?: OpenAIImageModelOptions | undefined);
+    protected _client?: OpenAI;
+    protected apiKeyEnvName: string;
+    get client(): OpenAI;
+    get credential(): {
+        url: string | undefined;
+        apiKey: string | undefined;
+        model: string;
+    };
+    get modelOptions(): Omit<Partial<OpenAIImageModelInput>, "model"> | undefined;
+    /**
+     * Process the input and generate a response
+     * @param input The input to process
+     * @returns The generated response
+     */
+    process(input: ImageModelInput): Promise<ImageModelOutput>;
+}

package/lib/cjs/openai-image-model.js

@@ -0,0 +1,97 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.OpenAIImageModel = void 0;
+const core_1 = require("@aigne/core");
+const camelize_js_1 = require("@aigne/core/utils/camelize.js");
+const type_utils_js_1 = require("@aigne/core/utils/type-utils.js");
+const zod_1 = require("zod");
+const openai_js_1 = require("./openai.js");
+const DEFAULT_MODEL = "dall-e-2";
+const openAIImageModelInputSchema = core_1.imageModelInputSchema.extend({});
+const openAIImageModelOptionsSchema = zod_1.z.object({
+    apiKey: zod_1.z.string().optional(),
+    baseURL: zod_1.z.string().optional(),
+    model: zod_1.z.string().optional(),
+});
+class OpenAIImageModel extends core_1.ImageModel {
+    options;
+    constructor(options) {
+        super({
+            ...options,
+            inputSchema: openAIImageModelInputSchema,
+            description: options?.description ?? "Draw or edit image by OpenAI image models",
+        });
+        this.options = options;
+        if (options)
+            (0, type_utils_js_1.checkArguments)(this.name, openAIImageModelOptionsSchema, options);
+    }
+    _client;
+    apiKeyEnvName = "OPENAI_API_KEY";
+    get client() {
+        if (this._client)
+            return this._client;
+        const { apiKey, url } = 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._client ??= new openai_js_1.CustomOpenAI({
+            baseURL: url,
+            apiKey,
+            ...this.options?.clientOptions,
+        });
+        return this._client;
+    }
+    get credential() {
+        return {
+            url: this.options?.baseURL || process.env.OPENAI_BASE_URL,
+            apiKey: this.options?.apiKey || process.env[this.apiKeyEnvName],
+            model: this.options?.model || DEFAULT_MODEL,
+        };
+    }
+    get modelOptions() {
+        return this.options?.modelOptions;
+    }
+    /**
+     * Process the input and generate a response
+     * @param input The input to process
+     * @returns The generated response
+     */
+    async process(input) {
+        const model = input.model || this.credential.model;
+        const inputKeys = [
+            "background",
+            "moderation",
+            "output_compression",
+            "output_format",
+            "prompt",
+            "quality",
+            "size",
+            "style",
+            "user",
+        ];
+        let responseFormat;
+        if (model !== "gpt-image-1") {
+            responseFormat = input.responseFormat === "base64" ? "b64_json" : "url";
+        }
+        const body = {
+            ...(0, camelize_js_1.snakelize)((0, type_utils_js_1.pick)({ ...this.modelOptions, ...input }, inputKeys)),
+            response_format: responseFormat,
+            model,
+        };
+        const response = await this.client.images.generate({ ...body });
+        return {
+            images: (response.data ?? []).map((image) => {
+                if (image.url)
+                    return { url: image.url };
+                if (image.b64_json)
+                    return { base64: image.b64_json };
+                throw new Error("Image response does not contain a valid URL or base64 data");
+            }),
+            usage: {
+                inputTokens: response.usage?.input_tokens || 0,
+                outputTokens: response.usage?.output_tokens || 0,
+            },
+            model,
+        };
+    }
+}
+exports.OpenAIImageModel = OpenAIImageModel;

package/lib/cjs/openai.d.ts

@@ -0,0 +1,4 @@
+import OpenAI, { type APIError } from "openai";
+export declare class CustomOpenAI extends OpenAI {
+    protected makeStatusError(status: number, error: object, message: string | undefined, headers: Headers): APIError;
+}

package/lib/cjs/openai.js

@@ -0,0 +1,17 @@
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.CustomOpenAI = void 0;
+const openai_1 = __importDefault(require("openai"));
+// Use a custom OpenAI client to handle API errors for better error messages
+class CustomOpenAI extends openai_1.default {
+    makeStatusError(status, error, message, headers) {
+        if (!("error" in error) || typeof error.error !== "string") {
+            message = JSON.stringify(error);
+        }
+        return super.makeStatusError(status, error, message, headers);
+    }
+}
+exports.CustomOpenAI = CustomOpenAI;

package/lib/dts/index.d.ts

@@ -1 +1,2 @@
 export * from "./openai-chat-model.js";
+export * from "./openai-image-model.js";

package/lib/dts/openai-chat-model.d.ts

@@ -1,6 +1,6 @@
 import { type AgentProcessResult, ChatModel, type ChatModelInput, type ChatModelInputMessage, type ChatModelInputTool, type ChatModelOptions, type ChatModelOutput } from "@aigne/core";
 import { type PromiseOrValue } from "@aigne/core/utils/type-utils.js";
-import OpenAI, { type ClientOptions } from "openai";
+import type { ClientOptions, OpenAI } from "openai";
 import type { ChatCompletionMessageParam, ChatCompletionTool } from "openai/resources";
 import { z } from "zod";
 export interface OpenAIChatModelCapabilities {
@@ -131,12 +131,12 @@ export declare class OpenAIChatModel extends ChatModel {
     protected supportsToolsEmptyParameters: boolean;
     protected supportsToolStreaming: boolean;
     protected supportsTemperature: boolean;
-    client(): Promise<OpenAI>;
-    getCredential(): Promise<{
+    get client(): OpenAI;
+    get credential(): {
         url: string | undefined;
         apiKey: string | undefined;
         model: string;
-    }>;
+    };
     get modelOptions(): ChatModelOptions | undefined;
     /**
      * Process the input and generate a response

package/lib/dts/openai-image-model.d.ts

@@ -0,0 +1,55 @@
+import { ImageModel, type ImageModelInput, type ImageModelOptions, type ImageModelOutput } from "@aigne/core";
+import { type Camelize } from "@aigne/core/utils/camelize.js";
+import type OpenAI from "openai";
+import type { ClientOptions } from "openai";
+export interface OpenAIImageModelInput extends ImageModelInput, Camelize<Omit<OpenAI.ImageGenerateParams, "prompt" | "model" | "n" | "response_format">> {
+}
+export interface OpenAIImageModelOutput extends ImageModelOutput {
+}
+export interface OpenAIImageModelOptions extends ImageModelOptions<OpenAIImageModelInput, OpenAIImageModelOutput> {
+    /**
+     * API key for OpenAI API
+     *
+     * If not provided, will look for OPENAI_API_KEY in environment variables
+     */
+    apiKey?: string;
+    /**
+     * Base URL for OpenAI API
+     *
+     * Useful for proxies or alternate endpoints
+     */
+    baseURL?: string;
+    /**
+     * OpenAI model to use
+     *
+     * Defaults to 'dall-e-2'
+     */
+    model?: string;
+    /**
+     * Additional model options to control behavior
+     */
+    modelOptions?: Omit<Partial<OpenAIImageModelInput>, "model">;
+    /**
+     * Client options for OpenAI API
+     */
+    clientOptions?: Partial<ClientOptions>;
+}
+export declare class OpenAIImageModel extends ImageModel<OpenAIImageModelInput, OpenAIImageModelOutput> {
+    options?: OpenAIImageModelOptions | undefined;
+    constructor(options?: OpenAIImageModelOptions | undefined);
+    protected _client?: OpenAI;
+    protected apiKeyEnvName: string;
+    get client(): OpenAI;
+    get credential(): {
+        url: string | undefined;
+        apiKey: string | undefined;
+        model: string;
+    };
+    get modelOptions(): Omit<Partial<OpenAIImageModelInput>, "model"> | undefined;
+    /**
+     * Process the input and generate a response
+     * @param input The input to process
+     * @returns The generated response
+     */
+    process(input: ImageModelInput): Promise<ImageModelOutput>;
+}

package/lib/dts/openai.d.ts

@@ -0,0 +1,4 @@
+import OpenAI, { type APIError } from "openai";
+export declare class CustomOpenAI extends OpenAI {
+    protected makeStatusError(status: number, error: object, message: string | undefined, headers: Headers): APIError;
+}

package/lib/esm/index.d.ts

@@ -1 +1,2 @@
 export * from "./openai-chat-model.js";
+export * from "./openai-image-model.js";

package/lib/esm/index.js

@@ -1 +1,2 @@
 export * from "./openai-chat-model.js";
+export * from "./openai-image-model.js";

package/lib/esm/openai-chat-model.d.ts

@@ -1,6 +1,6 @@
 import { type AgentProcessResult, ChatModel, type ChatModelInput, type ChatModelInputMessage, type ChatModelInputTool, type ChatModelOptions, type ChatModelOutput } from "@aigne/core";
 import { type PromiseOrValue } from "@aigne/core/utils/type-utils.js";
-import OpenAI, { type ClientOptions } from "openai";
+import type { ClientOptions, OpenAI } from "openai";
 import type { ChatCompletionMessageParam, ChatCompletionTool } from "openai/resources";
 import { z } from "zod";
 export interface OpenAIChatModelCapabilities {
@@ -131,12 +131,12 @@ export declare class OpenAIChatModel extends ChatModel {
     protected supportsToolsEmptyParameters: boolean;
     protected supportsToolStreaming: boolean;
     protected supportsTemperature: boolean;
-    client(): Promise<OpenAI>;
-    getCredential(): Promise<{
+    get client(): OpenAI;
+    get credential(): {
         url: string | undefined;
         apiKey: string | undefined;
         model: string;
-    }>;
+    };
     get modelOptions(): ChatModelOptions | undefined;
     /**
      * Process the input and generate a response

package/lib/esm/openai-chat-model.js

@@ -5,9 +5,9 @@ import { getJsonOutputPrompt } from "@aigne/core/utils/prompts.js";
 import { agentResponseStreamToObject } from "@aigne/core/utils/stream-utils.js";
 import { checkArguments, isNonNullable, } from "@aigne/core/utils/type-utils.js";
 import { Ajv } from "ajv";
-import OpenAI from "openai";
 import { v7 } from "uuid";
 import { z } from "zod";
+import { CustomOpenAI } from "./openai.js";
 const CHAT_MODEL_OPENAI_DEFAULT_MODEL = "gpt-4o-mini";
 const OPENAI_CHAT_MODEL_CAPABILITIES = {
     "o4-mini": { supportsParallelToolCalls: false, supportsTemperature: false },
@@ -72,8 +72,8 @@ export class OpenAIChatModel extends ChatModel {
     supportsToolsEmptyParameters = true;
     supportsToolStreaming = true;
     supportsTemperature = true;
-    async client() {
-        const { apiKey, url } = await this.getCredential();
+    get client() {
+        const { apiKey, url } = 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._client ??= new CustomOpenAI({
@@ -83,7 +83,7 @@ export class OpenAIChatModel extends ChatModel {
         });
         return this._client;
     }
-    async getCredential() {
+    get credential() {
         return {
             url: this.options?.baseURL || process.env.OPENAI_BASE_URL,
             apiKey: this.options?.apiKey || process.env[this.apiKeyEnvName] || this.apiKeyDefault,
@@ -104,7 +104,7 @@ export class OpenAIChatModel extends ChatModel {
     ajv = new Ajv();
     async _process(input) {
         const messages = await this.getRunMessages(input);
-        const { model } = await this.getCredential();
+        const model = input.modelOptions?.model || this.credential.model;
         const body = {
             model,
             temperature: this.supportsTemperature
@@ -125,8 +125,7 @@ export class OpenAIChatModel extends ChatModel {
             return await this.requestStructuredOutput(body, input.responseFormat);
         }
         const { jsonMode, responseFormat } = await this.getRunResponseFormat(input);
-        const client = await this.client();
-        const stream = (await client.chat.completions.create({
+        const stream = (await this.client.chat.completions.create({
             ...body,
             tools: toolsFromInputTools(input.tools, {
                 addTypeToEmptyParameters: !this.supportsToolsEmptyParameters,
@@ -205,8 +204,7 @@ export class OpenAIChatModel extends ChatModel {
         const { jsonMode, responseFormat: resolvedResponseFormat } = await this.getRunResponseFormat({
             responseFormat,
         });
-        const client = await this.client();
-        const res = (await client.chat.completions.create({
+        const res = (await this.client.chat.completions.create({
             ...body,
             response_format: resolvedResponseFormat,
         }));
@@ -418,13 +416,4 @@ function handleCompleteToolCall(toolCalls, call) {
         args: call.function?.arguments || "",
     });
 }
-// Use a custom OpenAI client to handle API errors for better error messages
-class CustomOpenAI extends OpenAI {
-    makeStatusError(status, error, message, headers) {
-        if (!("error" in error) || typeof error.error !== "string") {
-            message = JSON.stringify(error);
-        }
-        return super.makeStatusError(status, error, message, headers);
-    }
-}
 // safeParseJSON is now imported from @aigne/core

package/lib/esm/openai-image-model.d.ts

@@ -0,0 +1,55 @@
+import { ImageModel, type ImageModelInput, type ImageModelOptions, type ImageModelOutput } from "@aigne/core";
+import { type Camelize } from "@aigne/core/utils/camelize.js";
+import type OpenAI from "openai";
+import type { ClientOptions } from "openai";
+export interface OpenAIImageModelInput extends ImageModelInput, Camelize<Omit<OpenAI.ImageGenerateParams, "prompt" | "model" | "n" | "response_format">> {
+}
+export interface OpenAIImageModelOutput extends ImageModelOutput {
+}
+export interface OpenAIImageModelOptions extends ImageModelOptions<OpenAIImageModelInput, OpenAIImageModelOutput> {
+    /**
+     * API key for OpenAI API
+     *
+     * If not provided, will look for OPENAI_API_KEY in environment variables
+     */
+    apiKey?: string;
+    /**
+     * Base URL for OpenAI API
+     *
+     * Useful for proxies or alternate endpoints
+     */
+    baseURL?: string;
+    /**
+     * OpenAI model to use
+     *
+     * Defaults to 'dall-e-2'
+     */
+    model?: string;
+    /**
+     * Additional model options to control behavior
+     */
+    modelOptions?: Omit<Partial<OpenAIImageModelInput>, "model">;
+    /**
+     * Client options for OpenAI API
+     */
+    clientOptions?: Partial<ClientOptions>;
+}
+export declare class OpenAIImageModel extends ImageModel<OpenAIImageModelInput, OpenAIImageModelOutput> {
+    options?: OpenAIImageModelOptions | undefined;
+    constructor(options?: OpenAIImageModelOptions | undefined);
+    protected _client?: OpenAI;
+    protected apiKeyEnvName: string;
+    get client(): OpenAI;
+    get credential(): {
+        url: string | undefined;
+        apiKey: string | undefined;
+        model: string;
+    };
+    get modelOptions(): Omit<Partial<OpenAIImageModelInput>, "model"> | undefined;
+    /**
+     * Process the input and generate a response
+     * @param input The input to process
+     * @returns The generated response
+     */
+    process(input: ImageModelInput): Promise<ImageModelOutput>;
+}

package/lib/esm/openai-image-model.js

@@ -0,0 +1,93 @@
+import { ImageModel, imageModelInputSchema, } from "@aigne/core";
+import { snakelize } from "@aigne/core/utils/camelize.js";
+import { checkArguments, pick } from "@aigne/core/utils/type-utils.js";
+import { z } from "zod";
+import { CustomOpenAI } from "./openai.js";
+const DEFAULT_MODEL = "dall-e-2";
+const openAIImageModelInputSchema = imageModelInputSchema.extend({});
+const openAIImageModelOptionsSchema = z.object({
+    apiKey: z.string().optional(),
+    baseURL: z.string().optional(),
+    model: z.string().optional(),
+});
+export class OpenAIImageModel extends ImageModel {
+    options;
+    constructor(options) {
+        super({
+            ...options,
+            inputSchema: openAIImageModelInputSchema,
+            description: options?.description ?? "Draw or edit image by OpenAI image models",
+        });
+        this.options = options;
+        if (options)
+            checkArguments(this.name, openAIImageModelOptionsSchema, options);
+    }
+    _client;
+    apiKeyEnvName = "OPENAI_API_KEY";
+    get client() {
+        if (this._client)
+            return this._client;
+        const { apiKey, url } = 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._client ??= new CustomOpenAI({
+            baseURL: url,
+            apiKey,
+            ...this.options?.clientOptions,
+        });
+        return this._client;
+    }
+    get credential() {
+        return {
+            url: this.options?.baseURL || process.env.OPENAI_BASE_URL,
+            apiKey: this.options?.apiKey || process.env[this.apiKeyEnvName],
+            model: this.options?.model || DEFAULT_MODEL,
+        };
+    }
+    get modelOptions() {
+        return this.options?.modelOptions;
+    }
+    /**
+     * Process the input and generate a response
+     * @param input The input to process
+     * @returns The generated response
+     */
+    async process(input) {
+        const model = input.model || this.credential.model;
+        const inputKeys = [
+            "background",
+            "moderation",
+            "output_compression",
+            "output_format",
+            "prompt",
+            "quality",
+            "size",
+            "style",
+            "user",
+        ];
+        let responseFormat;
+        if (model !== "gpt-image-1") {
+            responseFormat = input.responseFormat === "base64" ? "b64_json" : "url";
+        }
+        const body = {
+            ...snakelize(pick({ ...this.modelOptions, ...input }, inputKeys)),
+            response_format: responseFormat,
+            model,
+        };
+        const response = await this.client.images.generate({ ...body });
+        return {
+            images: (response.data ?? []).map((image) => {
+                if (image.url)
+                    return { url: image.url };
+                if (image.b64_json)
+                    return { base64: image.b64_json };
+                throw new Error("Image response does not contain a valid URL or base64 data");
+            }),
+            usage: {
+                inputTokens: response.usage?.input_tokens || 0,
+                outputTokens: response.usage?.output_tokens || 0,
+            },
+            model,
+        };
+    }
+}

package/lib/esm/openai.d.ts

@@ -0,0 +1,4 @@
+import OpenAI, { type APIError } from "openai";
+export declare class CustomOpenAI extends OpenAI {
+    protected makeStatusError(status: number, error: object, message: string | undefined, headers: Headers): APIError;
+}

package/lib/esm/openai.js

@@ -0,0 +1,10 @@
+import OpenAI from "openai";
+// Use a custom OpenAI client to handle API errors for better error messages
+export class CustomOpenAI extends OpenAI {
+    makeStatusError(status, error, message, headers) {
+        if (!("error" in error) || typeof error.error !== "string") {
+            message = JSON.stringify(error);
+        }
+        return super.makeStatusError(status, error, message, headers);
+    }
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aigne/openai",
-  "version": "0.11.3",
+  "version": "0.12.0",
   "description": "AIGNE OpenAI SDK for integrating with OpenAI's GPT models and API services",
   "publishConfig": {
     "access": "public"
@@ -39,7 +39,7 @@
     "openai": "^5.8.3",
     "uuid": "^11.1.0",
     "zod": "^3.25.67",
-    "@aigne/core": "^1.50.1"
+    "@aigne/core": "^1.53.0"
   },
   "devDependencies": {
     "@types/bun": "^1.2.18",
@@ -47,7 +47,7 @@
     "npm-run-all": "^4.1.5",
     "rimraf": "^6.0.1",
     "typescript": "^5.8.3",
-    "@aigne/test-utils": "^0.5.29"
+    "@aigne/test-utils": "^0.5.32"
   },
   "scripts": {
     "lint": "tsc --noEmit",