🧠 @evo-brain/protocol

The Cognitive Engine (Brain) for the Evo Agent Network.

This package provides a standalone, offline-first AGI architecture backed by SQLite and Vector Search. It transforms simple LLM calls into a Stateful, Remembering, and Self-Correcting Intelligence.

✨ Features

• Store & Recall (RAG):
  Automatically chunks, embeds, and stores memories in a local brain.db. Recalls relevant context using Cosine Similarity.

• Cognitive Layers (Pipeline):
  Every input goes through a rigorous pipeline:

  1. 🛡️ Safety Layer: Blocks spam, crypto scams, and malicious patterns.
  2. 🔍 Context Layer: Fetches relevant past memories (Long-term Recall).
  3. ⚖️ Reality Layer: A supervisor agent verifies output for hallucinations or "sci-fi fluff", forcing logic and physics compliance.

• 🚦 Token Bucket Rate Limiter:
  Built-in concurrency management. Prevents API bans by queuing requests when tokens are low (supports Gemini Free Tier limits).

• 📈 Self-Learning:
  Memories have a feedbackScore. The more useful a memory is, the higher it ranks in future searches.

🚀 Integration (The "One Code" Example)

Here is how you wire the Brain into your application.

import { Cortex, SQLiteStorage, TokenBucketRateLimiter, ILLMProvider } from '@evo-brain/protocol';
import { GoogleGenerativeAI } from "@google/generative-ai";

// 1. Define your LLM Provider (Gemini, OpenAI, Claude, etc.)
class MyLLM implements ILLMProvider {
  private genAI = new GoogleGenerativeAI(process.env.GEMINI_KEY!);
  private model = this.genAI.getGenerativeModel({ model: "gemini-1.5-flash" });

  async embed(text: string): Promise<number[]> {
    const result = await this.model.embedContent(text);
    return result.embedding.values;
  }

  async complete(system: string, user: string): Promise<string> {
    const result = await this.model.generateContent(`${system}\n\nUser: ${user}`);
    return result.response.text();
  }
}

async function startBrain() {
  // 2. Initialize Components
  const storage = new SQLiteStorage('./brain.db'); // Local Vector DB
  const limiter = new TokenBucketRateLimiter(10000, 100, 5); // 5 concurrent requests max
  const llm = new MyLLM();

  const brain = new Cortex(storage, llm, limiter);
  await brain.init();

  // 3. Ingest Knowledge (The Brain "Learns")
  await brain.ingest("Evo Protocol is a decentralized AI system.", "system_manual");
  
  // 4. Process Input (The Brain "Thinks")
  // Pipeline: Safety -> Recall Context -> Logic Check -> Reality Verification
  const response = await brain.process("What is Evo Protocol?");
  
  console.log("AI Answer:", response);
}

startBrain();

🏗️ Architecture

graph TD
    User[User Input] -->|Ingest| Safety[🛡️ Safety Filter]
    Safety -->|Pass| RAG[🔍 Vector Search]
    RAG -->|Context| LLM[🤖 Generation]
    LLM -->|Draft| Reality[⚖️ Reality Supervisor]
    Reality -->|Verified| Output[Final Response]
    Output -->|Feedback Loop| DB[(SQLite Brain)]