V33X Memory Engine

Persistent AI Memory for Autonomous Trading Agents

Built on Cognee | Customized by @Vision33X

---

What This Is

An AI memory engine that never forgets. Feed it market data, trade history, whale movements, funding rates, news — it builds a living knowledge graph that connects everything and gets smarter over time.

This isn't RAG. This is a knowledge engine that understands relationships between data points, tracks how they change, and gives you answers that basic vector search can't.

Why I Forked This

Every AI trading system I've built loses context when sessions end. Memory compaction destroys nuance. Important patterns vanish. This solves that problem permanently.

My Use Cases:

• Store every trade signal, execution, and outcome — learn what actually works
• Track whale wallet movements across chains and connect them to price action
• Build persistent memory for autonomous trading agents (DQN + Transformer)
• Cross-reference funding rates, liquidation data, and kill zone timing
• Never lose context between AI sessions again

Quick Start

import cognee
import asyncio

async def main():
    # Feed it your data
    await cognee.add("BTC RSI hit 92 at 3PM ET kill zone. Shorted 20x. Hit TP in 4 minutes.")

    # Build the knowledge graph
    await cognee.cognify()

    # Query it later — it remembers everything
    results = await cognee.search("What happens when RSI > 90 during kill zones?")
    print(results)

asyncio.run(main())

The Stack

Layer	Tool	Purpose
AI Memory	Cognee (this repo)	Knowledge graphs + vector search
LLM	Claude / GPT-4o / Local (LM Studio)	Entity extraction + reasoning
Graph DB	Kuzu (local) / Neo4j	Relationship storage
Vector DB	LanceDB (local)	Semantic search
Hardware	RTX 5090 + 192GB RAM	Local inference + ML training

What It Supports

• 30+ data sources — PDFs, text, images (OCR), audio (transcription), URLs, code files
• 13 search modes — graph traversal, RAG, chain-of-thought, temporal, Cypher queries
• 8 LLM providers — OpenAI, Claude, Gemini, Ollama, Mistral, Bedrock, Groq, custom
• Multiple databases — Kuzu, Neo4j, LanceDB, ChromaDB, PGVector, PostgreSQL
• Multi-tenant — isolated knowledge bases per project/strategy
• MCP Server — direct integration with Claude Code and other MCP clients
• Runs local — Ollama + Kuzu + LanceDB = zero API costs on your own hardware

Installation

pip install cognee

Set your LLM key:

export LLM_API_KEY="your_key_here"

For local-only (no API costs):

export LLM_PROVIDER=ollama
export LLM_MODEL=llama3.1:8b
export LLM_ENDPOINT=http://localhost:11434/v1

Architecture

Raw Data (market feeds, trade logs, research, news)
    ↓
ADD — Ingest & classify documents
    ↓
COGNIFY — Extract entities, build knowledge graph
    ↓
SEARCH — Query with 13 retrieval strategies
    ↓
MEMIFY — Enrich with rules and context
    ↓
Living Knowledge Graph (grows smarter every cycle)

Currently Building

• V33X Autonomous Learner — DQN + Transformer that trades 24/7 and learns from its own results
• V33X Signal Engine — CoinGlass funding + RSI extremes + kill zone timing
• Whale Intelligence System — Track smart money across chains in real-time
• BloFin Copy Trading — Revenue from copiers seeing consistent green

Other V33X Projects

Repo	Description
V33X-Autonomous-Learner	Self-learning trading AI (PyTorch + DQN)
V33X-RSI-Scanner	Multi-exchange RSI extreme scanner
V33X-Pine-Scripts	TradingView indicators
V33X-Whale-Shield	Whale movement tracker
crypto-kill-zones	Kill zone timing analysis

---

Built during a bear market. Building for the next bull.

Follow @Vision33X | BloFin Partner

---

Based on Cognee by Topoteretes — Apache 2.0 License. Original documentation and full feature set available in their repo.