Enx (enx)

Real-time external memory and cognitive continuity layer for LLMs, autonomous agents, and model pretraining workflows.

Overview

Enx is a high-performance, real-time memory synchronization and cognitive continuity engine designed to bridge state across sessions, training checkpoints, and distributed agentic pipelines.

By leveraging FastEmbed (ONNX-backed), Enxi operates with zero PyTorch runtime overhead, delivering ultra-fast vector embeddings and surgical prompt contextualization out-of-the-box.

Key Features

• Zero PyTorch Overhead: Utilizes fastembed with ONNX runtime for ultra-lightweight, high-speed embedding generation without heavy deep learning framework dependencies.
• Multi-Tier Vector Storage: Built-in native connectors for Qdrant and Milvus to handle hot cache and cold archival memory tiers seamlessly.
• Dynamic Context Bridge: Instantly augments raw user prompts and training checkpoints with relevant historical memories.
• Production-Ready API Gateway: Powered by FastAPI and Pydantic v2 for high-throughput, validated async gateway communication.

Installation

Install Enx globally or inside your virtual environment straight from PyPI:

pip install enxi

Quickstart Usage
Here is how to initialize the Enxi SDK client to store, recall, and contextualize memories in your application:

Python
from sdk.client import EnxClient

# Initialize the client layer
client = EnxClient()

# 1. Store (Remember) a new memory point
point_id = client.remember(
    memory_id="node-session-001",
    content="Cognitive continuity layer successfully synchronized across nodes.",
    tier="cold",
    metadata={"module": "core", "sovereign_stack": True}
)
print(f"Committed memory with UUID: {point_id}")

# 2. Recall memories via semantic search
results = client.recall(
    query_text="How is the cognitive state synchronized?",
    tier="cold",
    limit=2
)
for idx, hit in enumerate(results, 1):
    print(f"{idx}. Score: {hit['score']:.3f} | Content: {hit['payload']['content']}")

# 3. Contextualize (Augment prompts or training loops)
augmented = client.contextualize(
    user_prompt="Explain the sovereign memory architecture.",
    tier="cold",
    limit=2
)
print(f"Augmented Prompt Payload:\n{augmented['augmented_prompt']}")
Project Structure
Plaintext
enxi/
├── pyproject.toml
├── README.md
├── src/
│   ├── api/          # FastAPI routers and gateway endpoints
│   ├── core/         # Engine, bridging, scraper, and consolidators
│   ├── sdk/          # Client SDK for seamless integration
│   └── storage/      # Qdrant and vector store connectors
└── examples/         # Pretraining hooks and SDK demo scripts