agentic-flow 2.0.7 → 3.0.0-alpha.2

package/dist/agentic-flow/src/router/providers/onnx.d.ts

@@ -0,0 +1,65 @@
+/**
+ * ONNX Runtime Provider for Local Model Inference
+ *
+ * Supports CPU and GPU execution providers for optimized local inference
+ * Compatible with Phi-3, Llama, and other ONNX models
+ */
+import type { LLMProvider, ChatParams, ChatResponse, StreamChunk } from '../types.js';
+export interface ONNXConfig {
+    modelPath?: string;
+    modelId?: string;
+    executionProviders?: string[];
+    sessionOptions?: any;
+    maxTokens?: number;
+    temperature?: number;
+}
+export declare class ONNXProvider implements LLMProvider {
+    name: string;
+    type: "custom";
+    supportsStreaming: boolean;
+    supportsTools: boolean;
+    supportsMCP: boolean;
+    private session;
+    private generator;
+    private config;
+    private executionProviders;
+    constructor(config?: ONNXConfig);
+    /**
+     * Detect available execution providers
+     */
+    private detectExecutionProviders;
+    /**
+     * Initialize ONNX session with model
+     */
+    private initializeSession;
+    /**
+     * Format messages for model input
+     */
+    private formatMessages;
+    /**
+     * Chat completion
+     */
+    chat(params: ChatParams): Promise<ChatResponse>;
+    /**
+     * Streaming generation
+     */
+    stream(params: ChatParams): AsyncGenerator<StreamChunk>;
+    /**
+     * Validate capabilities
+     */
+    validateCapabilities(features: string[]): boolean;
+    /**
+     * Get model info
+     */
+    getModelInfo(): {
+        modelId: string;
+        executionProviders: string[];
+        supportsGPU: boolean;
+        initialized: boolean;
+    };
+    /**
+     * Cleanup resources
+     */
+    dispose(): Promise<void>;
+}
+//# sourceMappingURL=onnx.d.ts.map

package/dist/agentic-flow/src/router/providers/onnx.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"onnx.d.ts","sourceRoot":"","sources":["../../../../../src/router/providers/onnx.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EACV,WAAW,EACX,UAAU,EACV,YAAY,EACZ,WAAW,EAIZ,MAAM,aAAa,CAAC;AA0BrB,MAAM,WAAW,UAAU;IACzB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,cAAc,CAAC,EAAE,GAAG,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,qBAAa,YAAa,YAAW,WAAW;IAC9C,IAAI,SAAU;IACd,IAAI,EAAG,QAAQ,CAAU;IACzB,iBAAiB,UAAQ;IACzB,aAAa,UAAS;IACtB,WAAW,UAAS;IAEpB,OAAO,CAAC,OAAO,CAAa;IAC5B,OAAO,CAAC,SAAS,CAAa;IAC9B,OAAO,CAAC,MAAM,CAAa;IAC3B,OAAO,CAAC,kBAAkB,CAAgB;gBAE9B,MAAM,GAAE,UAAe;IASnC;;OAEG;YACW,wBAAwB;IAgCtC;;OAEG;YACW,iBAAiB;IA8B/B;;OAEG;IACH,OAAO,CAAC,cAAc;IAsBtB;;OAEG;IACG,IAAI,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAAC,YAAY,CAAC;IA+DrD;;OAEG;IACI,MAAM,CAAC,MAAM,EAAE,UAAU,GAAG,cAAc,CAAC,WAAW,CAAC;IAqD9D;;OAEG;IACH,oBAAoB,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,OAAO;IAKjD;;OAEG;IACH,YAAY;;;;;;IASZ;;OAEG;IACG,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;CAS/B"}

package/dist/agentic-flow/src/router/providers/onnx.js

@@ -0,0 +1,265 @@
+/**
+ * ONNX Runtime Provider for Local Model Inference
+ *
+ * Supports CPU and GPU execution providers for optimized local inference
+ * Compatible with Phi-3, Llama, and other ONNX models
+ */
+// Dynamic imports for optional ONNX dependencies
+let ort;
+let transformers;
+async function ensureOnnxDependencies() {
+    if (!ort) {
+        try {
+            const ortModule = await import('onnxruntime-node');
+            ort = ortModule;
+        }
+        catch (e) {
+            throw new Error('onnxruntime-node not installed. Run: npm install onnxruntime-node');
+        }
+    }
+    if (!transformers) {
+        try {
+            const transformersModule = await import('@xenova/transformers');
+            transformers = transformersModule;
+            transformers.env.allowLocalModels = true;
+        }
+        catch (e) {
+            throw new Error('@xenova/transformers not installed. Run: npm install @xenova/transformers');
+        }
+    }
+}
+export class ONNXProvider {
+    name = 'onnx';
+    type = 'custom';
+    supportsStreaming = true;
+    supportsTools = false;
+    supportsMCP = false;
+    session = null;
+    generator = null;
+    config;
+    executionProviders = [];
+    constructor(config = {}) {
+        this.config = {
+            modelId: config.modelId || 'Xenova/Phi-3-mini-4k-instruct',
+            maxTokens: config.maxTokens || 512,
+            temperature: config.temperature || 0.7,
+            ...config
+        };
+    }
+    /**
+     * Detect available execution providers
+     */
+    async detectExecutionProviders() {
+        const providers = [];
+        // Try CUDA for NVIDIA GPUs
+        try {
+            if (process.platform === 'linux') {
+                providers.push('cuda');
+                this.executionProviders.push('cuda');
+            }
+        }
+        catch (e) {
+            // CUDA not available
+        }
+        // Try DirectML for Windows GPUs
+        try {
+            if (process.platform === 'win32') {
+                providers.push('dml');
+                this.executionProviders.push('dml');
+            }
+        }
+        catch (e) {
+            // DirectML not available
+        }
+        // Always fallback to CPU
+        providers.push('cpu');
+        this.executionProviders.push('cpu');
+        console.log(`🔧 ONNX Execution Providers: ${this.executionProviders.join(', ')}`);
+        return providers;
+    }
+    /**
+     * Initialize ONNX session with model
+     */
+    async initializeSession() {
+        if (this.generator)
+            return;
+        try {
+            await ensureOnnxDependencies();
+            console.log(`📦 Loading ONNX model: ${this.config.modelId}`);
+            // Use Transformers.js for easier model loading
+            this.generator = await transformers.pipeline('text-generation', this.config.modelId, {
+                quantized: true, // Use quantized models for better CPU performance
+            });
+            console.log(`✅ ONNX model loaded successfully`);
+        }
+        catch (error) {
+            const providerError = {
+                name: 'ONNXInitError',
+                message: `Failed to initialize ONNX model: ${error}`,
+                provider: 'onnx',
+                retryable: false
+            };
+            throw providerError;
+        }
+    }
+    /**
+     * Format messages for model input
+     */
+    formatMessages(messages) {
+        // Simple chat template for Phi-3
+        let prompt = '';
+        for (const msg of messages) {
+            const content = typeof msg.content === 'string'
+                ? msg.content
+                : msg.content.map(c => c.type === 'text' ? c.text : '').join('');
+            if (msg.role === 'user') {
+                prompt += `<|user|>\n${content}<|end|>\n`;
+            }
+            else if (msg.role === 'assistant') {
+                prompt += `<|assistant|>\n${content}<|end|>\n`;
+            }
+            else if (msg.role === 'system') {
+                prompt += `<|system|>\n${content}<|end|>\n`;
+            }
+        }
+        prompt += '<|assistant|>\n';
+        return prompt;
+    }
+    /**
+     * Chat completion
+     */
+    async chat(params) {
+        await this.initializeSession();
+        const startTime = Date.now();
+        const prompt = this.formatMessages(params.messages);
+        try {
+            const result = await this.generator(prompt, {
+                max_new_tokens: params.maxTokens || this.config.maxTokens,
+                temperature: params.temperature || this.config.temperature,
+                do_sample: true,
+                top_p: 0.9,
+            });
+            const generatedText = result[0].generated_text;
+            // Extract only the new assistant response
+            const assistantResponse = generatedText
+                .split('<|assistant|>')
+                .pop()
+                ?.split('<|end|>')[0]
+                ?.trim() || '';
+            const latency = Date.now() - startTime;
+            // Estimate token counts (rough approximation)
+            const inputTokens = Math.ceil(prompt.length / 4);
+            const outputTokens = Math.ceil(assistantResponse.length / 4);
+            const content = [{
+                    type: 'text',
+                    text: assistantResponse
+                }];
+            return {
+                id: `onnx-${Date.now()}`,
+                model: this.config.modelId || 'onnx-model',
+                content,
+                stopReason: 'end_turn',
+                usage: {
+                    inputTokens,
+                    outputTokens
+                },
+                metadata: {
+                    provider: 'onnx',
+                    model: this.config.modelId,
+                    latency,
+                    cost: 0, // Local inference is free
+                    executionProviders: this.executionProviders
+                }
+            };
+        }
+        catch (error) {
+            const providerError = {
+                name: 'ONNXInferenceError',
+                message: `ONNX inference failed: ${error}`,
+                provider: 'onnx',
+                retryable: true
+            };
+            throw providerError;
+        }
+    }
+    /**
+     * Streaming generation
+     */
+    async *stream(params) {
+        await this.initializeSession();
+        const prompt = this.formatMessages(params.messages);
+        try {
+            // Note: Transformers.js doesn't natively support streaming
+            // We'll simulate it by yielding tokens as they're generated
+            const result = await this.generator(prompt, {
+                max_new_tokens: params.maxTokens || this.config.maxTokens,
+                temperature: params.temperature || this.config.temperature,
+                do_sample: true,
+                top_p: 0.9,
+            });
+            const generatedText = result[0].generated_text;
+            const assistantResponse = generatedText
+                .split('<|assistant|>')
+                .pop()
+                ?.split('<|end|>')[0]
+                ?.trim() || '';
+            // Simulate streaming by chunking the response
+            const words = assistantResponse.split(' ');
+            for (let i = 0; i < words.length; i++) {
+                const chunk = words[i] + (i < words.length - 1 ? ' ' : '');
+                yield {
+                    type: 'content_block_delta',
+                    delta: {
+                        type: 'text_delta',
+                        text: chunk
+                    }
+                };
+                // Small delay to simulate real streaming
+                await new Promise(resolve => setTimeout(resolve, 10));
+            }
+            yield {
+                type: 'message_stop'
+            };
+        }
+        catch (error) {
+            const providerError = {
+                name: 'ONNXStreamError',
+                message: `ONNX streaming failed: ${error}`,
+                provider: 'onnx',
+                retryable: true
+            };
+            throw providerError;
+        }
+    }
+    /**
+     * Validate capabilities
+     */
+    validateCapabilities(features) {
+        const supported = ['chat', 'stream'];
+        return features.every(f => supported.includes(f));
+    }
+    /**
+     * Get model info
+     */
+    getModelInfo() {
+        return {
+            modelId: this.config.modelId,
+            executionProviders: this.executionProviders,
+            supportsGPU: this.executionProviders.includes('cuda') || this.executionProviders.includes('dml'),
+            initialized: this.generator !== null
+        };
+    }
+    /**
+     * Cleanup resources
+     */
+    async dispose() {
+        if (this.generator) {
+            this.generator = null;
+        }
+        if (this.session) {
+            await this.session.release();
+            this.session = null;
+        }
+    }
+}
+//# sourceMappingURL=onnx.js.map

package/dist/agentic-flow/src/router/providers/onnx.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"onnx.js","sourceRoot":"","sources":["../../../../../src/router/providers/onnx.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAYH,iDAAiD;AACjD,IAAI,GAAQ,CAAC;AACb,IAAI,YAAiB,CAAC;AAEtB,KAAK,UAAU,sBAAsB;IACnC,IAAI,CAAC,GAAG,EAAE,CAAC;QACT,IAAI,CAAC;YACH,MAAM,SAAS,GAAG,MAAM,MAAM,CAAC,kBAAyB,CAAC,CAAC;YAC1D,GAAG,GAAG,SAAS,CAAC;QAClB,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,MAAM,IAAI,KAAK,CAAC,mEAAmE,CAAC,CAAC;QACvF,CAAC;IACH,CAAC;IACD,IAAI,CAAC,YAAY,EAAE,CAAC;QAClB,IAAI,CAAC;YACH,MAAM,kBAAkB,GAAG,MAAM,MAAM,CAAC,sBAA6B,CAAC,CAAC;YACvE,YAAY,GAAG,kBAAkB,CAAC;YAClC,YAAY,CAAC,GAAG,CAAC,gBAAgB,GAAG,IAAI,CAAC;QAC3C,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,MAAM,IAAI,KAAK,CAAC,2EAA2E,CAAC,CAAC;QAC/F,CAAC;IACH,CAAC;AACH,CAAC;AAWD,MAAM,OAAO,YAAY;IACvB,IAAI,GAAG,MAAM,CAAC;IACd,IAAI,GAAG,QAAiB,CAAC;IACzB,iBAAiB,GAAG,IAAI,CAAC;IACzB,aAAa,GAAG,KAAK,CAAC;IACtB,WAAW,GAAG,KAAK,CAAC;IAEZ,OAAO,GAAQ,IAAI,CAAC;IACpB,SAAS,GAAQ,IAAI,CAAC;IACtB,MAAM,CAAa;IACnB,kBAAkB,GAAa,EAAE,CAAC;IAE1C,YAAY,SAAqB,EAAE;QACjC,IAAI,CAAC,MAAM,GAAG;YACZ,OAAO,EAAE,MAAM,CAAC,OAAO,IAAI,+BAA+B;YAC1D,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,GAAG;YAClC,WAAW,EAAE,MAAM,CAAC,WAAW,IAAI,GAAG;YACtC,GAAG,MAAM;SACV,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,wBAAwB;QACpC,MAAM,SAAS,GAAa,EAAE,CAAC;QAE/B,2BAA2B;QAC3B,IAAI,CAAC;YACH,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;gBACjC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACvB,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YACvC,CAAC;QACH,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,qBAAqB;QACvB,CAAC;QAED,gCAAgC;QAChC,IAAI,CAAC;YACH,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;gBACjC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;gBACtB,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YACtC,CAAC;QACH,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,yBAAyB;QAC3B,CAAC;QAED,yBAAyB;QACzB,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACtB,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAEpC,OAAO,CAAC,GAAG,CAAC,gCAAgC,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAElF,OAAO,SAAS,CAAC;IACnB,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,iBAAiB;QAC7B,IAAI,IAAI,CAAC,SAAS;YAAE,OAAO;QAE3B,IAAI,CAAC;YACH,MAAM,sBAAsB,EAAE,CAAC;YAE/B,OAAO,CAAC,GAAG,CAAC,0BAA0B,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;YAE7D,+CAA+C;YAC/C,IAAI,CAAC,SAAS,GAAG,MAAM,YAAY,CAAC,QAAQ,CAC1C,iBAAiB,EACjB,IAAI,CAAC,MAAM,CAAC,OAAO,EACnB;gBACE,SAAS,EAAE,IAAI,EAAE,kDAAkD;aACpE,CACF,CAAC;YAEF,OAAO,CAAC,GAAG,CAAC,kCAAkC,CAAC,CAAC;QAElD,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,GAAkB;gBACnC,IAAI,EAAE,eAAe;gBACrB,OAAO,EAAE,oCAAoC,KAAK,EAAE;gBACpD,QAAQ,EAAE,MAAM;gBAChB,SAAS,EAAE,KAAK;aACjB,CAAC;YACF,MAAM,aAAa,CAAC;QACtB,CAAC;IACH,CAAC;IAED;;OAEG;IACK,cAAc,CAAC,QAAmB;QACxC,iCAAiC;QACjC,IAAI,MAAM,GAAG,EAAE,CAAC;QAEhB,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;YAC3B,MAAM,OAAO,GAAG,OAAO,GAAG,CAAC,OAAO,KAAK,QAAQ;gBAC7C,CAAC,CAAC,GAAG,CAAC,OAAO;gBACb,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YAEnE,IAAI,GAAG,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBACxB,MAAM,IAAI,aAAa,OAAO,WAAW,CAAC;YAC5C,CAAC;iBAAM,IAAI,GAAG,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;gBACpC,MAAM,IAAI,kBAAkB,OAAO,WAAW,CAAC;YACjD,CAAC;iBAAM,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;gBACjC,MAAM,IAAI,eAAe,OAAO,WAAW,CAAC;YAC9C,CAAC;QACH,CAAC;QAED,MAAM,IAAI,iBAAiB,CAAC;QAC5B,OAAO,MAAM,CAAC;IAChB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,IAAI,CAAC,MAAkB;QAC3B,MAAM,IAAI,CAAC,iBAAiB,EAAE,CAAC;QAE/B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,MAAM,GAAG,IAAI,CAAC,cAAc,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QAEpD,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE;gBAC1C,cAAc,EAAE,MAAM,CAAC,SAAS,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS;gBACzD,WAAW,EAAE,MAAM,CAAC,WAAW,IAAI,IAAI,CAAC,MAAM,CAAC,WAAW;gBAC1D,SAAS,EAAE,IAAI;gBACf,KAAK,EAAE,GAAG;aACX,CAAC,CAAC;YAEH,MAAM,aAAa,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC;YAE/C,0CAA0C;YAC1C,MAAM,iBAAiB,GAAG,aAAa;iBACpC,KAAK,CAAC,eAAe,CAAC;iBACtB,GAAG,EAAE;gBACN,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC;gBACrB,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;YAEjB,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;YAEvC,8CAA8C;YAC9C,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;YACjD,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,iBAAiB,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;YAE7D,MAAM,OAAO,GAAmB,CAAC;oBAC/B,IAAI,EAAE,MAAM;oBACZ,IAAI,EAAE,iBAAiB;iBACxB,CAAC,CAAC;YAEH,OAAO;gBACL,EAAE,EAAE,QAAQ,IAAI,CAAC,GAAG,EAAE,EAAE;gBACxB,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,YAAY;gBAC1C,OAAO;gBACP,UAAU,EAAE,UAAU;gBACtB,KAAK,EAAE;oBACL,WAAW;oBACX,YAAY;iBACb;gBACD,QAAQ,EAAE;oBACR,QAAQ,EAAE,MAAM;oBAChB,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,OAAO;oBAC1B,OAAO;oBACP,IAAI,EAAE,CAAC,EAAE,0BAA0B;oBACnC,kBAAkB,EAAE,IAAI,CAAC,kBAAkB;iBAC5C;aACF,CAAC;QAEJ,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,GAAkB;gBACnC,IAAI,EAAE,oBAAoB;gBAC1B,OAAO,EAAE,0BAA0B,KAAK,EAAE;gBAC1C,QAAQ,EAAE,MAAM;gBAChB,SAAS,EAAE,IAAI;aAChB,CAAC;YACF,MAAM,aAAa,CAAC;QACtB,CAAC;IACH,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,CAAC,MAAM,CAAC,MAAkB;QAC9B,MAAM,IAAI,CAAC,iBAAiB,EAAE,CAAC;QAE/B,MAAM,MAAM,GAAG,IAAI,CAAC,cAAc,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QAEpD,IAAI,CAAC;YACH,2DAA2D;YAC3D,4DAA4D;YAC5D,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE;gBAC1C,cAAc,EAAE,MAAM,CAAC,SAAS,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS;gBACzD,WAAW,EAAE,MAAM,CAAC,WAAW,IAAI,IAAI,CAAC,MAAM,CAAC,WAAW;gBAC1D,SAAS,EAAE,IAAI;gBACf,KAAK,EAAE,GAAG;aACX,CAAC,CAAC;YAEH,MAAM,aAAa,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC;YAC/C,MAAM,iBAAiB,GAAG,aAAa;iBACpC,KAAK,CAAC,eAAe,CAAC;iBACtB,GAAG,EAAE;gBACN,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC;gBACrB,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;YAEjB,8CAA8C;YAC9C,MAAM,KAAK,GAAG,iBAAiB,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;YAC3C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBACtC,MAAM,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;gBAC3D,MAAM;oBACJ,IAAI,EAAE,qBAAqB;oBAC3B,KAAK,EAAE;wBACL,IAAI,EAAE,YAAY;wBAClB,IAAI,EAAE,KAAK;qBACZ;iBACF,CAAC;gBAEF,yCAAyC;gBACzC,MAAM,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC;YACxD,CAAC;YAED,MAAM;gBACJ,IAAI,EAAE,cAAc;aACrB,CAAC;QAEJ,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,GAAkB;gBACnC,IAAI,EAAE,iBAAiB;gBACvB,OAAO,EAAE,0BAA0B,KAAK,EAAE;gBAC1C,QAAQ,EAAE,MAAM;gBAChB,SAAS,EAAE,IAAI;aAChB,CAAC;YACF,MAAM,aAAa,CAAC;QACtB,CAAC;IACH,CAAC;IAED;;OAEG;IACH,oBAAoB,CAAC,QAAkB;QACrC,MAAM,SAAS,GAAG,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAC;QACrC,OAAO,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IACpD,CAAC;IAED;;OAEG;IACH,YAAY;QACV,OAAO;YACL,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC,OAAO;YAC5B,kBAAkB,EAAE,IAAI,CAAC,kBAAkB;YAC3C,WAAW,EAAE,IAAI,CAAC,kBAAkB,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,IAAI,CAAC,kBAAkB,CAAC,QAAQ,CAAC,KAAK,CAAC;YAChG,WAAW,EAAE,IAAI,CAAC,SAAS,KAAK,IAAI;SACrC,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,OAAO;QACX,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YACnB,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC;QACxB,CAAC;QACD,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACjB,MAAM,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC;YAC7B,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC;QACtB,CAAC;IACH,CAAC;CACF","sourcesContent":["/**\n * ONNX Runtime Provider for Local Model Inference\n *\n * Supports CPU and GPU execution providers for optimized local inference\n * Compatible with Phi-3, Llama, and other ONNX models\n */\n\nimport type {\n  LLMProvider,\n  ChatParams,\n  ChatResponse,\n  StreamChunk,\n  ProviderError,\n  Message,\n  ContentBlock\n} from '../types.js';\n\n// Dynamic imports for optional ONNX dependencies\nlet ort: any;\nlet transformers: any;\n\nasync function ensureOnnxDependencies() {\n  if (!ort) {\n    try {\n      const ortModule = await import('onnxruntime-node' as any);\n      ort = ortModule;\n    } catch (e) {\n      throw new Error('onnxruntime-node not installed. Run: npm install onnxruntime-node');\n    }\n  }\n  if (!transformers) {\n    try {\n      const transformersModule = await import('@xenova/transformers' as any);\n      transformers = transformersModule;\n      transformers.env.allowLocalModels = true;\n    } catch (e) {\n      throw new Error('@xenova/transformers not installed. Run: npm install @xenova/transformers');\n    }\n  }\n}\n\nexport interface ONNXConfig {\n  modelPath?: string;\n  modelId?: string; // HuggingFace model ID\n  executionProviders?: string[];\n  sessionOptions?: any;\n  maxTokens?: number;\n  temperature?: number;\n}\n\nexport class ONNXProvider implements LLMProvider {\n  name = 'onnx';\n  type = 'custom' as const;\n  supportsStreaming = true;\n  supportsTools = false;\n  supportsMCP = false;\n\n  private session: any = null;\n  private generator: any = null;\n  private config: ONNXConfig;\n  private executionProviders: string[] = [];\n\n  constructor(config: ONNXConfig = {}) {\n    this.config = {\n      modelId: config.modelId || 'Xenova/Phi-3-mini-4k-instruct',\n      maxTokens: config.maxTokens || 512,\n      temperature: config.temperature || 0.7,\n      ...config\n    };\n  }\n\n  /**\n   * Detect available execution providers\n   */\n  private async detectExecutionProviders(): Promise<string[]> {\n    const providers: string[] = [];\n\n    // Try CUDA for NVIDIA GPUs\n    try {\n      if (process.platform === 'linux') {\n        providers.push('cuda');\n        this.executionProviders.push('cuda');\n      }\n    } catch (e) {\n      // CUDA not available\n    }\n\n    // Try DirectML for Windows GPUs\n    try {\n      if (process.platform === 'win32') {\n        providers.push('dml');\n        this.executionProviders.push('dml');\n      }\n    } catch (e) {\n      // DirectML not available\n    }\n\n    // Always fallback to CPU\n    providers.push('cpu');\n    this.executionProviders.push('cpu');\n\n    console.log(`🔧 ONNX Execution Providers: ${this.executionProviders.join(', ')}`);\n\n    return providers;\n  }\n\n  /**\n   * Initialize ONNX session with model\n   */\n  private async initializeSession(): Promise<void> {\n    if (this.generator) return;\n\n    try {\n      await ensureOnnxDependencies();\n\n      console.log(`📦 Loading ONNX model: ${this.config.modelId}`);\n\n      // Use Transformers.js for easier model loading\n      this.generator = await transformers.pipeline(\n        'text-generation',\n        this.config.modelId,\n        {\n          quantized: true, // Use quantized models for better CPU performance\n        }\n      );\n\n      console.log(`✅ ONNX model loaded successfully`);\n\n    } catch (error) {\n      const providerError: ProviderError = {\n        name: 'ONNXInitError',\n        message: `Failed to initialize ONNX model: ${error}`,\n        provider: 'onnx',\n        retryable: false\n      };\n      throw providerError;\n    }\n  }\n\n  /**\n   * Format messages for model input\n   */\n  private formatMessages(messages: Message[]): string {\n    // Simple chat template for Phi-3\n    let prompt = '';\n\n    for (const msg of messages) {\n      const content = typeof msg.content === 'string'\n        ? msg.content\n        : msg.content.map(c => c.type === 'text' ? c.text : '').join('');\n\n      if (msg.role === 'user') {\n        prompt += `<|user|>\\n${content}<|end|>\\n`;\n      } else if (msg.role === 'assistant') {\n        prompt += `<|assistant|>\\n${content}<|end|>\\n`;\n      } else if (msg.role === 'system') {\n        prompt += `<|system|>\\n${content}<|end|>\\n`;\n      }\n    }\n\n    prompt += '<|assistant|>\\n';\n    return prompt;\n  }\n\n  /**\n   * Chat completion\n   */\n  async chat(params: ChatParams): Promise<ChatResponse> {\n    await this.initializeSession();\n\n    const startTime = Date.now();\n    const prompt = this.formatMessages(params.messages);\n\n    try {\n      const result = await this.generator(prompt, {\n        max_new_tokens: params.maxTokens || this.config.maxTokens,\n        temperature: params.temperature || this.config.temperature,\n        do_sample: true,\n        top_p: 0.9,\n      });\n\n      const generatedText = result[0].generated_text;\n\n      // Extract only the new assistant response\n      const assistantResponse = generatedText\n        .split('<|assistant|>')\n        .pop()\n        ?.split('<|end|>')[0]\n        ?.trim() || '';\n\n      const latency = Date.now() - startTime;\n\n      // Estimate token counts (rough approximation)\n      const inputTokens = Math.ceil(prompt.length / 4);\n      const outputTokens = Math.ceil(assistantResponse.length / 4);\n\n      const content: ContentBlock[] = [{\n        type: 'text',\n        text: assistantResponse\n      }];\n\n      return {\n        id: `onnx-${Date.now()}`,\n        model: this.config.modelId || 'onnx-model',\n        content,\n        stopReason: 'end_turn',\n        usage: {\n          inputTokens,\n          outputTokens\n        },\n        metadata: {\n          provider: 'onnx',\n          model: this.config.modelId,\n          latency,\n          cost: 0, // Local inference is free\n          executionProviders: this.executionProviders\n        }\n      };\n\n    } catch (error) {\n      const providerError: ProviderError = {\n        name: 'ONNXInferenceError',\n        message: `ONNX inference failed: ${error}`,\n        provider: 'onnx',\n        retryable: true\n      };\n      throw providerError;\n    }\n  }\n\n  /**\n   * Streaming generation\n   */\n  async *stream(params: ChatParams): AsyncGenerator<StreamChunk> {\n    await this.initializeSession();\n\n    const prompt = this.formatMessages(params.messages);\n\n    try {\n      // Note: Transformers.js doesn't natively support streaming\n      // We'll simulate it by yielding tokens as they're generated\n      const result = await this.generator(prompt, {\n        max_new_tokens: params.maxTokens || this.config.maxTokens,\n        temperature: params.temperature || this.config.temperature,\n        do_sample: true,\n        top_p: 0.9,\n      });\n\n      const generatedText = result[0].generated_text;\n      const assistantResponse = generatedText\n        .split('<|assistant|>')\n        .pop()\n        ?.split('<|end|>')[0]\n        ?.trim() || '';\n\n      // Simulate streaming by chunking the response\n      const words = assistantResponse.split(' ');\n      for (let i = 0; i < words.length; i++) {\n        const chunk = words[i] + (i < words.length - 1 ? ' ' : '');\n        yield {\n          type: 'content_block_delta',\n          delta: {\n            type: 'text_delta',\n            text: chunk\n          }\n        };\n\n        // Small delay to simulate real streaming\n        await new Promise(resolve => setTimeout(resolve, 10));\n      }\n\n      yield {\n        type: 'message_stop'\n      };\n\n    } catch (error) {\n      const providerError: ProviderError = {\n        name: 'ONNXStreamError',\n        message: `ONNX streaming failed: ${error}`,\n        provider: 'onnx',\n        retryable: true\n      };\n      throw providerError;\n    }\n  }\n\n  /**\n   * Validate capabilities\n   */\n  validateCapabilities(features: string[]): boolean {\n    const supported = ['chat', 'stream'];\n    return features.every(f => supported.includes(f));\n  }\n\n  /**\n   * Get model info\n   */\n  getModelInfo() {\n    return {\n      modelId: this.config.modelId,\n      executionProviders: this.executionProviders,\n      supportsGPU: this.executionProviders.includes('cuda') || this.executionProviders.includes('dml'),\n      initialized: this.generator !== null\n    };\n  }\n\n  /**\n   * Cleanup resources\n   */\n  async dispose(): Promise<void> {\n    if (this.generator) {\n      this.generator = null;\n    }\n    if (this.session) {\n      await this.session.release();\n      this.session = null;\n    }\n  }\n}\n"]}

package/dist/agentic-flow/src/router/providers/openrouter.d.ts

@@ -0,0 +1,21 @@
+import { LLMProvider, ChatParams, ChatResponse, StreamChunk, ProviderConfig } from '../types.js';
+export declare class OpenRouterProvider implements LLMProvider {
+    name: string;
+    type: "openrouter";
+    supportsStreaming: boolean;
+    supportsTools: boolean;
+    supportsMCP: boolean;
+    private client;
+    private config;
+    constructor(config: ProviderConfig);
+    validateCapabilities(features: string[]): boolean;
+    chat(params: ChatParams): Promise<ChatResponse>;
+    stream(params: ChatParams): AsyncGenerator<StreamChunk>;
+    private formatRequest;
+    private formatResponse;
+    private formatStreamChunk;
+    private mapFinishReason;
+    private calculateCost;
+    private handleError;
+}
+//# sourceMappingURL=openrouter.d.ts.map

package/dist/agentic-flow/src/router/providers/openrouter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"openrouter.d.ts","sourceRoot":"","sources":["../../../../../src/router/providers/openrouter.ts"],"names":[],"mappings":"AAEA,OAAO,EACL,WAAW,EACX,UAAU,EACV,YAAY,EACZ,WAAW,EACX,cAAc,EAIf,MAAM,aAAa,CAAC;AAGrB,qBAAa,kBAAmB,YAAW,WAAW;IACpD,IAAI,SAAgB;IACpB,IAAI,EAAG,YAAY,CAAU;IAC7B,iBAAiB,UAAQ;IACzB,aAAa,UAAQ;IACrB,WAAW,UAAS;IAEpB,OAAO,CAAC,MAAM,CAAgB;IAC9B,OAAO,CAAC,MAAM,CAAiB;gBAEnB,MAAM,EAAE,cAAc;IAmBlC,oBAAoB,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,OAAO;IAK3C,IAAI,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAAC,YAAY,CAAC;IAY9C,MAAM,CAAC,MAAM,EAAE,UAAU,GAAG,cAAc,CAAC,WAAW,CAAC;IAqC9D,OAAO,CAAC,aAAa;IA2ErB,OAAO,CAAC,cAAc;IA6CtB,OAAO,CAAC,iBAAiB;IAwCzB,OAAO,CAAC,eAAe;IASvB,OAAO,CAAC,aAAa;IAWrB,OAAO,CAAC,WAAW;CAWpB"}

package/dist/agentic-flow/src/router/providers/openrouter.js

@@ -0,0 +1,246 @@
+// OpenRouter provider implementation
+import axios from 'axios';
+import { mapModelId } from '../model-mapping.js';
+export class OpenRouterProvider {
+    name = 'openrouter';
+    type = 'openrouter';
+    supportsStreaming = true;
+    supportsTools = true;
+    supportsMCP = false; // Requires translation
+    client;
+    config;
+    constructor(config) {
+        this.config = config;
+        if (!config.apiKey) {
+            throw new Error('OpenRouter API key is required');
+        }
+        this.client = axios.create({
+            baseURL: config.baseUrl || 'https://openrouter.ai/api/v1',
+            headers: {
+                'Authorization': `Bearer ${config.apiKey}`,
+                'HTTP-Referer': 'https://github.com/ruvnet/agentic-flow',
+                'X-Title': 'Agentic Flow Multi-Model Router',
+                'Content-Type': 'application/json'
+            },
+            timeout: config.timeout || 180000
+        });
+    }
+    validateCapabilities(features) {
+        const supported = ['chat', 'streaming', 'tools'];
+        return features.every(f => supported.includes(f));
+    }
+    async chat(params) {
+        try {
+            const requestBody = this.formatRequest(params);
+            const response = await this.client.post('/chat/completions', requestBody);
+            return this.formatResponse(response.data, params.model);
+        }
+        catch (error) {
+            throw this.handleError(error);
+        }
+    }
+    async *stream(params) {
+        try {
+            const requestBody = this.formatRequest(params, true);
+            const response = await this.client.post('/chat/completions', requestBody, {
+                responseType: 'stream'
+            });
+            for await (const chunk of response.data) {
+                const lines = chunk.toString().split('\n').filter((line) => line.trim() !== '');
+                for (const line of lines) {
+                    if (line.startsWith('data: ')) {
+                        const data = line.slice(6);
+                        if (data === '[DONE]') {
+                            yield { type: 'message_stop' };
+                            return;
+                        }
+                        try {
+                            const parsed = JSON.parse(data);
+                            const streamChunk = this.formatStreamChunk(parsed);
+                            if (streamChunk) {
+                                yield streamChunk;
+                            }
+                        }
+                        catch (e) {
+                            // Skip invalid JSON
+                        }
+                    }
+                }
+            }
+        }
+        catch (error) {
+            throw this.handleError(error);
+        }
+    }
+    formatRequest(params, stream = false) {
+        const messages = params.messages.map(msg => ({
+            role: msg.role,
+            content: typeof msg.content === 'string'
+                ? msg.content
+                : msg.content.map(block => {
+                    if (block.type === 'text')
+                        return { type: 'text', text: block.text };
+                    if (block.type === 'tool_use')
+                        return {
+                            type: 'function',
+                            function: {
+                                name: block.name,
+                                arguments: JSON.stringify(block.input)
+                            }
+                        };
+                    if (block.type === 'tool_result')
+                        return {
+                            type: 'function_result',
+                            content: block.content,
+                            is_error: block.is_error
+                        };
+                    return block;
+                })
+        }));
+        // Map model ID to OpenRouter format
+        const openrouterModel = mapModelId(params.model, 'openrouter');
+        const body = {
+            model: openrouterModel,
+            messages,
+            temperature: params.temperature ?? 0.7,
+            max_tokens: params.maxTokens,
+            stream
+        };
+        // Add tools if provided
+        if (params.tools && params.tools.length > 0) {
+            body.tools = params.tools.map(tool => ({
+                type: 'function',
+                function: {
+                    name: tool.name,
+                    description: tool.description,
+                    parameters: tool.input_schema
+                }
+            }));
+            if (params.toolChoice) {
+                if (params.toolChoice === 'auto' || params.toolChoice === 'none') {
+                    body.tool_choice = params.toolChoice;
+                }
+                else if (typeof params.toolChoice === 'object') {
+                    body.tool_choice = {
+                        type: 'function',
+                        function: { name: params.toolChoice.name }
+                    };
+                }
+            }
+        }
+        // Add OpenRouter-specific preferences
+        if (this.config.preferences) {
+            if (this.config.preferences.requireParameters) {
+                body.require_parameters = true;
+            }
+            if (this.config.preferences.dataCollection) {
+                body['X-Data-Collection'] = this.config.preferences.dataCollection;
+            }
+            if (this.config.preferences.order) {
+                body.provider = {
+                    order: this.config.preferences.order
+                };
+            }
+        }
+        return body;
+    }
+    formatResponse(data, model) {
+        const choice = data.choices[0];
+        const message = choice.message;
+        const content = [];
+        // Handle text content
+        if (message.content) {
+            content.push({
+                type: 'text',
+                text: message.content
+            });
+        }
+        // Handle tool calls
+        if (message.tool_calls) {
+            for (const toolCall of message.tool_calls) {
+                content.push({
+                    type: 'tool_use',
+                    id: toolCall.id,
+                    name: toolCall.function.name,
+                    input: JSON.parse(toolCall.function.arguments)
+                });
+            }
+        }
+        const stopReason = this.mapFinishReason(choice.finish_reason);
+        return {
+            id: data.id,
+            model: data.model || model,
+            content,
+            stopReason,
+            usage: {
+                inputTokens: data.usage?.prompt_tokens || 0,
+                outputTokens: data.usage?.completion_tokens || 0
+            },
+            metadata: {
+                provider: 'openrouter',
+                cost: this.calculateCost(data.usage),
+                latency: 0 // Will be set by router
+            }
+        };
+    }
+    formatStreamChunk(data) {
+        const choice = data.choices?.[0];
+        if (!choice)
+            return null;
+        const delta = choice.delta;
+        if (delta?.content) {
+            return {
+                type: 'content_block_delta',
+                delta: {
+                    type: 'text_delta',
+                    text: delta.content
+                }
+            };
+        }
+        if (delta?.tool_calls) {
+            const toolCall = delta.tool_calls[0];
+            return {
+                type: 'content_block_delta',
+                delta: {
+                    type: 'input_json_delta',
+                    partial_json: toolCall.function?.arguments || ''
+                }
+            };
+        }
+        if (choice.finish_reason) {
+            return {
+                type: 'message_stop',
+                usage: data.usage ? {
+                    inputTokens: data.usage.prompt_tokens || 0,
+                    outputTokens: data.usage.completion_tokens || 0
+                } : undefined
+            };
+        }
+        return null;
+    }
+    mapFinishReason(reason) {
+        switch (reason) {
+            case 'stop': return 'end_turn';
+            case 'length': return 'max_tokens';
+            case 'tool_calls': return 'tool_use';
+            default: return 'end_turn';
+        }
+    }
+    calculateCost(usage) {
+        if (!usage)
+            return 0;
+        // OpenRouter pricing varies by model
+        // Using average pricing: $0.01/1K input, $0.03/1K output
+        const inputCost = (usage.prompt_tokens || 0) * 0.00001;
+        const outputCost = (usage.completion_tokens || 0) * 0.00003;
+        return inputCost + outputCost;
+    }
+    handleError(error) {
+        const providerError = new Error(error.response?.data?.error?.message || error.message || 'OpenRouter request failed');
+        providerError.provider = 'openrouter';
+        providerError.statusCode = error.response?.status;
+        providerError.retryable = error.response?.status >= 500 || error.code === 'ECONNABORTED';
+        return providerError;
+    }
+}
+//# sourceMappingURL=openrouter.js.map