@ecfjs/ai — Enterprise AI Engine & LLM Platform

@ecfjs/ai is the official first-party enterprise AI engine for ECF (Enterprise Core Framework). It provides multi-provider LLM chat (AI.chat()), token streaming (AI.stream()), state-managed conversation memory (AI.memory()), structured output validation, versioned prompt templates (AI.prompt("support@v2")), dedicated EmbeddingManager, modular RAG pipeline (AI.rag()), post-retrieval ReRanker, code/markdown chunkers (Chunk.code()), provider failover fallback, cost & token telemetry, model registry (AI.model("gpt-4o")), AI middleware pipeline, semantic caching, strict tool calling, autonomous agent planner with reflection loops (Planner ➔ Executor ➔ Memory ➔ Reflection), AI safety & moderation (AI.moderate()), Model Context Protocol (MCP) abstraction (AI.mcp()), DevTools AI collector (AiCollector), testing fake (AI.fake()), and 8 provider drivers (OpenAI, Anthropic, Gemini, Ollama, OpenRouter, Groq, Memory, Null).

🌟 Key Capabilities

import { AI } from "@ecfjs/ai";

// 1. Multi-Provider Chat & Streaming
const res = await AI.chat("Summarize enterprise security policies", { driver: "openai" });

// 2. State-Managed Conversation Memory
AI.memory("ticket-102").addMessage("user", "Hello").addMessage("assistant", "Hi there!");

// 3. Prompt Template Versioning
AI.registerPrompt("welcome", "Hello {{name}}, welcome to {{app}}!", "v2");
const text = AI.prompt("welcome@v2", { name: "Jane", app: "ECF" });

// 4. Modular RAG & ReRanker
const ragResult = await AI.rag().execute("How do I configure queues?", { topK: 3 });

// 5. Autonomous Tool-Calling Agent
const agent = AI.agent({
  tools: {
    getWeather: { description: "Fetch weather", execute: async ({ city }) => "Sunny 25C" }
  }
});
const output = await agent.run("Find weather in Tokyo");

📄 License

MIT Licensed.