task loop

renderer/src/components/main-panel/chat/core/xml-wrapper.ts

@@ -288,9 +288,6 @@ export async function handleXmlWrapperToolcall(toolcall: XmlToolCall): Promise<T
             state: MessageState.InvalidXml
         }
     }
-
-    // 进行调用，根据结果返回不同的值
-    console.log(toolcall);
     
     const toolResponse = await mcpClientAdapter.callTool(toolcall.name, toolcall.parameters);
     return handleToolResponse(toolResponse);

service/src/hook/adapter.ts

@@ -152,3 +152,60 @@ export class TaskLoopAdapter {
     }
 }
 
+import { TaskLoop } from '../../task-loop.js';
+import * as fs from 'fs';
+import chalk from 'chalk';
+
+export class OAgent {
+    private adapter: TaskLoopAdapter;
+    constructor() {
+        this.adapter = new TaskLoopAdapter();
+    }
+
+    public addMcp(mcpOption: IConnectionArgs) {
+        this.adapter.addMcp(mcpOption);
+    }
+
+    /**
+     * @description Load MCP configuration from file.
+     * Supports multiple MCP backends and a default LLM model configuration.
+     * 
+     * Configuration below can be generated from OpenMCP Client directly: https://kirigaya.cn/openmcp
+     *
+     * @example
+     * Example configuration:
+     * {
+     *   "version": "1.0.0",
+     *   "mcpServers": {
+     *     "openmemory": {
+     *       "command": "npx",
+     *       "args": ["-y", "openmemory"],
+     *       "env": {
+     *         "OPENMEMORY_API_KEY": "YOUR_API_KEY",
+     *         "CLIENT_NAME": "openmemory"
+     *       },
+     *       "description": "A MCP for long-term memory support",
+     *       "prompt": "You are a helpful assistant."
+     *     }
+     *   },
+     *   "llm": {
+     *     "baseURL": "https://api.deepseek.com",
+     *     "apiToken": "YOUR_API_KEY",
+     *     "model": "deepseek-chat"
+     *   }
+     * }
+     *
+     * @param configPath - Path to the configuration file
+     */
+    public loadMcpConfig(configPath: string) {
+        if (!fs.existsSync(configPath)) {
+            console.log(
+                chalk.red(`× configPath not exists! ${configPath}`),
+            );
+            throw new Error(`× configPath not exists! ${configPath}`);
+        }
+
+        const config = JSON.parse(fs.readFileSync(configPath, 'utf-8'));
+
+    }
+}

service/task-loop.d.ts

@@ -0,0 +1,303 @@
+/* eslint-disable */
+import type { OpenAI } from 'openai';
+
+export type ChatCompletionChunk = OpenAI.Chat.Completions.ChatCompletionChunk;
+export type ChatCompletionCreateParamsBase = OpenAI.Chat.Completions.ChatCompletionCreateParams & { id?: string };
+
+interface SchemaProperty {
+	title: string;
+	type: string;
+	description?: string;
+}
+
+interface InputSchema {
+	type: string;
+	properties: Record<string, SchemaProperty>;
+	required?: string[];
+	title?: string;
+	$defs?: any;
+}
+
+interface ToolItem {
+	name: string;
+	description: string;
+	inputSchema: InputSchema;
+    enabled: boolean;
+	anyOf?: any;
+}
+
+interface IExtraInfo {
+    created: number,
+    state: MessageState,
+    serverName: string,
+    usage?: ChatCompletionChunk['usage'];
+    enableXmlWrapper: boolean;
+    [key: string]: any;
+}
+
+
+interface ToolMessage {
+    role: 'tool';
+    index: number;
+    content: ToolCallContent[];
+    tool_call_id?: string
+    name?: string // 工具名称，当 role 为 tool
+    tool_calls?: ToolCall[],
+    extraInfo: IExtraInfo
+}
+
+interface TextMessage {
+    role: 'user' | 'assistant' | 'system';
+    content: string;
+    tool_call_id?: string
+    name?: string // 工具名称，当 role 为 tool
+    tool_calls?: ToolCall[],
+    extraInfo: IExtraInfo
+}
+
+export type ChatMessage = ToolMessage | TextMessage;
+
+interface ChatStorage {
+    messages: ChatMessage[]
+    settings: ChatSetting
+}
+
+interface EnableToolItem {
+    name: string;
+    description: string;
+    enabled: boolean;
+    inputSchema: InputSchema;
+}
+
+
+export type Ref<T> = {
+    value: T;
+};
+
+export interface ToolCall {
+    id?: string;
+    index?: number;
+    type: string;
+    function: {
+        name: string;
+        arguments: string;
+    }
+}
+
+export interface ToolCallContent {
+    type: string;
+    text: string;
+	[key: string]: any;
+}
+
+export interface ToolCallResult {
+    state: MessageState;
+    content: ToolCallContent[];
+}
+
+export enum MessageState {
+    ServerError = 'server internal error',
+    ReceiveChunkError = 'receive chunk error',
+    Timeout = 'timeout',
+    MaxEpochs = 'max epochs',
+    Unknown = 'unknown error',
+    Abort = 'abort',
+    ToolCall = 'tool call failed',
+    None = 'none',
+    Success = 'success',
+    ParseJsonError = 'parse json error'
+}
+
+export interface IErrorMssage {
+    state: MessageState;
+    msg: string;
+}
+
+export interface IDoConversationResult {
+    stop: boolean;
+}
+
+export interface TaskLoopOptions {
+    /**
+     * The maximum number of epochs (conversation rounds) to perform.
+     */
+    maxEpochs?: number;
+
+    /**
+     * The maximum number of retries allowed when parsing JSON responses fails.
+     */
+    maxJsonParseRetry?: number;
+
+    /**
+     * A custom adapter that can be used to modify behavior or integrate with different environments.
+     */
+    adapter?: any;
+
+    /**
+     * Verbosity level for logging:
+     * 0 - Silent, 1 - Errors only, 2 - Warnings and errors, 3 - Full debug output.
+     */
+    verbose?: 0 | 1 | 2 | 3;
+}
+
+interface ChatSetting {
+    /**
+     * Index of the selected language model from a list of available models.
+     */
+    modelIndex: number;
+
+    /**
+     * System-level prompt used to guide the behavior of the assistant.
+     */
+    systemPrompt: string;
+
+    /**
+     * List of tools that are enabled and available during the chat.
+     */
+    enableTools: EnableToolItem[];
+
+    /**
+     * Sampling temperature for generating responses.
+     * Higher values (e.g., 0.8) make output more random; lower values (e.g., 0.2) make it more focused and deterministic.
+     */
+    temperature: number;
+
+    /**
+     * Whether web search is enabled for enhancing responses with real-time information.
+     */
+    enableWebSearch: boolean;
+
+    /**
+     * Maximum length of the conversation context to keep.
+     */
+    contextLength: number;
+
+    /**
+     * Whether multiple tools can be called in parallel within a single message.
+     */
+    parallelToolCalls: boolean;
+
+    /**
+     * Whether to wrap tool call responses in XML format.
+     */
+    enableXmlWrapper: boolean;
+}
+
+/**
+ * @description 对任务循环进行的抽象封装
+ */
+export class TaskLoop {
+    constructor(taskOptions?: TaskLoopOptions);
+
+    /**
+     * @description make chat data
+     * @param tabStorage 
+     */
+    makeChatData(tabStorage: any): ChatCompletionCreateParamsBase | undefined;
+
+    /**
+     * @description stop the task loop
+     */
+    abort(): void;
+
+    /**
+     * @description Register a callback function triggered on error
+     * @param handler 
+     */
+    registerOnError(handler: (msg: IErrorMssage) => void): void;
+
+    /**
+     * @description Register a callback function triggered on chunk
+     * @param handler 
+     */
+    registerOnChunk(handler: (chunk: ChatCompletionChunk) => void): void;
+
+    /**
+     * @description Register a callback function triggered at the beginning of each epoch
+     * @param handler 
+     */
+    registerOnDone(handler: () => void): void;
+
+    /**
+     * @description Register a callback function triggered at the beginning of each epoch
+     * @param handler 
+     */
+    registerOnEpoch(handler: () => void): void;
+
+    /**
+     * @description Registers a callback function that is triggered when a tool call is completed. This method allows you to intercept and modify the output of the tool call.
+     * @param handler 
+     */
+    registerOnToolCalled(handler: (toolCallResult: ToolCallResult) => ToolCallResult): void;
+
+    /**
+     * @description Register a callback triggered after tool call finishes. You can intercept and modify the output.
+     * @param handler 
+     */
+    registerOnToolCall(handler: (toolCall: ToolCall) => ToolCall): void;
+
+    /**
+     * @description Get current LLM configuration
+     */
+    getLlmConfig(): any;
+
+    /**
+     * @description Set the current LLM configuration, for Node.js environment
+     * @param config 
+     * @example
+     * setLlmConfig({
+     *     id: 'openai',
+     *     baseUrl: 'https://api.openai.com/v1',
+     *     userToken: 'sk-xxx',
+     *     userModel: 'gpt-3.5-turbo',
+     * })
+     */
+    setLlmConfig(config: any): void;
+
+    /**
+     * @description Set proxy server
+     * @param maxEpochs 
+     */
+    setMaxEpochs(maxEpochs: number): void;
+
+    /**
+     * @description bind streaming content and tool calls
+     */
+    bindStreaming(content: Ref<string>, toolCalls: Ref<ToolCall[]>): void;
+
+    /**
+     * @description not finish
+     */
+    connectToService(): Promise<void>;
+
+    /**
+     * @description 设置代理服务器
+     * @param proxyServer 
+     */
+    setProxyServer(proxyServer: string): void;
+
+    /**
+     * @description Get all available tool list
+     */
+    listTools(): Promise<ToolItem[]>;
+
+    /**
+     * @description Start the loop and asynchronously update the DOM
+     */
+    start(tabStorage: any, userMessage: string): Promise<void>;
+
+    /**
+     * @description Create single conversation context
+     */
+    createStorage(settings?: ChatSetting): Promise<ChatStorage>;
+}
+
+export declare const getToolSchema: any;
+export declare const useMessageBridge: any;
+export declare const llmManager: any;
+export declare const llms: any;
+export declare const pinkLog: any;
+export declare const redLog: any;
+export declare const ElMessage: any;
+export declare const handleToolCalls: any;
+export declare const getPlatform: any;