📚 API Reference¶

Complete reference for NeuroLink's TypeScript API.

Core Functions¶

`createBestAIProvider(requestedProvider?, modelName?)`¶

Creates the best available AI provider based on environment configuration and provider availability. All providers inherit from BaseProvider and include built-in tool support.

function createBestAIProvider(
  requestedProvider?: string,
  modelName?: string,
): AIProvider;

Parameters:

requestedProvider (optional): Preferred provider name ('openai', 'bedrock', 'sagemaker', 'vertex', 'anthropic', 'azure', 'google-ai', 'huggingface', 'ollama', 'mistral', 'litellm', or 'auto')
modelName (optional): Specific model to use

Returns: AIProvider instance

Examples:

import { createBestAIProvider } from "@juspay/neurolink";

// Auto-select best available provider
const provider = createBestAIProvider();

// Prefer specific provider
const openaiProvider = createBestAIProvider("openai");

// Prefer specific provider and model
const googleProvider = createBestAIProvider("google-ai", "gemini-2.5-flash");

// Use more comprehensive model for detailed responses
const detailedProvider = createBestAIProvider("google-ai", "gemini-2.5-pro");

// Use LiteLLM proxy for access to 100+ models
const litellmProvider = createBestAIProvider("litellm", "openai/gpt-4o");
const claudeProvider = createBestAIProvider(
  "litellm",
  "anthropic/claude-3-5-sonnet",
);

// Use Amazon SageMaker for custom deployed models
const sagemakerProvider = createBestAIProvider("sagemaker", "my-custom-model");

`createAIProviderWithFallback(primary, fallback, modelName?)`¶

Creates a provider with automatic fallback mechanism.

function createAIProviderWithFallback(
  primary: string,
  fallback: string,
  modelName?: string,
): { primary: AIProvider; fallback: AIProvider };

Parameters:

primary: Primary provider name
fallback: Fallback provider name
modelName (optional): Model name for both providers

Returns: Object with primary and fallback provider instances

Example:

import { createAIProviderWithFallback } from "@juspay/neurolink";

const { primary, fallback } = createAIProviderWithFallback("bedrock", "openai");

try {
  const result = await primary.generate({ input: { text: "Hello AI!" } });
} catch (error) {
  console.log("Primary failed, trying fallback...");
  const result = await fallback.generate({ input: { text: "Hello AI!" } });
}

BaseProvider Class¶

All AI providers inherit from BaseProvider, which provides unified tool support and consistent behavior across all providers.

Key Features¶

Automatic Tool Support: All providers include six built-in tools without additional configuration
Unified Interface: Consistent generate() and stream() methods across all providers
Analytics & Evaluation: Built-in support for usage analytics and quality evaluation
Error Handling: Standardized error handling and recovery

Built-in Tools¶

Every provider automatically includes these tools:

interface BuiltInTools {
  getCurrentTime: {
    description: "Get the current date and time";
    parameters: { timezone?: string };
  };
  readFile: {
    description: "Read contents of a file";
    parameters: { path: string };
  };
  listDirectory: {
    description: "List contents of a directory";
    parameters: { path: string };
  };
  calculateMath: {
    description: "Perform mathematical calculations";
    parameters: { expression: string };
  };
  writeFile: {
    description: "Write content to a file";
    parameters: { path: string; content: string };
  };
  searchFiles: {
    description: "Search for files by pattern";
    parameters: { pattern: string; path?: string };
  };
}

Example Usage¶

// All providers automatically have tool support
const provider = createBestAIProvider("openai");

// Tools are used automatically when appropriate
const result = await provider.generate({
  input: { text: "What time is it?" },
});
// Result will use getCurrentTime tool automatically

// Disable tools if needed
const resultNoTools = await provider.generate({
  input: { text: "What time is it?" },
  disableTools: true,
});
// Result will use training data instead of real-time tools

// LiteLLM Provider - Access 100+ Models
const litellmProvider = createBestAIProvider("litellm");
const litellmResult = await litellmProvider.generate({
  input: { text: "Hello from multiple AI models!" },
  // Automatically uses configured LiteLLM model (e.g., gemini-2.5-pro, claude-sonnet-4)
});

AIProviderFactory¶

Factory class for creating specific provider instances with BaseProvider inheritance.

`createProvider(providerName, modelName?)`¶

Creates a specific provider instance.

static createProvider(
  providerName: string,
  modelName?: string
): AIProvider

Parameters:

providerName: Provider name ('openai', 'bedrock', 'sagemaker', 'vertex', 'anthropic', 'azure', 'google-ai', 'huggingface', 'ollama', 'mistral', 'litellm')
modelName (optional): Specific model to use

Returns: AIProvider instance

Examples:

import { AIProviderFactory } from "@juspay/neurolink";

// Create specific providers
const openai = AIProviderFactory.createProvider("openai", "gpt-4o");
const bedrock = AIProviderFactory.createProvider(
  "bedrock",
  "claude-3-7-sonnet",
);
const sagemaker = AIProviderFactory.createProvider("sagemaker", "my-endpoint");
const vertex = AIProviderFactory.createProvider("vertex", "gemini-2.5-flash");

// LiteLLM Provider - Access multiple models through proxy
const litellm = AIProviderFactory.createProvider("litellm", "gemini-2.5-pro");
const claudeLiteLLM = AIProviderFactory.createProvider(
  "litellm",
  "claude-sonnet-4",
);
const llamaLiteLLM = AIProviderFactory.createProvider(
  "litellm",
  "llama4-scout",
);

// Use default models
const defaultOpenAI = AIProviderFactory.createProvider("openai");

`createProviderWithFallback(primary, fallback, modelName?)`¶

Creates provider with fallback (same as standalone function).

static createProviderWithFallback(
  primary: string,
  fallback: string,
  modelName?: string
): { primary: AIProvider; fallback: AIProvider }

AIProvider Interface¶

All providers implement the AIProvider interface with these methods:

interface AIProvider {
  generate(options: GenerateOptions): Promise<GenerateResult>;
  stream(options: StreamOptions): Promise<StreamResult>; // PRIMARY streaming method

  // Legacy compatibility
  gen?(options: GenerateOptions): Promise<GenerateResult>;
}

🔗 CLI-SDK Consistency¶

All providers now include method aliases that match CLI command names for consistent developer experience:

generate() - Primary method for content generation (matches neurolink generate CLI command)
gen() - Short alias for generate() (matches neurolink gen CLI command)

🆕 NeuroLink Class API¶

`addMCPServer(serverId, config)`¶

NEW! Programmatically add MCP servers at runtime for dynamic tool ecosystem management.

async addMCPServer(
  serverId: string,
  config: {
    command: string;
    args?: string[];
    env?: Record<string, string>;
    cwd?: string;
  }
): Promise<void>

Parameters:

serverId: Unique identifier for the MCP server
config.command: Command to execute (e.g., 'npx', 'node')
config.args: Optional command arguments array
config.env: Optional environment variables
config.cwd: Optional working directory

Examples:

import { NeuroLink } from "@juspay/neurolink";
const neurolink = new NeuroLink();

// Add Bitbucket integration
await neurolink.addMCPServer("bitbucket", {
  command: "npx",
  args: ["-y", "@nexus2520/bitbucket-mcp-server"],
  env: {
    BITBUCKET_USERNAME: "your-username",
    BITBUCKET_APP_PASSWORD: "your-app-password",
  },
});

// Add custom database server
await neurolink.addMCPServer("database", {
  command: "node",
  args: ["./custom-db-mcp-server.js"],
  env: { DB_CONNECTION_STRING: "postgresql://..." },
  cwd: "/path/to/server",
});

// Add any MCP-compatible server
await neurolink.addMCPServer("slack", {
  command: "npx",
  args: ["-y", "@slack/mcp-server"],
  env: { SLACK_BOT_TOKEN: "xoxb-..." },
});

Use Cases:

External service integration (Bitbucket, Slack, Jira)
Custom tool development
Dynamic workflow configuration
Enterprise application toolchain management

`getMCPStatus()`¶

Get current MCP server status and statistics.

async getMCPStatus(): Promise<{
  totalServers: number;
  availableServers: number;
  totalTools: number;
}>

`getUnifiedRegistry()`¶

Access the unified MCP registry for advanced server management.

getUnifiedRegistry(): UnifiedMCPRegistry

These methods have identical signatures and behavior to generate().

// All three methods are equivalent:
const result1 = await provider.generate({ input: { text: "Hello" } });
const result2 = await provider.generate({ input: { text: "Hello" } });
const result3 = await provider.gen({ input: { text: "Hello" } });

`generate(options)`¶

Generate text content synchronously.

async generate(options: GenerateOptions): Promise<GenerateResult>

Parameters:

interface GenerateOptions {
  input: { text: string };
  temperature?: number;
  maxTokens?: number;
  systemPrompt?: string;
  schema?: any; // For structured output
  timeout?: number | string; // Timeout in ms or human-readable format (e.g., '30s', '2m', '1h')
  disableTools?: boolean; // Disable tool usage for this request
  enableAnalytics?: boolean; // Enable usage analytics
  enableEvaluation?: boolean; // Enable AI quality scoring
  context?: Record<string, any>; // Custom context for analytics
}

Returns:

interface GenerateResult {
  content: string;
  provider: string;
  model: string;
  usage?: {
    promptTokens: number;
    completionTokens: number;
    totalTokens: number;
  };
  responseTime?: number;

  // 🆕 NEW: AI Enhancement Features
  analytics?: {
    provider: string;
    model: string;
    tokens: { input: number; output: number; total: number };
    cost?: number;
    responseTime: number;
    context?: Record<string, any>;
  };

  evaluation?: {
    relevanceScore: number; // 1-10 scale
    accuracyScore: number; // 1-10 scale
    completenessScore: number; // 1-10 scale
    overallScore: number; // 1-10 scale
    alertLevel?: string; // 'none', 'low', 'medium', 'high'
    reasoning?: string; // AI reasoning for the evaluation

    // Enhanced evaluation fields (when available)
    domainAlignment?: number; // 1-10 scale
    terminologyAccuracy?: number; // 1-10 scale
    toolEffectiveness?: number; // 1-10 scale
    alertSeverity?: string; // Legacy field
    contextUtilization?: {
      conversationUsed: boolean;
      toolsUsed: boolean;
      domainKnowledgeUsed: boolean;
    };
  };
}

🆕 Enterprise Configuration Interfaces¶

`NeuroLinkConfig`¶

Main configuration interface for enterprise features:

interface NeuroLinkConfig {
  providers: ProviderConfig;
  performance: PerformanceConfig;
  analytics: AnalyticsConfig;
  backup: BackupConfig;
  validation: ValidationConfig;
}

`ExecutionContext`¶

Rich context interface for all MCP operations:

interface ExecutionContext {
  sessionId?: string;
  userId?: string;
  aiProvider?: string;
  permissions?: string[];
  cacheOptions?: CacheOptions;
  fallbackOptions?: FallbackOptions;
  metadata?: Record<string, unknown>;
  priority?: "low" | "normal" | "high";
  timeout?: number;
  retries?: number;
  correlationId?: string;
  requestId?: string;
  userAgent?: string;
  clientVersion?: string;
  environment?: string;
}

`ToolInfo`¶

Comprehensive tool metadata interface:

interface ToolInfo {
  name: string;
  description?: string;
  serverId?: string;
  category?: string;
  version?: string;
  parameters?: unknown;
  capabilities?: string[];
  lastUsed?: Date;
  usageCount?: number;
  averageExecutionTime?: number;
}

`ConfigUpdateOptions`¶

Flexible configuration update options:

interface ConfigUpdateOptions {
  createBackup?: boolean;
  validateBeforeUpdate?: boolean;
  mergeStrategy?: "replace" | "merge" | "deep-merge";
  backupRetention?: number;
  onValidationError?: (errors: ValidationError[]) => void;
  onBackupCreated?: (backupPath: string) => void;
}

`McpRegistry`¶

Registry interface with optional methods for maximum flexibility:

interface McpRegistry {
  registerServer?(serverId: string, config?: unknown, context?: ExecutionContext): Promise<void>;
  executeTool?<T>(toolName: string, args?: unknown, context?: ExecutionContext): Promise<T>;
  listTools?(context?: ExecutionContext): Promise<ToolInfo[]>;
  getStats?(): Record<string, { count: number; averageTime: number; totalTime: number }>;
  unregisterServer?(serverId: string): Promise<void>;
  getServerInfo?(serverId: string): Promise<unknown>;
}
}

🌐 Enterprise Real-time Services API¶

`createEnhancedChatService(options)`¶

Creates an enhanced chat service with WebSocket and SSE support for real-time applications.

function createEnhancedChatService(options: {
  provider: AIProvider;
  enableSSE?: boolean;
  enableWebSocket?: boolean;
  streamingConfig?: StreamingConfig;
}): EnhancedChatService;

Parameters:

interface EnhancedChatServiceOptions {
  provider: AIProvider; // AI provider instance
  enableSSE?: boolean; // Enable Server-Sent Events (default: true)
  enableWebSocket?: boolean; // Enable WebSocket support (default: false)
  streamingConfig?: {
    bufferSize?: number; // Buffer size in bytes (default: 8192)
    compressionEnabled?: boolean; // Enable compression (default: true)
    latencyTarget?: number; // Target latency in ms (default: 100)
  };
}

Returns: EnhancedChatService instance

Example:

import {
  createEnhancedChatService,
  createBestAIProvider,
} from "@juspay/neurolink";

const provider = await createBestAIProvider();
const chatService = createEnhancedChatService({
  provider,
  enableWebSocket: true,
  enableSSE: true,
  streamingConfig: {
    bufferSize: 4096,
    compressionEnabled: true,
    latencyTarget: 50, // 50ms target latency
  },
});

// Stream chat with enhanced capabilities
await chatService.streamChat({
  prompt: "Generate a story",
  onChunk: (chunk) => console.log(chunk),
  onComplete: (result) => console.log("Complete:", result),
});

`NeuroLinkWebSocketServer`¶

Professional-grade WebSocket server for real-time AI applications.

class NeuroLinkWebSocketServer {
  constructor(options?: WebSocketOptions);
  joinRoom(connectionId: string, roomId: string): boolean;
  broadcastToRoom(roomId: string, message: WebSocketMessage): void;
  createStreamingChannel(
    connectionId: string,
    channelId: string,
  ): StreamingChannel;
  sendMessage(connectionId: string, message: WebSocketMessage): boolean;
  on(event: string, handler: Function): void;
}

Constructor Options:

interface WebSocketOptions {
  port?: number; // Server port (default: 8080)
  maxConnections?: number; // Max concurrent connections (default: 1000)
  heartbeatInterval?: number; // Heartbeat interval in ms (default: 30000)
  enableCompression?: boolean; // Enable WebSocket compression (default: true)
  bufferSize?: number; // Message buffer size (default: 8192)
}

Example:

import { NeuroLinkWebSocketServer } from "@juspay/neurolink";

const wsServer = new NeuroLinkWebSocketServer({
  port: 8080,
  maxConnections: 1000,
  enableCompression: true,
});

// Handle connections
wsServer.on("connection", ({ connectionId, userAgent }) => {
  console.log(`New connection: ${connectionId}`);
  wsServer.joinRoom(connectionId, "general-chat");
});

// Handle chat messages
wsServer.on("chat-message", async ({ connectionId, message }) => {
  // Process with AI and broadcast response
  const aiResponse = await processWithAI(message.data.prompt);
  wsServer.broadcastToRoom("general-chat", {
    type: "ai-response",
    data: { text: aiResponse },
  });
});

📊 Enterprise Telemetry API¶

`initializeTelemetry(config)`¶

Initializes enterprise telemetry with OpenTelemetry integration. Zero overhead when disabled.

function initializeTelemetry(config: TelemetryConfig): TelemetryResult;

Parameters:

interface TelemetryConfig {
  serviceName: string; // Service name for telemetry
  endpoint?: string; // OpenTelemetry endpoint
  enableTracing?: boolean; // Enable distributed tracing (default: true)
  enableMetrics?: boolean; // Enable metrics collection (default: true)
  enableLogs?: boolean; // Enable log collection (default: true)
  samplingRate?: number; // Trace sampling rate 0-1 (default: 0.1)
}

Returns:

interface TelemetryResult {
  success: boolean;
  tracingEnabled: boolean;
  metricsEnabled: boolean;
  logsEnabled: boolean;
  endpoint?: string;
  error?: string;
}

Example:

import { initializeTelemetry } from "@juspay/neurolink";

const telemetry = initializeTelemetry({
  serviceName: "my-ai-application",
  endpoint: "http://localhost:4318",
  enableTracing: true,
  enableMetrics: true,
  enableLogs: true,
  samplingRate: 0.1, // Sample 10% of traces
});

if (telemetry.success) {
  console.log("Telemetry initialized successfully");
} else {
  console.error("Telemetry initialization failed:", telemetry.error);
}

`getTelemetryStatus()`¶

Returns current telemetry status and configuration.

function getTelemetryStatus(): Promise<TelemetryStatus>;

Returns:

interface TelemetryStatus {
  enabled: boolean; // Whether telemetry is active
  endpoint?: string; // Current endpoint
  service: string; // Service name
  version: string; // NeuroLink version
  features: {
    tracing: boolean;
    metrics: boolean;
    logs: boolean;
  };
  stats?: {
    tracesCollected: number;
    metricsCollected: number;
    logsCollected: number;
  };
}

Example:

import { getTelemetryStatus } from "@juspay/neurolink";

const status = await getTelemetryStatus();
console.log("Telemetry enabled:", status.enabled);
console.log("Service:", status.service);
console.log("Features:", status.features);

if (status.stats) {
  console.log("Traces collected:", status.stats.tracesCollected);
  console.log("Metrics collected:", status.stats.metricsCollected);
}

🔧 Enhanced Generation Options¶

The base GenerateOptions interface now supports enterprise features:

interface GenerateOptions {
  input: { text: string };
  temperature?: number;
  maxTokens?: number;
  systemPrompt?: string;
  schema?: any;
  timeout?: number | string;
  disableTools?: boolean; // Disable tool usage for this request

  // 🆕 NEW: AI Enhancement Features
  enableAnalytics?: boolean; // Enable usage analytics
  enableEvaluation?: boolean; // Enable AI quality scoring
  context?: Record<string, any>; // Custom context for analytics
}

Enhanced Usage Example:

const result = await provider.generate({
  input: { text: "Write a business proposal" },
  enableAnalytics: true,
  enableEvaluation: true,
  context: {
    userId: "12345",
    session: "business-meeting",
    department: "sales",
  },
});

// Access enhancement data
console.log("📊 Analytics:", result.analytics);
// { provider: 'openai', model: 'gpt-4o', tokens: {...}, cost: 0.02, responseTime: 2340 }

console.log("⭐ Evaluation:", result.evaluation);
// { relevanceScore: 9, accuracyScore: 8, completenessScore: 9, overallScore: 8.7 }

Example:

const result = await provider.generate({
  input: { text: "Explain quantum computing in simple terms" },
  temperature: 0.7,
  maxTokens: 500,
  systemPrompt: "You are a helpful science teacher",
});

console.log(result.content);
console.log(`Used ${result.usage?.totalTokens} tokens`);
console.log(`Provider: ${result.provider}, Model: ${result.model}`);

`stream(options)` - Recommended for New Code¶

Generate content with streaming responses using future-ready multi-modal interface.

async stream(options: StreamOptions): Promise<StreamResult>

Parameters:

interface StreamOptions {
  input: { text: string }; // Current scope: text input (future: multi-modal)
  output?: {
    format?: "text" | "structured" | "json";
    streaming?: {
      chunkSize?: number;
      bufferSize?: number;
      enableProgress?: boolean;
    };
  };
  provider?: string;
  model?: string;
  temperature?: number;
  maxTokens?: number;
  timeout?: number | string;
}

Returns:

interface StreamResult {
  stream: AsyncIterable<{ content: string }>;
  provider?: string;
  model?: string;
  metadata?: {
    streamId?: string;
    startTime?: number;
    totalChunks?: number;
  };
}

Example:

const result = await provider.stream({
  input: { text: "Write a story about AI and humanity" },
  provider: "openai",
  temperature: 0.8,
});

for await (const chunk of result.stream) {
  process.stdout.write(chunk.content);
}

Flexible Parameter Support¶

NeuroLink supports both object-based and string-based parameters for convenience:

// Object format (recommended for complex options)
const result1 = await provider.generate({
  input: { text: "Hello" },
  temperature: 0.7,
  maxTokens: 100,
});

// String format (convenient for simple prompts)
const result2 = await provider.generate({ input: { text: "Hello" } });

Using Timeouts¶

NeuroLink supports flexible timeout configuration for all AI operations:

// Numeric milliseconds
const result1 = await provider.generate({
  input: { text: "Write a story" },
  timeout: 30000, // 30 seconds
});

// Human-readable formats
const result2 = await provider.generate({
  input: { text: "Complex calculation" },
  timeout: "2m", // 2 minutes
});

// Streaming with longer timeout
const stream = await provider.stream({ input: { text:
  prompt: "Generate long content",
  timeout: "5m", // 5 minutes for streaming
});

// Provider-specific default timeouts
const provider = createBestAIProvider("ollama"); // Uses 5m default timeout

Supported Timeout Formats:

Milliseconds: 5000, 30000
Seconds: '30s', '1.5s'
Minutes: '2m', '0.5m'
Hours: '1h', '0.5h'

Usage Examples¶

Basic Usage¶

import { createBestAIProvider } from "@juspay/neurolink";

// Simple text generation
const provider = createBestAIProvider();
const result = await provider.generate({
  input: { text: "Write a haiku about coding" },
});
console.log(result.content);

Dynamic Model Usage (v1.8.0+)¶

import { AIProviderFactory, DynamicModelRegistry } from "@juspay/neurolink";

// Initialize factory and registry
const factory = new AIProviderFactory();
const registry = new DynamicModelRegistry();

// Use model aliases for convenient access
const provider1 = await factory.createProvider({
  provider: "anthropic",
  model: "claude-latest", // Auto-resolves to latest Claude model
});

// Capability-based model selection
const provider2 = await factory.createProvider({
  provider: "auto",
  capability: "vision", // Automatically selects best vision model
  optimizeFor: "cost", // Prefer cost-effective options
});

// Advanced model resolution
const bestCodingModel = await registry.findBestModel({
  capability: "code",
  maxPrice: 0.005, // Max $0.005 per 1K tokens
  provider: "anthropic", // Prefer Anthropic models
});

console.log(
  `Selected: ${bestCodingModel.modelId} (${bestCodingModel.reasoning})`,
);

Cost-Optimized Generation¶

import { DynamicModelRegistry } from "@juspay/neurolink";

const registry = new DynamicModelRegistry();

// Get the cheapest model for general tasks
const cheapestModel = await registry.getCheapestModel("general");
const provider = await factory.createProvider({
  provider: cheapestModel.provider,
  model: cheapestModel.id,
});

// Generate text with cost optimization
const result = await provider.generate({
  input: { text: "Summarize the benefits of renewable energy" },
  maxTokens: 200, // Control output length for cost
});

console.log(
  `Generated with ${result.model} - Cost: $${calculateCost(result.usage, cheapestModel.pricing)}`,
);

Vision Capabilities with Dynamic Selection¶

// Automatically select best vision model
const visionProvider = await factory.createProvider({
  capability: "vision",
  optimizeFor: "quality", // Prefer highest quality vision model
});

const result = await visionProvider.generate({
  input: { text: "Describe what you see in this image" },
  images: ["data:image/jpeg;base64,/9j/4AAQSkZJRgABA..."], // Base64 image
  maxTokens: 500,
});

Function Calling with Smart Model Selection¶

// Select model optimized for function calling
const functionProvider = await factory.createProvider({
  capability: "functionCalling",
  optimizeFor: "speed", // Fast function execution
});

const result = await functionProvider.generate({
  input: { text: "What's the weather in San Francisco?" },
  schema: {
    type: "object",
    properties: {
      location: { type: "string" },
      temperature: { type: "number" },
      conditions: { type: "string" },
    },
  },
});

console.log(JSON.parse(result.content)); // Structured weather data

Model Discovery and Search¶

import { DynamicModelRegistry } from "@juspay/neurolink";

const registry = new DynamicModelRegistry();

// Search for vision models under $0.001 per 1K tokens
const affordableVisionModels = await registry.searchModels({
  capability: "vision",
  maxPrice: 0.001,
  excludeDeprecated: true,
});

console.log("Affordable Vision Models:");
affordableVisionModels.forEach((model) => {
  console.log(`- ${model.name}: $${model.pricing.input}/1K tokens`);
});

// Get all models from a specific provider
const anthropicModels = await registry.searchModels({
  provider: "anthropic",
});

// Resolve aliases to actual model IDs
const resolvedModel = await registry.resolveModel("claude-latest");
console.log(`claude-latest resolves to: ${resolvedModel}`);

Streaming with Dynamic Models¶

// Use fastest model for streaming
const streamingProvider = await factory.createProvider({
  model: "fastest", // Alias for fastest available model
});

const stream = await streamingProvider.stream({
  input: { text:
  prompt: "Write a story about space exploration",
  maxTokens: 1000,
});

// Process streaming response
for await (const chunk of stream.textStream) {
  process.stdout.write(chunk);
}

Provider Fallback with Dynamic Models¶

// Primary: Best quality model, Fallback: Fastest cheap model
const primaryProvider = await factory.createProvider({
  provider: "anthropic",
  model: "claude-latest",
});

const fallbackProvider = await factory.createProvider({
  model: "fastest",
});

try {
  const result = await primaryProvider.generate({
    input: { text: "Complex reasoning task" },
  });
  console.log(result.content);
} catch (error) {
  console.log("Primary failed, using fallback...");
  const result = await fallbackProvider.generate({
    input: { text: "Complex reasoning task" },
  });
  console.log(result.content);
}

Supported Models¶

OpenAI Models¶

type OpenAIModel =
  | "gpt-4o" // Default - Latest multimodal model
  | "gpt-4o-mini" // Cost-effective variant
  | "gpt-4-turbo"; // High-performance model

Amazon Bedrock Models¶

type BedrockModel =
  | "claude-3-7-sonnet" // Default - Latest Claude model
  | "claude-3-5-sonnet" // Previous generation
  | "claude-3-haiku"; // Fast, lightweight model

Note: Bedrock requires full inference profile ARNs in environment variables.

Amazon SageMaker Models¶

type SageMakerModel = string; // Any custom model deployed to SageMaker endpoint
// Model names are user-defined based on endpoint configuration
// Examples:
// - 'my-custom-llm' (your custom fine-tuned model)
// - 'company-domain-model' (domain-specific model)
// - 'multilingual-model' (custom multilingual model)

Note: SageMaker supports any custom model you deploy. Model names are determined by your endpoint configuration and can be any identifier you choose.

Google Vertex AI Models¶

type VertexModel =
  | "gemini-2.5-flash" // Default - Fast, efficient
  | "claude-sonnet-4@20250514"; // High-quality reasoning

Google AI Studio Models¶

type GoogleAIModel =
  | "gemini-2.5-pro" // Default - Latest Gemini Pro
  | "gemini-2.5-flash"; // Fast, efficient responses

Azure OpenAI Models¶

type AzureModel = string; // Deployment-specific models
// Common deployments:
// - 'gpt-4o' (default)
// - 'gpt-4-turbo'
// - 'gpt-35-turbo'

Hugging Face Models¶

type HuggingFaceModel = string; // Any model from Hugging Face Hub
// Popular models:
// - 'microsoft/DialoGPT-medium' (default)
// - 'gpt2'
// - 'distilgpt2'
// - 'EleutherAI/gpt-neo-2.7B'

Ollama Models¶

type OllamaModel = string; // Any locally installed model
// Popular models:
// - 'llama2' (default)
// - 'codellama'
// - 'mistral'
// - 'vicuna'

Mistral AI Models¶

type MistralModel =
  | "mistral-tiny"
  | "mistral-small" // Default
  | "mistral-medium"
  | "mistral-large";

LiteLLM Models¶

type LiteLLMModel = string; // Uses provider/model format
// Popular models:
// - 'openai/gpt-4o' (default: openai/gpt-4o-mini)
// - 'anthropic/claude-3-5-sonnet'
// - 'google/gemini-2.0-flash'
// - 'mistral/mistral-large'
// - 'meta/llama-3.1-70b'
// Note: Requires LiteLLM proxy server configuration

Dynamic Model System (v1.8.0+)¶

Overview¶

NeuroLink now supports a dynamic model configuration system that replaces static TypeScript enums with runtime-configurable model definitions. This enables:

✅ Runtime Model Updates - Add/remove models without code changes
✅ Smart Model Resolution - Use aliases like "claude-latest", "best-coding", "fastest"
✅ Cost Optimization - Automatic best-value model selection
✅ Provider Agnostic - Unified model interface across all providers
✅ Type Safety - Zod schema validation for all configurations

Model Configuration Server¶

The dynamic system includes a REST API server for model configurations:

# Start the model configuration server
npm run start:model-server

# Server runs on http://localhost:3001
# API endpoints:
# GET /models - List all models
# GET /models/search?capability=vision - Search by capability
# GET /models/provider/anthropic - Get provider models
# GET /models/resolve/claude-latest - Resolve aliases

Model Configuration Schema¶

Models are defined in config/models.json with comprehensive metadata:

interface ModelConfig {
  id: string; // Unique model identifier
  name: string; // Display name
  provider: string; // Provider name (anthropic, openai, etc.)
  pricing: {
    input: number; // Cost per 1K input tokens
    output: number; // Cost per 1K output tokens
  };
  capabilities: string[]; // ['functionCalling', 'vision', 'code']
  contextWindow: number; // Maximum context length
  deprecated: boolean; // Whether model is deprecated
  aliases: string[]; // Alternative names
  metadata: {
    description: string;
    useCase: string; // 'general', 'coding', 'vision', etc.
    speed: "fast" | "medium" | "slow";
    quality: "high" | "medium" | "low";
  };
}

Smart Model Resolution¶

The dynamic system provides intelligent model resolution:

import { DynamicModelRegistry } from "@juspay/neurolink";

const registry = new DynamicModelRegistry();

// Resolve aliases to actual model IDs
await registry.resolveModel("claude-latest"); // → 'claude-3-5-sonnet'
await registry.resolveModel("fastest"); // → 'gpt-4o-mini'
await registry.resolveModel("best-coding"); // → 'claude-3-5-sonnet'

// Find best model for specific criteria
await registry.findBestModel({
  capability: "vision",
  maxPrice: 0.001, // Maximum cost per 1K tokens
  provider: "anthropic", // Optional provider preference
});

// Get models by capability
await registry.getModelsByCapability("functionCalling");

// Cost-optimized model selection
await registry.getCheapestModel("general"); // Cheapest general-purpose model
await registry.getFastestModel("coding"); // Fastest coding model

Dynamic Model Usage in AI Factory¶

The AI factory automatically uses the dynamic model system:

import { AIProviderFactory } from "@juspay/neurolink";

const factory = new AIProviderFactory();

// Use model aliases
const provider1 = await factory.createProvider({
  provider: "anthropic",
  model: "claude-latest", // Resolves to latest Claude model
});

// Use capability-based selection
const provider2 = await factory.createProvider({
  provider: "auto",
  model: "best-vision", // Selects best vision model
  optimizeFor: "cost", // Prefer cost-effective models
});

// Use direct model IDs (still supported)
const provider3 = await factory.createProvider({
  provider: "openai",
  model: "gpt-4o", // Direct model specification
});

Configuration Management¶

Environment Variables for Dynamic Models¶

// Model server configuration
MODEL_SERVER_URL?: string        // Default: 'http://localhost:3001'
MODEL_CONFIG_PATH?: string       // Default: './config/models.json'
ENABLE_DYNAMIC_MODELS?: string   // Default: 'true'

// Model selection preferences
DEFAULT_MODEL_PREFERENCE?: 'cost' | 'speed' | 'quality'  // Default: 'quality'
FALLBACK_MODEL?: string          // Model to use if preferred unavailable

Configuration File Structure¶

The config/models.json file defines all available models:

{
  "models": [
    {
      "id": "claude-3-5-sonnet",
      "name": "Claude 3.5 Sonnet",
      "provider": "anthropic",
      "pricing": { "input": 0.003, "output": 0.015 },
      "capabilities": ["functionCalling", "vision", "code"],
      "contextWindow": 200000,
      "deprecated": false,
      "aliases": ["claude-latest", "best-coding", "claude-sonnet"],
      "metadata": {
        "description": "Most capable Claude model",
        "useCase": "general",
        "speed": "medium",
        "quality": "high"
      }
    }
  ],
  "aliases": {
    "claude-latest": "claude-3-5-sonnet",
    "fastest": "gpt-4o-mini",
    "cheapest": "claude-3-haiku",
    "best-vision": "gpt-4o",
    "best-coding": "claude-3-5-sonnet"
  }
}

CLI Integration¶

The CLI provides comprehensive dynamic model management:

# List all models with pricing
neurolink models list

# Search models by capability
neurolink models search --capability functionCalling
neurolink models search --capability vision --max-price 0.001

# Get best model for use case
neurolink models best --use-case coding
neurolink models best --use-case vision

# Resolve aliases
neurolink models resolve anthropic claude-latest
neurolink models resolve google fastest

# Test with dynamic model selection
neurolink generate "Hello" --model best-coding
neurolink generate "Describe this" --capability vision --optimize-cost

Type Definitions for Dynamic Models¶

interface DynamicModelOptions {
  // Specify exact model ID
  model?: string;

  // OR specify requirements for automatic selection
  capability?: "functionCalling" | "vision" | "code" | "general";
  maxPrice?: number; // Maximum cost per 1K tokens
  optimizeFor?: "cost" | "speed" | "quality";
  provider?: string; // Preferred provider
}

interface ModelResolutionResult {
  modelId: string; // Resolved model ID
  provider: string; // Provider name
  reasoning: string; // Why this model was selected
  pricing: {
    input: number;
    output: number;
  };
  capabilities: string[];
}

interface ModelSearchOptions {
  capability?: string;
  provider?: string;
  maxPrice?: number;
  minContextWindow?: number;
  excludeDeprecated?: boolean;
}

Migration from Static Models¶

For existing code using static model enums, the transition is seamless:

// OLD: Static enum usage (still works)
const provider = await factory.createProvider({
  provider: "anthropic",
  model: "claude-3-5-sonnet",
});

// NEW: Dynamic model usage (recommended)
const provider = await factory.createProvider({
  provider: "anthropic",
  model: "claude-latest", // Auto-resolves to latest Claude
});

// ADVANCED: Capability-based selection
const provider = await factory.createProvider({
  provider: "auto",
  capability: "vision",
  optimizeFor: "cost",
});

The dynamic model system maintains backward compatibility while enabling powerful new capabilities for intelligent model selection and cost optimization.

Environment Configuration¶

Required Environment Variables¶

// OpenAI
OPENAI_API_KEY: string

// Amazon Bedrock
AWS_ACCESS_KEY_ID: string
AWS_SECRET_ACCESS_KEY: string
AWS_REGION?: string              // Default: 'us-east-2'
AWS_SESSION_TOKEN?: string       // For temporary credentials
BEDROCK_MODEL?: string           // Inference profile ARN

// Amazon SageMaker
AWS_ACCESS_KEY_ID: string                    // AWS access key (shared with Bedrock)
AWS_SECRET_ACCESS_KEY: string                // AWS secret key (shared with Bedrock)
AWS_REGION?: string                          // Default: 'us-east-1'
SAGEMAKER_DEFAULT_ENDPOINT: string           // SageMaker endpoint name
SAGEMAKER_TIMEOUT?: string                   // Default: '30000'
SAGEMAKER_MAX_RETRIES?: string               // Default: '3'
AWS_SESSION_TOKEN?: string                   // For temporary credentials

// Google Vertex AI (choose one authentication method)
GOOGLE_APPLICATION_CREDENTIALS?: string           // Method 1: File path
GOOGLE_SERVICE_ACCOUNT_KEY?: string              // Method 2: JSON string
GOOGLE_AUTH_CLIENT_EMAIL?: string                // Method 3a: Individual vars
GOOGLE_AUTH_PRIVATE_KEY?: string                 // Method 3b: Individual vars
GOOGLE_VERTEX_PROJECT: string                    // Required for all methods
GOOGLE_VERTEX_LOCATION?: string                  // Default: 'us-east5'

// Google AI Studio
GOOGLE_AI_API_KEY: string                        // API key from AI Studio

// Anthropic
ANTHROPIC_API_KEY?: string                       // Direct Anthropic API

// Azure OpenAI
AZURE_OPENAI_API_KEY?: string                    // Azure OpenAI API key
AZURE_OPENAI_ENDPOINT?: string                   // Azure OpenAI endpoint
AZURE_OPENAI_DEPLOYMENT_ID?: string              // Deployment ID

// Hugging Face
HUGGINGFACE_API_KEY: string                      // HF token from huggingface.co
HUGGINGFACE_MODEL?: string                       // Default: 'microsoft/DialoGPT-medium'

// Ollama (Local)
OLLAMA_BASE_URL?: string                         // Default: 'http://localhost:11434'
OLLAMA_MODEL?: string                            // Default: 'llama2'

// Mistral AI
MISTRAL_API_KEY: string                          // API key from mistral.ai
MISTRAL_MODEL?: string                           // Default: 'mistral-small'

// LiteLLM (100+ Models via Proxy)
LITELLM_BASE_URL?: string                        // Default: 'http://localhost:4000'
LITELLM_API_KEY?: string                         // Default: 'sk-anything'
LITELLM_MODEL?: string                           // Default: 'openai/gpt-4o-mini'

// Dynamic Model System (v1.8.0+)
MODEL_SERVER_URL?: string                        // Default: 'http://localhost:3001'
MODEL_CONFIG_PATH?: string                       // Default: './config/models.json'
ENABLE_DYNAMIC_MODELS?: string                   // Default: 'true'
DEFAULT_MODEL_PREFERENCE?: 'cost' | 'speed' | 'quality'  // Default: 'quality'
FALLBACK_MODEL?: string                          // Model to use if preferred unavailable

Optional Configuration Variables¶

// Provider preferences
DEFAULT_PROVIDER?: 'auto' | 'openai' | 'bedrock' | 'sagemaker' | 'vertex' | 'anthropic' | 'azure' | 'google-ai' | 'huggingface' | 'ollama' | 'mistral' | 'litellm'
FALLBACK_PROVIDER?: 'openai' | 'bedrock' | 'sagemaker' | 'vertex' | 'anthropic' | 'azure' | 'google-ai' | 'huggingface' | 'ollama' | 'mistral' | 'litellm'

// Feature toggles
ENABLE_STREAMING?: 'true' | 'false'
ENABLE_FALLBACK?: 'true' | 'false'

// Debugging
NEUROLINK_DEBUG?: 'true' | 'false'
LOG_LEVEL?: 'error' | 'warn' | 'info' | 'debug'

Type Definitions¶

Core Types¶

type ProviderName =
  | "openai"
  | "bedrock"
  | "sagemaker"
  | "vertex"
  | "anthropic"
  | "azure"
  | "google-ai"
  | "huggingface"
  | "ollama"
  | "mistral"
  | "litellm";

interface AIProvider {
  generate(options: GenerateOptions): Promise<GenerateResult>;
  stream(options: StreamOptions | string): Promise<StreamResult>; // PRIMARY streaming method
}

interface GenerateOptions {
  input: { text: string };
  temperature?: number; // 0.0 to 1.0, default: 0.7
  maxTokens?: number; // Default: 1000
  systemPrompt?: string; // System message
  schema?: any; // For structured output
  timeout?: number | string; // Timeout in ms or human-readable format
  disableTools?: boolean; // Disable tool usage
  enableAnalytics?: boolean; // Enable usage analytics
  enableEvaluation?: boolean; // Enable AI quality scoring
  context?: Record<string, any>; // Custom context for analytics
}

interface GenerateResult {
  content: string;
  provider: string;
  model: string;
  usage?: TokenUsage;
  responseTime?: number; // Milliseconds
  analytics?: {
    provider: string;
    model: string;
    tokens: { input: number; output: number; total: number };
    cost?: number;
    responseTime: number;
    context?: Record<string, any>;
  };
  evaluation?: {
    relevanceScore: number; // 1-10 scale
    accuracyScore: number; // 1-10 scale
    completenessScore: number; // 1-10 scale
    overallScore: number; // 1-10 scale
    alertLevel?: string; // 'none', 'low', 'medium', 'high'
    reasoning?: string; // AI reasoning for the evaluation
  };
}

interface TokenUsage {
  promptTokens: number;
  completionTokens: number;
  totalTokens: number;
}

Dynamic Model Types (v1.8.0+)¶

interface ModelConfig {
  id: string; // Unique model identifier
  name: string; // Display name
  provider: string; // Provider name (anthropic, openai, etc.)
  pricing: {
    input: number; // Cost per 1K input tokens
    output: number; // Cost per 1K output tokens
  };
  capabilities: string[]; // ['functionCalling', 'vision', 'code']
  contextWindow: number; // Maximum context length
  deprecated: boolean; // Whether model is deprecated
  aliases: string[]; // Alternative names
  metadata: {
    description: string;
    useCase: string; // 'general', 'coding', 'vision', etc.
    speed: "fast" | "medium" | "slow";
    quality: "high" | "medium" | "low";
  };
}

interface DynamicModelOptions {
  // Specify exact model ID
  model?: string;

  // OR specify requirements for automatic selection
  capability?: "functionCalling" | "vision" | "code" | "general";
  maxPrice?: number; // Maximum cost per 1K tokens
  optimizeFor?: "cost" | "speed" | "quality";
  provider?: string; // Preferred provider
}

interface ModelResolutionResult {
  modelId: string; // Resolved model ID
  provider: string; // Provider name
  reasoning: string; // Why this model was selected
  pricing: {
    input: number;
    output: number;
  };
  capabilities: string[];
}

interface ModelSearchOptions {
  capability?: string;
  provider?: string;
  maxPrice?: number;
  minContextWindow?: number;
  excludeDeprecated?: boolean;
}

interface DynamicModelRegistry {
  resolveModel(alias: string): Promise<string>;
  findBestModel(options: DynamicModelOptions): Promise<ModelResolutionResult>;
  getModelsByCapability(capability: string): Promise<ModelConfig[]>;
  getCheapestModel(useCase: string): Promise<ModelConfig>;
  getFastestModel(useCase: string): Promise<ModelConfig>;
  searchModels(options: ModelSearchOptions): Promise<ModelConfig[]>;
  getModelConfig(modelId: string): Promise<ModelConfig | null>;
  getAllModels(): Promise<ModelConfig[]>;
}

Provider Tool Support Status¶

Due to the factory pattern refactoring, all providers now have consistent tool support through BaseProvider:

Provider	Tool Support	Notes
OpenAI	✅ Full	All tools work correctly
Google AI	✅ Full	Excellent tool execution
Anthropic	✅ Full	Reliable tool usage
Azure OpenAI	✅ Full	Same as OpenAI
Mistral	✅ Full	Good tool support
HuggingFace	⚠️ Partial	Model sees tools but may describe instead of execute
Vertex AI	⚠️ Partial	Tools available but may not execute
Ollama	❌ Limited	Requires specific models like gemma3n
Bedrock	✅ Full*	Requires valid AWS credentials

Provider-Specific Types¶

// OpenAI specific
interface OpenAIOptions extends GenerateOptions {
  user?: string; // User identifier
  stop?: string | string[]; // Stop sequences
  topP?: number; // Nucleus sampling
  frequencyPenalty?: number; // Reduce repetition
  presencePenalty?: number; // Encourage diversity
}

// Bedrock specific
interface BedrockOptions extends GenerateOptions {
  region?: string; // AWS region override
  inferenceProfile?: string; // Inference profile ARN
}

// SageMaker specific
interface SageMakerOptions extends GenerateOptions {
  endpoint?: string; // Override default endpoint
  region?: string; // AWS region override
  contentType?: string; // Request content type (default: application/json)
  accept?: string; // Response accept type (default: application/json)
  customAttributes?: string; // Custom attributes for the request
  targetModel?: string; // Target model for multi-model endpoints
}

// Vertex AI specific
interface VertexOptions extends GenerateOptions {
  project?: string; // GCP project override
  location?: string; // GCP location override
  safetySettings?: any[]; // Safety filter settings
}

// Google AI Studio specific
interface GoogleAIOptions extends GenerateOptions {
  safetySettings?: any[]; // Safety filter settings
  generationConfig?: {
    // Additional generation settings
    stopSequences?: string[];
    candidateCount?: number;
    topK?: number;
    topP?: number;
  };
}

// Anthropic specific
interface AnthropicOptions extends GenerateOptions {
  stopSequences?: string[]; // Custom stop sequences
  metadata?: {
    // Usage tracking
    userId?: string;
  };
}

// Azure OpenAI specific
interface AzureOptions extends GenerateOptions {
  deploymentId?: string; // Override deployment
  apiVersion?: string; // API version override
  user?: string; // User tracking
}

// Hugging Face specific
interface HuggingFaceOptions extends GenerateOptions {
  waitForModel?: boolean; // Wait for model to load
  useCache?: boolean; // Use cached responses
  options?: {
    // Model-specific options
    useGpu?: boolean;
    precision?: string;
  };
}

// Ollama specific
interface OllamaOptions extends GenerateOptions {
  format?: string; // Response format (e.g., 'json')
  context?: number[]; // Conversation context
  stream?: boolean; // Enable streaming
  raw?: boolean; // Raw mode (no templating)
  keepAlive?: string; // Model keep-alive duration
}

// Mistral AI specific
interface MistralOptions extends GenerateOptions {
  topP?: number; // Nucleus sampling
  randomSeed?: number; // Reproducible outputs
  safeMode?: boolean; // Enable safe mode
  safePrompt?: boolean; // Add safe prompt
}

Error Handling¶

Error Types¶

class AIProviderError extends Error {
  provider: string;
  originalError?: Error;
}

class TimeoutError extends AIProviderError {
  // Thrown when operation exceeds specified timeout
  timeout: number; // Timeout in milliseconds
  operation?: string; // Operation that timed out (e.g., 'generate', 'stream')
}

class ConfigurationError extends AIProviderError {
  // Thrown when provider configuration is invalid
}

class AuthenticationError extends AIProviderError {
  // Thrown when authentication fails
}

class RateLimitError extends AIProviderError {
  // Thrown when rate limits are exceeded
  retryAfter?: number; // Seconds to wait before retrying
}

class QuotaExceededError extends AIProviderError {
  // Thrown when usage quotas are exceeded
}

Error Handling Patterns¶

import {
  AIProviderError,
  ConfigurationError,
  AuthenticationError,
  RateLimitError,
  TimeoutError,
} from "@juspay/neurolink";

try {
  const result = await provider.generate({
    prompt: "Hello",
    timeout: "30s",
  });
} catch (error) {
  if (error instanceof TimeoutError) {
    console.error(`Operation timed out after ${error.timeout}ms`);
    console.error(`Provider: ${error.provider}, Operation: ${error.operation}`);
  } else if (error instanceof ConfigurationError) {
    console.error("Provider not configured:", error.message);
  } else if (error instanceof AuthenticationError) {
    console.error("Authentication failed:", error.message);
  } else if (error instanceof RateLimitError) {
    console.error(`Rate limit exceeded. Retry after ${error.retryAfter}s`);
  } else if (error instanceof AIProviderError) {
    console.error(`Provider ${error.provider} failed:`, error.message);
  } else {
    console.error("Unexpected error:", error);
  }
}

Advanced Usage Patterns¶

Custom Provider Selection¶

interface ProviderSelector {
  selectProvider(available: ProviderName[]): ProviderName;
}

class CustomSelector implements ProviderSelector {
  selectProvider(available: ProviderName[]): ProviderName {
    // Custom logic for provider selection
    if (available.includes("bedrock")) return "bedrock";
    if (available.includes("openai")) return "openai";
    return available[0];
  }
}

// Usage with custom selector
const provider = createBestAIProvider(); // Uses default selection logic

Middleware Support¶

interface AIMiddleware {
  beforeRequest?(options: GenerateOptions): GenerateOptions;
  afterResponse?(result: GenerateResult): GenerateResult;
  onError?(error: Error): Error;
}

class LoggingMiddleware implements AIMiddleware {
  beforeRequest(options: GenerateOptions): GenerateOptions {
    console.log(
      `Generating text for prompt: ${options.prompt.slice(0, 50)}...`,
    );
    return options;
  }

  afterResponse(result: GenerateResult): GenerateResult {
    console.log(
      `Generated ${result.text.length} characters using ${result.provider}`,
    );
    return result;
  }
}

// Note: Middleware is a planned feature for future versions

Batch Processing¶

async function processBatch(prompts: string[], options: GenerateOptions = {}) {
  const provider = createBestAIProvider();
  const results = [];

  for (const prompt of prompts) {
    try {
      const result = await provider.generate({ ...options, prompt });
      results.push({ success: true, ...result });
    } catch (error) {
      results.push({
        success: false,
        prompt,
        error: error.message,
      });
    }

    // Rate limiting: wait 1 second between requests
    await new Promise((resolve) => setTimeout(resolve, 1000));
  }

  return results;
}

// Usage
const prompts = [
  "Explain photosynthesis",
  "What is machine learning?",
  "Describe the solar system",
];

const results = await processBatch(prompts, {
  temperature: 0.7,
  maxTokens: 200,
  timeout: "45s", // Set reasonable timeout for batch operations
});

Response Caching¶

class CachedProvider implements AIProvider {
  private cache = new Map<string, GenerateResult>();
  private provider: AIProvider;

  constructor(provider: AIProvider) {
    this.provider = provider;
  }

  async generate(options: GenerateOptions): Promise<GenerateResult> {
    const key = JSON.stringify(options);

    if (this.cache.has(key)) {
      return { ...this.cache.get(key)!, fromCache: true };
    }

    const result = await this.provider.generate(options);
    this.cache.set(key, result);
    return result;
  }

  async stream(options: StreamOptions): Promise<StreamResult> {
    // Streaming responses are not cached
    return this.provider.stream(options);
  }
}

// Usage
const baseProvider = createBestAIProvider();
const cachedProvider = new CachedProvider(baseProvider);

TypeScript Integration¶

Type-Safe Configuration¶

interface NeuroLinkConfig {
  defaultProvider?: ProviderName;
  fallbackProvider?: ProviderName;
  defaultOptions?: Partial<GenerateOptions>;
  enableFallback?: boolean;
  enableStreaming?: boolean;
  debug?: boolean;
}

const config: NeuroLinkConfig = {
  defaultProvider: "openai",
  fallbackProvider: "bedrock",
  defaultOptions: {
    temperature: 0.7,
    maxTokens: 500,
  },
  enableFallback: true,
  debug: false,
};

Generic Provider Interface¶

interface TypedAIProvider<
  TOptions = GenerateOptions,
  TResult = GenerateResult,
> {
  generate(options: TOptions): Promise<TResult>;
}

// Custom typed provider
interface CustomOptions extends GenerateOptions {
  customParameter?: string;
}

interface CustomResult extends GenerateResult {
  customData?: any;
}

const typedProvider: TypedAIProvider<CustomOptions, CustomResult> =
  createBestAIProvider() as any;

MCP (Model Context Protocol) APIs¶

NeuroLink supports MCP through built-in tools and SDK custom tool registration.

✅ Current Status¶

Built-in Tools: ✅ FULLY FUNCTIONAL

✅ Time tool - Returns current time in human-readable format
✅ Built-in utilities - All system tools working correctly
✅ CLI integration - Direct tool execution via CLI
✅ Function calling - Tools properly registered and callable

External MCP Tools: 🔍 DISCOVERY PHASE

✅ Auto-discovery working - 58+ external servers found
✅ Configuration parsing - Resilient JSON parser handles all formats
✅ Cross-platform support - macOS, Linux, Windows configurations
🔧 Tool activation - External servers discovered but in placeholder mode
🔧 Communication protocol - Under active development for full activation

Current Working Examples¶

# ✅ Working: Test built-in tools
neurolink generate "What time is it?" --debug
neurolink generate "What tools do you have access to?" --debug

# ✅ Working: Discover external MCP servers
neurolink mcp discover --format table

# ✅ Working: Build and test system
npm run build && npm run test:run -- test/mcp-comprehensive.test.ts

MCP CLI Commands¶

All MCP functionality is available through the NeuroLink CLI:

# ✅ Working: Built-in tool testing
neurolink generate "What time is it?" --debug

# ✅ Working: Server discovery and management
neurolink mcp discover [--format table|json|yaml]  # Auto-discover MCP servers
neurolink mcp list [--status]     # List discovered servers with optional status

# 🔧 In Development: Server management and execution
neurolink mcp install <server>    # Install popular MCP servers (discovery phase)
neurolink mcp add <name> <command> # Add custom MCP server
neurolink mcp remove <server>     # Remove MCP server
neurolink mcp test <server>       # Test server connectivity
neurolink mcp tools <server>      # List available tools for server
neurolink mcp execute <server> <tool> [args] # Execute specific tool

# Configuration management
neurolink mcp config             # Show MCP configuration
neurolink mcp config --reset     # Reset MCP configuration

MCP Server Types¶

Built-in Server Support¶

NeuroLink includes built-in installation support for popular MCP servers:

type PopularMCPServer =
  | "filesystem" // File operations
  | "github" // GitHub integration
  | "postgres" // PostgreSQL database
  | "puppeteer" // Web browsing
  | "brave-search"; // Web search

Additional MCP Servers While not included in the auto-install feature, any MCP-compatible server can be manually added, including:

git - Git operations
fetch - Web fetching
google-drive - Google Drive integration
atlassian - Jira/Confluence integration
slack - Slack integration
Any custom MCP server

Use neurolink mcp add <name> <command> to add these servers manually.

Custom Server Support¶

Add any MCP-compatible server:

# Python server
neurolink mcp add myserver "python /path/to/server.py"

# Node.js server
neurolink mcp add nodeserver "node /path/to/server.js"

# Docker container
neurolink mcp add dockerserver "docker run my-mcp-server"

# SSE (Server-Sent Events) endpoint
neurolink mcp add sseserver "sse://https://api.example.com/mcp"

MCP Configuration¶

Configuration File¶

MCP servers are configured in .mcp-config.json:

interface MCPConfig {
  mcpServers: {
    [serverName: string]: {
      command: string; // Command to start server
      args?: string[]; // Optional command arguments
      env?: Record<string, string>; // Environment variables
      cwd?: string; // Working directory
      timeout?: number; // Connection timeout (ms)
      retry?: number; // Retry attempts
      enabled?: boolean; // Server enabled status
    };
  };
  global?: {
    timeout?: number; // Global timeout
    maxConnections?: number; // Max concurrent connections
    logLevel?: "debug" | "info" | "warn" | "error";
  };
}

Example Configuration¶

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/"],
      "timeout": 5000,
      "enabled": true
    },

      "timeout": 8000,
      "enabled": false
    }
  },
  "global": {
    "timeout": 10000,
    "maxConnections": 5,
    "logLevel": "info"
  }
}

MCP Environment Variables¶

Configure MCP server authentication through environment variables:

# Database connections
MYSQL_CONNECTION_STRING=mysql://user:pass@localhost/db

# Web services
BRAVE_API_KEY=BSA...
GOOGLE_API_KEY=AIza...

MCP Tool Execution¶

Available Tool Categories¶

interface MCPToolCategory {
  filesystem: {
    read_file: { path: string };
    write_file: { path: string; content: string };
    list_directory: { path: string };
    search_files: { query: string; path?: string };
  };

  github: {
    get_repository: { owner: string; repo: string };
    create_issue: { owner: string; repo: string; title: string; body?: string };
    list_issues: { owner: string; repo: string; state?: "open" | "closed" };
    create_pull_request: {
      owner: string;
      repo: string;
      title: string;
      head: string;
      base: string;
    };
  };

  database: {
    execute_query: { query: string; params?: any[] };
    list_tables: {};
    describe_table: { table: string };
  };

  web: {
    navigate: { url: string };
    click: { selector: string };
    type: { selector: string; text: string };
    screenshot: { name?: string };
  };
}

Tool Execution Examples¶

# File operations
neurolink mcp exec filesystem read_file --params '{"path": "/path/to/file.txt"}'
neurolink mcp exec filesystem list_directory --params '{"path": "/home/user"}'

# GitHub operations
neurolink mcp exec github get_repository --params '{"owner": "juspay", "repo": "neurolink"}'
neurolink mcp exec github create_issue --params '{"owner": "juspay", "repo": "neurolink", "title": "New feature request"}'

# Database operations
neurolink mcp exec postgres execute_query --params '{"query": "SELECT * FROM users LIMIT 10"}'
neurolink mcp exec postgres list_tables --params '{}'

# Web operations
neurolink mcp exec puppeteer navigate --params '{"url": "https://example.com"}'
neurolink mcp exec puppeteer screenshot --params '{"name": "homepage"}'

MCP Demo Server Integration¶

FULLY FUNCTIONAL: NeuroLink's demo server (neurolink-demo/server.js) includes working MCP API endpoints that you can use immediately:

How to Access These APIs¶

# 1. Start the demo server
cd neurolink-demo
node server.js
# Server runs at http://localhost:9876

# 2. Use any HTTP client to call the APIs
curl http://localhost:9876/api/mcp/servers
curl -X POST http://localhost:9876/api/mcp/install -d '{"serverName": "filesystem"}'

Available MCP API Endpoints¶

// ALL ENDPOINTS WORKING IN DEMO SERVER
interface MCPDemoEndpoints {
  "GET /api/mcp/servers": {
    // List all configured MCP servers with live status
    response: {
      servers: Array<{
        name: string;
        status: "connected" | "disconnected" | "error";
        tools: string[];
        lastConnected?: string;
      }>;
    };
  };

  "POST /api/mcp/install": {
    // Install popular MCP servers (filesystem, github, postgres, etc.)
    body: { serverName: string };
    response: {
      success: boolean;
      message: string;
      configuration?: Record<string, any>;
    };
  };

  "DELETE /api/mcp/servers/:name": {
    // Remove MCP servers
    params: { name: string };
    response: {
      success: boolean;
      message: string;
    };
  };

  "POST /api/mcp/test/:name": {
    // Test server connectivity and get diagnostics
    params: { name: string };
    response: {
      success: boolean;
      status: "connected" | "disconnected" | "error";
      responseTime?: number;
      error?: string;
    };
  };

  "GET /api/mcp/tools/:name": {
    // Get available tools for specific server
    params: { name: string };
    response: {
      success: boolean;
      tools: Array<{
        name: string;
        description: string;
        parameters: Record<string, any>;
      }>;
    };
  };

  "POST /api/mcp/execute": {
    // Execute MCP tools via HTTP API
    body: {
      serverName: string;
      toolName: string;
      params: Record<string, any>;
    };
    response: {
      success: boolean;
      result?: any;
      error?: string;
      executionTime?: number;
    };
  };

  "POST /api/mcp/servers/custom": {
    // Add custom MCP servers
    body: {
      name: string;
      command: string;
      options?: Record<string, any>;
    };
    response: {
      success: boolean;
      message: string;
    };
  };

  "GET /api/mcp/status": {
    // Get comprehensive MCP system status
    response: {
      summary: {
        totalServers: number;
        availableServers: number;
        cliAvailable: boolean;
      };
      servers: Record<string, any>;
    };
  };

  "POST /api/mcp/workflow": {
    // Execute predefined MCP workflows
    body: {
      workflowType: string;
      description?: string;
      servers?: string[];
    };
    response: {
      success: boolean;
      workflowType: string;
      steps: string[];
      result: string;
      data: any;
    };
  };
}

Real-World Usage Examples¶

1. File Operations via HTTP API

# Install filesystem server
curl -X POST http://localhost:9876/api/mcp/install \
  -H "Content-Type: application/json" \
  -d '{"serverName": "filesystem"}'

# Read a file via HTTP
curl -X POST http://localhost:9876/api/mcp/execute \
  -H "Content-Type: application/json" \
  -d '{
    "serverName": "filesystem",
    "toolName": "read_file",
    "params": {"path": "index.md"}
  }'

# List directory contents
curl -X POST http://localhost:9876/api/mcp/execute \
  -H "Content-Type: application/json" \
  -d '{
    "serverName": "filesystem",
    "toolName": "list_directory",
    "params": {"path": "."}
  }'

2. GitHub Integration via HTTP API

# Install GitHub server
curl -X POST http://localhost:9876/api/mcp/install \
  -H "Content-Type: application/json" \
  -d '{"serverName": "github"}'

# Get repository information
curl -X POST http://localhost:9876/api/mcp/execute \
  -H "Content-Type: application/json" \
  -d '{
    "serverName": "github",
    "toolName": "get_repository",
    "params": {"owner": "juspay", "repo": "neurolink"}
  }'

3. Web Interface Integration

// JavaScript example for web applications
async function callMCPTool(serverName, toolName, params) {
  const response = await fetch("http://localhost:9876/api/mcp/execute", {
    method: "POST",
    headers: { "Content-Type": "application/json" },
    body: JSON.stringify({ serverName, toolName, params }),
  });

  const result = await response.json();
  return result;
}

// Use in your web app
const fileContent = await callMCPTool("filesystem", "read_file", {
  path: "/path/to/file.txt",
});

What You Can Use This For¶

1. Web Application MCP Integration

Build web dashboards that manage MCP servers
Create file management interfaces
Integrate GitHub operations into web apps
Build database administration tools

2. API-First MCP Development

Test MCP tools without CLI setup
Prototype MCP integrations quickly
Build custom MCP management interfaces
Create automated workflows via HTTP

3. Cross-Platform MCP Access

Access MCP tools from any programming language
Build mobile apps that use MCP functionality
Create browser extensions with MCP features
Integrate with existing web services

4. Educational and Testing

Learn MCP concepts through web interface
Test MCP server configurations
Debug MCP tool interactions
Demonstrate MCP capabilities to others

Getting Started¶

# 1. Clone and setup
git clone https://github.com/juspay/neurolink
cd neurolink/neurolink-demo

# 2. Install dependencies
npm install

# 3. Configure environment (optional)
cp .env.example .env
# Add any needed API keys

# 4. Start server
node server.js

# 5. Test APIs
curl http://localhost:9876/api/mcp/status
curl http://localhost:9876/api/mcp/servers

The demo server provides a production-ready MCP HTTP API that you can integrate into any application or service.

MCP Error Handling¶

class MCPError extends Error {
  server: string;
  tool?: string;
  originalError?: Error;
}

class MCPConnectionError extends MCPError {
  // Thrown when server connection fails
}

class MCPToolError extends MCPError {
  // Thrown when tool execution fails
}

class MCPConfigurationError extends MCPError {
  // Thrown when server configuration is invalid
}

// Error handling example
try {
  const result = await executeCommand(
    'neurolink mcp execute filesystem read_file --path="/nonexistent"',
  );
} catch (error) {
  if (error instanceof MCPConnectionError) {
    console.error(`Failed to connect to server ${error.server}`);
  } else if (error instanceof MCPToolError) {
    console.error(
      `Tool ${error.tool} failed on server ${error.server}: ${error.message}`,
    );
  }
}

MCP Integration Best Practices¶

Server Management¶

# Test connectivity before using
neurolink mcp test filesystem

# Install servers explicitly
neurolink mcp install github
neurolink mcp install postgres

# Monitor server status
neurolink mcp list --status

Environment Setup¶

# Test configuration
neurolink mcp test

Error Recovery¶

# Reset configuration if needed
neurolink mcp config --reset

# Reinstall problematic servers
neurolink mcp remove filesystem
neurolink mcp install filesystem
neurolink mcp test filesystem

Performance Optimization¶

# Limit concurrent connections in config
{
  "global": {
    "maxConnections": 3,
    "timeout": 5000
  }
}

# Disable unused servers
{
  "mcpServers": {
    "heavyServer": {
      "command": "...",
      "enabled": false
    }
  }
}

← Back to Main README | Next: Visual Demos →

📚 API Reference¶

Core Functions¶

createBestAIProvider(requestedProvider?, modelName?)¶

createAIProviderWithFallback(primary, fallback, modelName?)¶

BaseProvider Class¶

Key Features¶

Built-in Tools¶

Example Usage¶

AIProviderFactory¶

createProvider(providerName, modelName?)¶

createProviderWithFallback(primary, fallback, modelName?)¶

AIProvider Interface¶

🔗 CLI-SDK Consistency¶

🆕 NeuroLink Class API¶

addMCPServer(serverId, config)¶

getMCPStatus()¶

getUnifiedRegistry()¶

generate(options)¶

🆕 Enterprise Configuration Interfaces¶

NeuroLinkConfig¶

ExecutionContext¶

ToolInfo¶

ConfigUpdateOptions¶

McpRegistry¶

🌐 Enterprise Real-time Services API¶

createEnhancedChatService(options)¶

NeuroLinkWebSocketServer¶

📊 Enterprise Telemetry API¶

initializeTelemetry(config)¶

getTelemetryStatus()¶

🔧 Enhanced Generation Options¶

stream(options) - Recommended for New Code¶

Flexible Parameter Support¶

Using Timeouts¶

Usage Examples¶

Basic Usage¶

Dynamic Model Usage (v1.8.0+)¶

Cost-Optimized Generation¶

Vision Capabilities with Dynamic Selection¶

Function Calling with Smart Model Selection¶

Model Discovery and Search¶

Streaming with Dynamic Models¶

Provider Fallback with Dynamic Models¶

Supported Models¶

OpenAI Models¶

Amazon Bedrock Models¶

Amazon SageMaker Models¶

Google Vertex AI Models¶

Google AI Studio Models¶

Azure OpenAI Models¶

Hugging Face Models¶

Ollama Models¶

Mistral AI Models¶

LiteLLM Models¶

Dynamic Model System (v1.8.0+)¶

Overview¶

Model Configuration Server¶

Model Configuration Schema¶

Smart Model Resolution¶

Dynamic Model Usage in AI Factory¶

Configuration Management¶

Environment Variables for Dynamic Models¶

Configuration File Structure¶

CLI Integration¶

Type Definitions for Dynamic Models¶

Migration from Static Models¶

Environment Configuration¶

Required Environment Variables¶

Optional Configuration Variables¶

Type Definitions¶

Core Types¶

Dynamic Model Types (v1.8.0+)¶

Provider Tool Support Status¶

Provider-Specific Types¶

Error Handling¶

Error Types¶

Error Handling Patterns¶

Advanced Usage Patterns¶

Custom Provider Selection¶

Middleware Support¶

`createBestAIProvider(requestedProvider?, modelName?)`¶

`createAIProviderWithFallback(primary, fallback, modelName?)`¶

`createProvider(providerName, modelName?)`¶

`createProviderWithFallback(primary, fallback, modelName?)`¶

`addMCPServer(serverId, config)`¶

`getMCPStatus()`¶

`getUnifiedRegistry()`¶

`generate(options)`¶

`NeuroLinkConfig`¶

`ExecutionContext`¶

`ToolInfo`¶

`ConfigUpdateOptions`¶

`McpRegistry`¶

`createEnhancedChatService(options)`¶

`NeuroLinkWebSocketServer`¶

`initializeTelemetry(config)`¶

`getTelemetryStatus()`¶

`stream(options)` - Recommended for New Code¶