@carlonicora/nestjs-neo4jsonapi 0.0.1

package/dist/core/llm/services/llm.service.js

@@ -0,0 +1,457 @@
+"use strict";
+var __decorate = (this && this.__decorate) || function (decorators, target, key, desc) {
+    var c = arguments.length, r = c < 3 ? target : desc === null ? desc = Object.getOwnPropertyDescriptor(target, key) : desc, d;
+    if (typeof Reflect === "object" && typeof Reflect.decorate === "function") r = Reflect.decorate(decorators, target, key, desc);
+    else for (var i = decorators.length - 1; i >= 0; i--) if (d = decorators[i]) r = (c < 3 ? d(r) : c > 3 ? d(target, key, r) : d(target, key)) || r;
+    return c > 3 && r && Object.defineProperty(target, key, r), r;
+};
+var __metadata = (this && this.__metadata) || function (k, v) {
+    if (typeof Reflect === "object" && typeof Reflect.metadata === "function") return Reflect.metadata(k, v);
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.LLMService = void 0;
+const messages_1 = require("@langchain/core/messages");
+const prompts_1 = require("@langchain/core/prompts");
+const common_1 = require("@nestjs/common");
+const agentmessage_type_1 = require("../../../common/enums/agentmessage.type");
+const model_service_1 = require("../../llm/services/model.service");
+const schema_utils_1 = require("../../llm/utils/schema.utils");
+/**
+ * Type guard to validate raw response structure
+ */
+function isValidRaw(raw) {
+    return typeof raw === "object" && raw !== null;
+}
+let LLMService = class LLMService {
+    constructor(modelService) {
+        this.modelService = modelService;
+        this._sessionTokens = {
+            input: 0,
+            output: 0,
+            total: 0,
+            callCount: 0,
+        };
+    }
+    /**
+     * Converts AgentMessageType to LangChain BaseMessage
+     */
+    _convertToBaseMessage(role, content) {
+        switch (role) {
+            case agentmessage_type_1.AgentMessageType.System:
+                return new messages_1.SystemMessage(content);
+            case agentmessage_type_1.AgentMessageType.Assistant:
+                return new messages_1.AIMessage(content);
+            case agentmessage_type_1.AgentMessageType.User:
+                return new messages_1.HumanMessage(content);
+            default:
+                return new messages_1.HumanMessage(content);
+        }
+    }
+    /**
+     * Trims history to prevent context overflow
+     */
+    _trimHistory(history, maxMessages) {
+        if (!maxMessages || history.length <= maxMessages) {
+            return history;
+        }
+        // Keep the most recent messages
+        const trimmed = history.slice(-maxMessages);
+        return trimmed;
+    }
+    /**
+     * Auto-generates instructions from input parameters
+     *
+     * Formats parameters as "key: value" pairs separated by double newlines.
+     * Handles primitives, objects, and arrays intelligently.
+     *
+     * IMPORTANT: For objects/arrays, curly braces are escaped with double braces
+     * ({{ and }}) to prevent ChatPromptTemplate from treating them as template
+     * variables. ChatPromptTemplate will render {{ as literal { in the final prompt.
+     *
+     * @param inputParams - Parameters to format
+     * @returns Formatted instruction string with escaped braces, or empty string if no params
+     */
+    _autoGenerateInstructions(inputParams) {
+        const keys = Object.keys(inputParams);
+        if (keys.length === 0) {
+            return "";
+        }
+        // const formattedPairs = keys.map((key) => {
+        //   const value = inputParams[key];
+        //   // Format the value based on its type
+        //   let formattedValue: string;
+        //   if (value === null || value === undefined) {
+        //     formattedValue = String(value);
+        //   } else if (typeof value === "object") {
+        //     // For objects/arrays, use JSON stringify with formatting
+        //     // CRITICAL: Escape curly braces for ChatPromptTemplate
+        //     // Single braces {} are interpreted as template variables
+        //     // Double braces {{}} render as literal {} in the output
+        //     formattedValue = JSON.stringify(value, null, 2).replace(/{/g, "{{").replace(/}/g, "}}");
+        //   } else {
+        //     formattedValue = String(value);
+        //   }
+        //   return `${key}: ${formattedValue}`;
+        // });
+        const formattedPairs = keys.map((key) => {
+            const value = inputParams[key];
+            let formattedValue;
+            if (value === null || value === undefined) {
+                formattedValue = String(value);
+            }
+            else if (typeof value === "object") {
+                formattedValue = JSON.stringify(value, null, 2).replace(/{/g, "{{").replace(/}/g, "}}");
+            }
+            else {
+                // ✅ FIX: Escape braces in string values too!
+                formattedValue = String(value).replace(/{/g, "{{").replace(/}/g, "}}");
+            }
+            return `${key}: ${formattedValue}`;
+        });
+        return formattedPairs.join("\n\n");
+    }
+    /**
+     * Generates schema-guided instructions with inline descriptions
+     *
+     * This method enhances auto-generated instructions by including field descriptions
+     * from the input schema. This provides the LLM with semantic context about each
+     * input parameter, improving understanding and adherence to constraints.
+     *
+     * Benefits:
+     * - LLM understands field purposes (e.g., "use likes to SUBTLY influence tone")
+     * - LLM receives explicit constraints (e.g., "FORBIDDEN - never repeat these")
+     * - Reduces need for redundant explanations in system prompts
+     * - Single source of truth for input semantics
+     *
+     * Format: "fieldName (description): value"
+     *
+     * @param inputParams - Parameters to format (actual values)
+     * @param inputSchema - Optional Zod schema with descriptions
+     * @returns Formatted instruction string with inline descriptions
+     *
+     * @example Without schema (fallback to auto-generation):
+     * ```typescript
+     * _generateSchemaGuidedInstructions({ name: "Alice" })
+     * // Returns: "name: Alice"
+     * ```
+     *
+     * @example With schema (includes descriptions):
+     * ```typescript
+     * const schema = z.object({
+     *   name: z.string().describe("The user's name"),
+     *   recentActions: z.array(z.string()).describe("FORBIDDEN - never repeat")
+     * });
+     * _generateSchemaGuidedInstructions({ name: "Alice", recentActions: ["wave"] }, schema)
+     * // Returns:
+     * // "name (The user's name): Alice
+     * //
+     * // recentActions (FORBIDDEN - never repeat): ["wave"]"
+     * ```
+     */
+    _generateSchemaGuidedInstructions(inputParams, inputSchema) {
+        const keys = Object.keys(inputParams);
+        if (keys.length === 0) {
+            return "";
+        }
+        // If no schema provided, fall back to basic auto-generation
+        if (!inputSchema) {
+            return this._autoGenerateInstructions(inputParams);
+        }
+        // Extract schema metadata (field descriptions)
+        const schemaMetadata = (0, schema_utils_1.extractSchemaMetadata)(inputSchema);
+        // Format each field with its description (if available)
+        const formattedPairs = keys.map((key) => {
+            const value = inputParams[key];
+            const fieldMetadata = schemaMetadata.fields[key];
+            return (0, schema_utils_1.formatFieldWithDescription)(key, value, fieldMetadata?.description);
+        });
+        return formattedPairs.join("\n\n");
+    }
+    /**
+     * Creates message array for ChatPromptTemplate using MessagesPlaceholder pattern
+     */
+    _createMessages(params) {
+        const templateMessages = [];
+        // Add system prompts
+        params.systemPrompts.forEach((systemPrompt) => {
+            templateMessages.push([agentmessage_type_1.AgentMessageType.System, systemPrompt.replace(/{/g, "{{").replace(/}/g, "}}")]);
+        });
+        // Add placeholder for conversation history (modern LangChain pattern)
+        templateMessages.push(new prompts_1.MessagesPlaceholder("chat_history"));
+        // Determine final instructions: use provided or generate with schema guidance
+        const finalInstructions = params.instructions || this._generateSchemaGuidedInstructions(params.inputParams, params.inputSchema);
+        // Add instructions with {placeholders} intact - ChatPromptTemplate will substitute them
+        templateMessages.push([agentmessage_type_1.AgentMessageType.User, finalInstructions]);
+        // Prepare history messages
+        let historyToUse = params.history || [];
+        // Trim history if needed
+        if (params.maxHistoryMessages) {
+            historyToUse = this._trimHistory(historyToUse, params.maxHistoryMessages);
+        }
+        // Convert to BaseMessage format
+        const historyMessages = historyToUse.map((entry) => this._convertToBaseMessage(entry.role, entry.content));
+        return {
+            template: templateMessages,
+            historyMessages,
+        };
+    }
+    /**
+     * Calls the LLM with structured input/output using LangChain.
+     *
+     * This method:
+     * 1. Builds a chat prompt from system prompts, history, and user instructions
+     * 2. Auto-generates instructions from inputParams if not provided
+     * 3. Trims history if maxHistoryMessages is specified (prevents context overflow)
+     * 4. Substitutes {placeholders} in instructions with values from inputParams
+     * 5. Calls the LLM with structured output enforcement (via function calling)
+     * 6. Implements automatic retry logic with exponential backoff
+     * 7. Returns the parsed response with token usage metadata
+     * 8. Tracks session-level token usage
+     *
+     * @template T - The expected output type (inferred from outputSchema)
+     *
+     * @param params - Call parameters
+     * @param params.inputParams - Variables to substitute in instruction template, or to auto-generate
+     *                              Keys match {placeholders} in instructions (if provided)
+     *                              Example: {character: {...}, userMessage: "Hello"}
+     * @param params.inputSchema - Optional Zod schema for input validation and context injection
+     *                              Field descriptions are extracted and included in prompts
+     * @param params.outputSchema - Zod schema defining expected LLM response structure
+     * @param params.systemPrompts - Array of system prompts to set context/behavior
+     * @param params.instructions - Optional user instructions template with {placeholders}
+     *                               If omitted, auto-generates from inputParams (with schema descriptions if provided)
+     *                               Example: "Character: {character}\nUser says: {userMessage}"
+     * @param params.temperature - Optional temperature override (0-2, default from config)
+     * @param params.history - Optional conversation history as role/content pairs
+     * @param params.maxHistoryMessages - Optional limit on history size (default: unlimited)
+     * @param params.maxTokens - Optional max tokens for response
+     * @param params.timeout - Optional timeout in milliseconds
+     * @param params.metadata - Optional metadata for LangSmith tracking
+     * @param params.stopSequences - Optional stop sequences
+     * @param params.validateInput - Optional flag to enable input validation (default: false)
+     * @param params.tools - Optional array of tools to bind to the LLM
+     * @param params.maxToolIterations - Optional max tool call iterations (default: 5)
+     *
+     * @returns Promise resolving to parsed output + token usage metadata
+     * @throws {Error} If LLM call fails or returns invalid structured output
+     *
+     * @example Simple case (auto-generated instructions):
+     * ```typescript
+     * const response = await llm.call({
+     *   inputParams: { character: {...}, userMessage: "Hello" },
+     *   outputSchema: z.object({ response: z.string() }),
+     *   systemPrompts: ["You are a helpful assistant"],
+     *   // No instructions - auto-generates: "character: {...}\n\nuserMessage: Hello"
+     * });
+     * ```
+     *
+     * @example Custom instructions with placeholders:
+     * ```typescript
+     * const response = await llm.call({
+     *   inputParams: {
+     *     character: { name: "Zoe", description: "..." },
+     *     userMessage: "Hello"
+     *   },
+     *   outputSchema: z.object({ response: z.string() }),
+     *   systemPrompts: ["You are a helpful assistant"],
+     *   instructions: "Character: {character}\nUser says: {userMessage}\nRespond in character:",
+     *   temperature: 0.7,
+     *   maxHistoryMessages: 20,
+     *   metadata: { node_type: "character" },
+     *   history: [
+     *     { role: AgentMessageType.User, content: "Previous message" },
+     *     { role: AgentMessageType.Assistant, content: "Previous response" }
+     *   ]
+     * });
+     * ```
+     */
+    async call(params) {
+        try {
+            // Optional: Validate input parameters against schema
+            if (params.inputSchema && params.validateInput) {
+                try {
+                    params.inputParams = params.inputSchema.parse(params.inputParams);
+                }
+                catch (validationError) {
+                    console.error("[LLMService] Input validation failed:", validationError);
+                    throw new Error(`Invalid input parameters: ${validationError instanceof Error ? validationError.message : "Unknown validation error"}`);
+                }
+            }
+            // Create messages with modern MessagesPlaceholder pattern (with schema-guided instructions)
+            const { template, historyMessages } = this._createMessages({
+                systemPrompts: params.systemPrompts,
+                instructions: params.instructions,
+                inputParams: params.inputParams,
+                inputSchema: params.inputSchema,
+                history: params.history,
+                maxHistoryMessages: params.maxHistoryMessages,
+            });
+            const prompt = prompts_1.ChatPromptTemplate.fromMessages(template);
+            // Get base model
+            const baseModel = this.modelService.getLLM({
+                temperature: params.temperature,
+            });
+            // Build config options for the invocation
+            const configOptions = {};
+            if (params.maxTokens)
+                configOptions.maxTokens = params.maxTokens;
+            if (params.stopSequences)
+                configOptions.stop = params.stopSequences;
+            if (params.metadata)
+                configOptions.metadata = params.metadata;
+            if (params.timeout)
+                configOptions.timeout = params.timeout;
+            // Track token usage across tool iterations
+            let totalInputTokens = 0;
+            let totalOutputTokens = 0;
+            // Build initial messages for the conversation
+            const conversationMessages = await prompt.formatMessages({
+                ...params.inputParams,
+                chat_history: historyMessages,
+            });
+            // If tools are provided, handle tool calling loop
+            if (params.tools && params.tools.length > 0) {
+                const maxIterations = params.maxToolIterations ?? 5;
+                // Build tool map for execution
+                const toolMap = new Map();
+                for (const tool of params.tools) {
+                    toolMap.set(tool.name, tool);
+                }
+                // Bind tools to model
+                const modelWithTools = baseModel.bindTools(params.tools);
+                // Tool calling loop
+                for (let iteration = 0; iteration < maxIterations; iteration++) {
+                    // Call model with tools
+                    const toolResponse = Object.keys(configOptions).length > 0
+                        ? await modelWithTools.invoke(conversationMessages, configOptions)
+                        : await modelWithTools.invoke(conversationMessages);
+                    // Track token usage
+                    const responseUsage = toolResponse.usage_metadata;
+                    if (responseUsage) {
+                        totalInputTokens += responseUsage.input_tokens ?? 0;
+                        totalOutputTokens += responseUsage.output_tokens ?? 0;
+                    }
+                    // Check for tool calls
+                    const toolCalls = toolResponse.tool_calls ?? [];
+                    if (toolCalls.length === 0) {
+                        // No more tool calls - break to get final structured response
+                        break;
+                    }
+                    // Add AI message with tool calls to conversation
+                    conversationMessages.push(toolResponse);
+                    // Execute each tool call
+                    for (const toolCall of toolCalls) {
+                        const tool = toolMap.get(toolCall.name);
+                        if (!tool) {
+                            console.warn(`[LLMService] Tool not found: ${toolCall.name}`);
+                            conversationMessages.push(new messages_1.ToolMessage({
+                                content: `Tool "${toolCall.name}" not found`,
+                                tool_call_id: toolCall.id ?? "",
+                            }));
+                            continue;
+                        }
+                        try {
+                            const result = await tool.invoke(toolCall.args);
+                            const resultStr = typeof result === "string" ? result : JSON.stringify(result);
+                            conversationMessages.push(new messages_1.ToolMessage({
+                                content: resultStr,
+                                tool_call_id: toolCall.id ?? "",
+                            }));
+                        }
+                        catch (error) {
+                            console.error(`[LLMService] Tool error: ${toolCall.name}`, error);
+                            conversationMessages.push(new messages_1.ToolMessage({
+                                content: `Tool error: ${error instanceof Error ? error.message : "Unknown error"}`,
+                                tool_call_id: toolCall.id ?? "",
+                            }));
+                        }
+                    }
+                }
+            }
+            // Get final structured response (unified path for both tool and non-tool flows)
+            const structuredLlm = baseModel.withStructuredOutput(params.outputSchema, {
+                includeRaw: true,
+            });
+            const response = (await structuredLlm.invoke(conversationMessages, Object.keys(configOptions).length > 0 ? configOptions : undefined));
+            // Extract token usage with type guard (includes tool iteration tokens)
+            const raw = isValidRaw(response.raw) ? response.raw : undefined;
+            const input = totalInputTokens + (raw?.usage_metadata?.input_tokens ?? 0);
+            const output = totalOutputTokens + (raw?.usage_metadata?.output_tokens ?? 0);
+            // Update session tracking
+            this._sessionTokens.input += input;
+            this._sessionTokens.output += output;
+            this._sessionTokens.total += input + output;
+            this._sessionTokens.callCount += 1;
+            // Warn if high token usage
+            const totalTokens = input + output;
+            if (totalTokens > 8000) {
+                console.warn(`[LLMService] High token usage detected: ${totalTokens} tokens in this call`);
+            }
+            // Enhanced error handling with detailed diagnostics
+            if (!response.parsed) {
+                const rawContent = raw?.content || "No content";
+                const finishReason = raw?.response_metadata?.finish_reason;
+                console.error("[LLMService] Parsing failed:", {
+                    rawContentPreview: rawContent.substring(0, 500),
+                    finishReason,
+                    schemaName: params.outputSchema.constructor.name,
+                });
+                // Attempt fallback parsing
+                try {
+                    console.warn("[LLMService] Attempting fallback JSON parsing");
+                    const manualParse = JSON.parse(rawContent);
+                    const validated = params.outputSchema.parse(manualParse);
+                    console.warn("[LLMService] Fallback parsing succeeded");
+                    return {
+                        ...validated,
+                        tokenUsage: { input, output },
+                    };
+                }
+                catch (fallbackError) {
+                    throw new Error(`LLM failed to return structured output. ` +
+                        `Finish reason: ${finishReason}. ` +
+                        `Raw content preview: ${rawContent.substring(0, 200)}...` +
+                        `Fallback parsing error: ${fallbackError instanceof Error ? fallbackError.message : String(fallbackError)}`);
+                }
+            }
+            return {
+                ...response.parsed,
+                tokenUsage: {
+                    input,
+                    output,
+                },
+            };
+        }
+        catch (error) {
+            console.error("[LLMService] Error calling LLM:", error);
+            throw new Error(`LLM service error: ${error instanceof Error ? error.message : "Unknown error"}`);
+        }
+    }
+    /**
+     * Get session-level token usage statistics
+     *
+     * @returns Session usage data including total tokens and call count
+     */
+    getSessionUsage() {
+        return { ...this._sessionTokens };
+    }
+    /**
+     * Reset session token tracking
+     *
+     * Useful when starting a new conversation or game session
+     */
+    resetSession() {
+        this._sessionTokens = {
+            input: 0,
+            output: 0,
+            total: 0,
+            callCount: 0,
+        };
+    }
+};
+exports.LLMService = LLMService;
+exports.LLMService = LLMService = __decorate([
+    (0, common_1.Injectable)(),
+    __metadata("design:paramtypes", [model_service_1.ModelService])
+], LLMService);
+//# sourceMappingURL=llm.service.js.map

package/dist/core/llm/services/llm.service.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llm.service.js","sourceRoot":"","sources":["../../../../src/core/llm/services/llm.service.ts"],"names":[],"mappings":";;;;;;;;;;;;AAAA,uDAA4G;AAC5G,qDAAkF;AAElF,2CAA4C;AAE5C,+EAA2E;AAC3E,oEAAgE;AAChE,+DAAiG;AAiBjG;;GAEG;AACH,SAAS,UAAU,CAAC,GAAY;IAC9B,OAAO,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,CAAC;AACjD,CAAC;AA0CM,IAAM,UAAU,GAAhB,MAAM,UAAU;IAGrB,YAA6B,YAA0B;QAA1B,iBAAY,GAAZ,YAAY,CAAc;QACrD,IAAI,CAAC,cAAc,GAAG;YACpB,KAAK,EAAE,CAAC;YACR,MAAM,EAAE,CAAC;YACT,KAAK,EAAE,CAAC;YACR,SAAS,EAAE,CAAC;SACb,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,qBAAqB,CAAC,IAAsB,EAAE,OAAe;QACnE,QAAQ,IAAI,EAAE,CAAC;YACb,KAAK,oCAAgB,CAAC,MAAM;gBAC1B,OAAO,IAAI,wBAAa,CAAC,OAAO,CAAC,CAAC;YACpC,KAAK,oCAAgB,CAAC,SAAS;gBAC7B,OAAO,IAAI,oBAAS,CAAC,OAAO,CAAC,CAAC;YAChC,KAAK,oCAAgB,CAAC,IAAI;gBACxB,OAAO,IAAI,uBAAY,CAAC,OAAO,CAAC,CAAC;YACnC;gBACE,OAAO,IAAI,uBAAY,CAAC,OAAO,CAAC,CAAC;QACrC,CAAC;IACH,CAAC;IAED;;OAEG;IACK,YAAY,CAClB,OAA2D,EAC3D,WAAoB;QAEpB,IAAI,CAAC,WAAW,IAAI,OAAO,CAAC,MAAM,IAAI,WAAW,EAAE,CAAC;YAClD,OAAO,OAAO,CAAC;QACjB,CAAC;QAED,gCAAgC;QAChC,MAAM,OAAO,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,WAAW,CAAC,CAAC;QAE5C,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;;;;;;;;;;;OAYG;IACK,yBAAyB,CAAC,WAAgC;QAChE,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAEtC,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACtB,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,6CAA6C;QAC7C,oCAAoC;QAEpC,0CAA0C;QAC1C,gCAAgC;QAChC,iDAAiD;QACjD,sCAAsC;QACtC,4CAA4C;QAC5C,gEAAgE;QAChE,8DAA8D;QAC9D,gEAAgE;QAChE,+DAA+D;QAC/D,+FAA+F;QAC/F,aAAa;QACb,sCAAsC;QACtC,MAAM;QAEN,wCAAwC;QACxC,MAAM;QACN,MAAM,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE;YACtC,MAAM,KAAK,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;YAE/B,IAAI,cAAsB,CAAC;YAC3B,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;gBAC1C,cAAc,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;YACjC,CAAC;iBAAM,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;gBACrC,cAAc,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YAC1F,CAAC;iBAAM,CAAC;gBACN,6CAA6C;gBAC7C,cAAc,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YACzE,CAAC;YAED,OAAO,GAAG,GAAG,KAAK,cAAc,EAAE,CAAC;QACrC,CAAC,CAAC,CAAC;QAEH,OAAO,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACrC,CAAC;IAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;OAqCG;IACK,iCAAiC,CAAC,WAAgC,EAAE,WAAqB;QAC/F,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAEtC,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACtB,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,4DAA4D;QAC5D,IAAI,CAAC,WAAW,EAAE,CAAC;YACjB,OAAO,IAAI,CAAC,yBAAyB,CAAC,WAAW,CAAC,CAAC;QACrD,CAAC;QAED,+CAA+C;QAC/C,MAAM,cAAc,GAAG,IAAA,oCAAqB,EAAC,WAAW,CAAC,CAAC;QAE1D,wDAAwD;QACxD,MAAM,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE;YACtC,MAAM,KAAK,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;YAC/B,MAAM,aAAa,GAAG,cAAc,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAEjD,OAAO,IAAA,yCAA0B,EAAC,GAAG,EAAE,KAAK,EAAE,aAAa,EAAE,WAAW,CAAC,CAAC;QAC5E,CAAC,CAAC,CAAC;QAEH,OAAO,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACrC,CAAC;IAED;;OAEG;IACK,eAAe,CAAC,MAOvB;QAIC,MAAM,gBAAgB,GAA4D,EAAE,CAAC;QAErF,qBAAqB;QACrB,MAAM,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC,YAAY,EAAE,EAAE;YAC5C,gBAAgB,CAAC,IAAI,CAAC,CAAC,oCAAgB,CAAC,MAAM,EAAE,YAAY,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC;QACzG,CAAC,CAAC,CAAC;QAEH,sEAAsE;QACtE,gBAAgB,CAAC,IAAI,CAAC,IAAI,6BAAmB,CAAC,cAAc,CAAC,CAAC,CAAC;QAE/D,8EAA8E;QAC9E,MAAM,iBAAiB,GACrB,MAAM,CAAC,YAAY,IAAI,IAAI,CAAC,iCAAiC,CAAC,MAAM,CAAC,WAAW,EAAE,MAAM,CAAC,WAAW,CAAC,CAAC;QAExG,wFAAwF;QACxF,gBAAgB,CAAC,IAAI,CAAC,CAAC,oCAAgB,CAAC,IAAI,EAAE,iBAAiB,CAAC,CAAC,CAAC;QAElE,2BAA2B;QAC3B,IAAI,YAAY,GAAG,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC;QAExC,yBAAyB;QACzB,IAAI,MAAM,CAAC,kBAAkB,EAAE,CAAC;YAC9B,YAAY,GAAG,IAAI,CAAC,YAAY,CAAC,YAAY,EAAE,MAAM,CAAC,kBAAkB,CAAC,CAAC;QAC5E,CAAC;QAED,gCAAgC;QAChC,MAAM,eAAe,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,qBAAqB,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC;QAE3G,OAAO;YACL,QAAQ,EAAE,gBAAgB;YAC1B,eAAe;SAChB,CAAC;IACJ,CAAC;IAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;OAqEG;IACH,KAAK,CAAC,IAAI,CAAI,MAAwB;QACpC,IAAI,CAAC;YACH,qDAAqD;YACrD,IAAI,MAAM,CAAC,WAAW,IAAI,MAAM,CAAC,aAAa,EAAE,CAAC;gBAC/C,IAAI,CAAC;oBACH,MAAM,CAAC,WAAW,GAAG,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;gBACpE,CAAC;gBAAC,OAAO,eAAe,EAAE,CAAC;oBACzB,OAAO,CAAC,KAAK,CAAC,uCAAuC,EAAE,eAAe,CAAC,CAAC;oBACxE,MAAM,IAAI,KAAK,CACb,6BAA6B,eAAe,YAAY,KAAK,CAAC,CAAC,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC,0BAA0B,EAAE,CACvH,CAAC;gBACJ,CAAC;YACH,CAAC;YAED,4FAA4F;YAC5F,MAAM,EAAE,QAAQ,EAAE,eAAe,EAAE,GAAG,IAAI,CAAC,eAAe,CAAC;gBACzD,aAAa,EAAE,MAAM,CAAC,aAAa;gBACnC,YAAY,EAAE,MAAM,CAAC,YAAY;gBACjC,WAAW,EAAE,MAAM,CAAC,WAAW;gBAC/B,WAAW,EAAE,MAAM,CAAC,WAAW;gBAC/B,OAAO,EAAE,MAAM,CAAC,OAAO;gBACvB,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;aAC9C,CAAC,CAAC;YAEH,MAAM,MAAM,GAAG,4BAAkB,CAAC,YAAY,CAAC,QAAQ,CAAC,CAAC;YAEzD,iBAAiB;YACjB,MAAM,SAAS,GAAG,IAAI,CAAC,YAAY,CAAC,MAAM,CAAC;gBACzC,WAAW,EAAE,MAAM,CAAC,WAAW;aAChC,CAAC,CAAC;YAEH,0CAA0C;YAC1C,MAAM,aAAa,GAAwB,EAAE,CAAC;YAC9C,IAAI,MAAM,CAAC,SAAS;gBAAE,aAAa,CAAC,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;YACjE,IAAI,MAAM,CAAC,aAAa;gBAAE,aAAa,CAAC,IAAI,GAAG,MAAM,CAAC,aAAa,CAAC;YACpE,IAAI,MAAM,CAAC,QAAQ;gBAAE,aAAa,CAAC,QAAQ,GAAG,MAAM,CAAC,QAAQ,CAAC;YAC9D,IAAI,MAAM,CAAC,OAAO;gBAAE,aAAa,CAAC,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;YAE3D,2CAA2C;YAC3C,IAAI,gBAAgB,GAAG,CAAC,CAAC;YACzB,IAAI,iBAAiB,GAAG,CAAC,CAAC;YAE1B,8CAA8C;YAC9C,MAAM,oBAAoB,GAAkB,MAAM,MAAM,CAAC,cAAc,CAAC;gBACtE,GAAG,MAAM,CAAC,WAAW;gBACrB,YAAY,EAAE,eAAe;aAC9B,CAAC,CAAC;YAEH,kDAAkD;YAClD,IAAI,MAAM,CAAC,KAAK,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAC5C,MAAM,aAAa,GAAG,MAAM,CAAC,iBAAiB,IAAI,CAAC,CAAC;gBAEpD,+BAA+B;gBAC/B,MAAM,OAAO,GAAG,IAAI,GAAG,EAAiC,CAAC;gBACzD,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;oBAChC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;gBAC/B,CAAC;gBAED,sBAAsB;gBACtB,MAAM,cAAc,GAAG,SAAS,CAAC,SAAS,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;gBAEzD,oBAAoB;gBACpB,KAAK,IAAI,SAAS,GAAG,CAAC,EAAE,SAAS,GAAG,aAAa,EAAE,SAAS,EAAE,EAAE,CAAC;oBAC/D,wBAAwB;oBACxB,MAAM,YAAY,GAChB,MAAM,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,MAAM,GAAG,CAAC;wBACnC,CAAC,CAAC,MAAM,cAAc,CAAC,MAAM,CAAC,oBAAoB,EAAE,aAAa,CAAC;wBAClE,CAAC,CAAC,MAAM,cAAc,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC;oBAExD,oBAAoB;oBACpB,MAAM,aAAa,GAAI,YAA0C,CAAC,cAAc,CAAC;oBACjF,IAAI,aAAa,EAAE,CAAC;wBAClB,gBAAgB,IAAI,aAAa,CAAC,YAAY,IAAI,CAAC,CAAC;wBACpD,iBAAiB,IAAI,aAAa,CAAC,aAAa,IAAI,CAAC,CAAC;oBACxD,CAAC;oBAED,uBAAuB;oBACvB,MAAM,SAAS,GAAI,YAA0B,CAAC,UAAU,IAAI,EAAE,CAAC;oBAE/D,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;wBAC3B,8DAA8D;wBAC9D,MAAM;oBACR,CAAC;oBAED,iDAAiD;oBACjD,oBAAoB,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;oBAExC,yBAAyB;oBACzB,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;wBACjC,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;wBAExC,IAAI,CAAC,IAAI,EAAE,CAAC;4BACV,OAAO,CAAC,IAAI,CAAC,gCAAgC,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC;4BAC9D,oBAAoB,CAAC,IAAI,CACvB,IAAI,sBAAW,CAAC;gCACd,OAAO,EAAE,SAAS,QAAQ,CAAC,IAAI,aAAa;gCAC5C,YAAY,EAAE,QAAQ,CAAC,EAAE,IAAI,EAAE;6BAChC,CAAC,CACH,CAAC;4BACF,SAAS;wBACX,CAAC;wBAED,IAAI,CAAC;4BACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;4BAChD,MAAM,SAAS,GAAG,OAAO,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;4BAC/E,oBAAoB,CAAC,IAAI,CACvB,IAAI,sBAAW,CAAC;gCACd,OAAO,EAAE,SAAS;gCAClB,YAAY,EAAE,QAAQ,CAAC,EAAE,IAAI,EAAE;6BAChC,CAAC,CACH,CAAC;wBACJ,CAAC;wBAAC,OAAO,KAAK,EAAE,CAAC;4BACf,OAAO,CAAC,KAAK,CAAC,4BAA4B,QAAQ,CAAC,IAAI,EAAE,EAAE,KAAK,CAAC,CAAC;4BAClE,oBAAoB,CAAC,IAAI,CACvB,IAAI,sBAAW,CAAC;gCACd,OAAO,EAAE,eAAe,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,EAAE;gCAClF,YAAY,EAAE,QAAQ,CAAC,EAAE,IAAI,EAAE;6BAChC,CAAC,CACH,CAAC;wBACJ,CAAC;oBACH,CAAC;gBACH,CAAC;YACH,CAAC;YAED,gFAAgF;YAChF,MAAM,aAAa,GAAG,SAAS,CAAC,oBAAoB,CAAC,MAAM,CAAC,YAAY,EAAE;gBACxE,UAAU,EAAE,IAAI;aACjB,CAAC,CAAC;YAEH,MAAM,QAAQ,GAAG,CAAC,MAAM,aAAa,CAAC,MAAM,CAC1C,oBAAoB,EACpB,MAAM,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS,CAClE,CAA2C,CAAC;YAE7C,uEAAuE;YACvE,MAAM,GAAG,GAAG,UAAU,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC;YAChE,MAAM,KAAK,GAAG,gBAAgB,GAAG,CAAC,GAAG,EAAE,cAAc,EAAE,YAAY,IAAI,CAAC,CAAC,CAAC;YAC1E,MAAM,MAAM,GAAG,iBAAiB,GAAG,CAAC,GAAG,EAAE,cAAc,EAAE,aAAa,IAAI,CAAC,CAAC,CAAC;YAE7E,0BAA0B;YAC1B,IAAI,CAAC,cAAc,CAAC,KAAK,IAAI,KAAK,CAAC;YACnC,IAAI,CAAC,cAAc,CAAC,MAAM,IAAI,MAAM,CAAC;YACrC,IAAI,CAAC,cAAc,CAAC,KAAK,IAAI,KAAK,GAAG,MAAM,CAAC;YAC5C,IAAI,CAAC,cAAc,CAAC,SAAS,IAAI,CAAC,CAAC;YAEnC,2BAA2B;YAC3B,MAAM,WAAW,GAAG,KAAK,GAAG,MAAM,CAAC;YACnC,IAAI,WAAW,GAAG,IAAI,EAAE,CAAC;gBACvB,OAAO,CAAC,IAAI,CAAC,2CAA2C,WAAW,sBAAsB,CAAC,CAAC;YAC7F,CAAC;YAED,oDAAoD;YACpD,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC;gBACrB,MAAM,UAAU,GAAG,GAAG,EAAE,OAAO,IAAI,YAAY,CAAC;gBAChD,MAAM,YAAY,GAAG,GAAG,EAAE,iBAAiB,EAAE,aAAa,CAAC;gBAE3D,OAAO,CAAC,KAAK,CAAC,8BAA8B,EAAE;oBAC5C,iBAAiB,EAAE,UAAU,CAAC,SAAS,CAAC,CAAC,EAAE,GAAG,CAAC;oBAC/C,YAAY;oBACZ,UAAU,EAAE,MAAM,CAAC,YAAY,CAAC,WAAW,CAAC,IAAI;iBACjD,CAAC,CAAC;gBAEH,2BAA2B;gBAC3B,IAAI,CAAC;oBACH,OAAO,CAAC,IAAI,CAAC,+CAA+C,CAAC,CAAC;oBAC9D,MAAM,WAAW,GAAG,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;oBAC3C,MAAM,SAAS,GAAG,MAAM,CAAC,YAAY,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;oBAEzD,OAAO,CAAC,IAAI,CAAC,yCAAyC,CAAC,CAAC;oBAExD,OAAO;wBACL,GAAI,SAAe;wBACnB,UAAU,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE;qBAC9B,CAAC;gBACJ,CAAC;gBAAC,OAAO,aAAa,EAAE,CAAC;oBACvB,MAAM,IAAI,KAAK,CACb,0CAA0C;wBACxC,kBAAkB,YAAY,IAAI;wBAClC,wBAAwB,UAAU,CAAC,SAAS,CAAC,CAAC,EAAE,GAAG,CAAC,KAAK;wBACzD,2BAA2B,aAAa,YAAY,KAAK,CAAC,CAAC,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC,EAAE,CAC9G,CAAC;gBACJ,CAAC;YACH,CAAC;YAED,OAAO;gBACL,GAAI,QAAQ,CAAC,MAAY;gBACzB,UAAU,EAAE;oBACV,KAAK;oBACL,MAAM;iBACP;aACF,CAAC;QACJ,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,CAAC,KAAK,CAAC,iCAAiC,EAAE,KAAK,CAAC,CAAC;YACxD,MAAM,IAAI,KAAK,CAAC,sBAAsB,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,EAAE,CAAC,CAAC;QACpG,CAAC;IACH,CAAC;IAED;;;;OAIG;IACH,eAAe;QACb,OAAO,EAAE,GAAG,IAAI,CAAC,cAAc,EAAE,CAAC;IACpC,CAAC;IAED;;;;OAIG;IACH,YAAY;QACV,IAAI,CAAC,cAAc,GAAG;YACpB,KAAK,EAAE,CAAC;YACR,MAAM,EAAE,CAAC;YACT,KAAK,EAAE,CAAC;YACR,SAAS,EAAE,CAAC;SACb,CAAC;IACJ,CAAC;CACF,CAAA;AAxfY,gCAAU;qBAAV,UAAU;IADtB,IAAA,mBAAU,GAAE;qCAIgC,4BAAY;GAH5C,UAAU,CAwftB"}

package/dist/core/llm/services/model.service.d.ts

@@ -0,0 +1,60 @@
+import { EmbeddingsInterface } from "@langchain/core/embeddings";
+import { BaseChatModel } from "@langchain/core/language_models/chat_models";
+import { ConfigService } from "@nestjs/config";
+import { ClsService } from "nestjs-cls";
+import { BaseConfigInterface } from "../../../config/interfaces";
+export declare class ModelService {
+    private readonly clsService;
+    private readonly configService;
+    private _modelCache;
+    constructor(clsService: ClsService, configService: ConfigService<BaseConfigInterface>);
+    private get aiConfig();
+    /**
+     * Gets a configured LLM instance based on the current config.
+     *
+     * Uses caching to reuse model instances with the same configuration,
+     * improving performance by avoiding repeated instantiation.
+     *
+     * Supports multiple providers:
+     * - `llamacpp`/`local`: Local llama.cpp server (OpenAI-compatible API)
+     * - `openrouter`: OpenRouter cloud service
+     *
+     * @param params - Optional parameters
+     * @param params.temperature - Temperature for text generation (0-2, default: 0.2)
+     *                             Lower = more deterministic, Higher = more creative
+     * @returns Configured BaseChatModel instance from LangChain (cached if available)
+     * @throws {Error} If the configured LLM type is not supported
+     *
+     * @example
+     * ```typescript
+     * const chatModelService = new ChatModelService();
+     * const model = chatModelService.getLLM({ temperature: 0.8 });
+     * // Second call with same temperature returns cached instance
+     * const sameModel = chatModelService.getLLM({ temperature: 0.8 });
+     * ```
+     */
+    getLLM(params?: {
+        temperature?: number;
+    }): BaseChatModel;
+    /**
+     * Clears the model cache.
+     *
+     * Useful when configuration changes or to free up memory.
+     */
+    clearCache(): void;
+    /**
+     * Gets the number of cached model instances.
+     *
+     * @returns Number of cached models
+     */
+    getCacheSize(): number;
+    getEmbedder(): EmbeddingsInterface;
+    getTranscriber(): any;
+    getEmbedderDimensions(): number;
+    transcribeAudio(params: {
+        filePath: string;
+        prompt: string;
+        language?: string;
+    }): Promise<any>;
+}
+//# sourceMappingURL=model.service.d.ts.map

package/dist/core/llm/services/model.service.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"model.service.d.ts","sourceRoot":"","sources":["../../../../src/core/llm/services/model.service.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,4BAA4B,CAAC;AACjE,OAAO,EAAE,aAAa,EAAE,MAAM,6CAA6C,CAAC;AAG5E,OAAO,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAE/C,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AAExC,OAAO,EAAE,mBAAmB,EAAqB,MAAM,4BAA4B,CAAC;AAapF,qBACa,YAAY;IAIrB,OAAO,CAAC,QAAQ,CAAC,UAAU;IAC3B,OAAO,CAAC,QAAQ,CAAC,aAAa;IAJhC,OAAO,CAAC,WAAW,CAA6B;gBAG7B,UAAU,EAAE,UAAU,EACtB,aAAa,EAAE,aAAa,CAAC,mBAAmB,CAAC;IAKpE,OAAO,KAAK,QAAQ,GAEnB;IAED;;;;;;;;;;;;;;;;;;;;;;;OAuBG;IACH,MAAM,CAAC,MAAM,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,aAAa;IAuDxD;;;;OAIG;IACH,UAAU,IAAI,IAAI;IAIlB;;;;OAIG;IACH,YAAY,IAAI,MAAM;IAItB,WAAW,IAAI,mBAAmB;IAmClC,cAAc,IAAI,GAAG;IAoBrB,qBAAqB,IAAI,MAAM;IAIzB,eAAe,CAAC,MAAM,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,GAAG,CAAC;CAQrG"}