paradoxlf 0.15.0

A latent memory and active inference engine for AI agents.

Paradox: Latent Memory & Simulation Engine

Paradox is a lightweight, hardware-agnostic cognitive architecture for AI agents. It provides a dynamic "Latent Memory" that doesn't just store data but allows for active simulation, evolution, and reasoning.

🎯 The Main Point

"Extreme Efficiency through Abstraction."

We have too much data and not enough hardware. Paradox solves this by replacing heavy "Real Objects" with lightweight "Latent Vectors", allowing you to perform Supercomputer-scale tasks on a Laptop.

• Don't store the Cake (100MB Object).
• Store the Recipe (1KB Vector).
• Bake it on demand.

🚀 Key Features

• Multimodal Intelligence (v0.7.0): Unified encoding for Images and Text using CLIP.
• Semantic Proximity (v0.8.0): Weighted "Attention" search to prioritize specific features (e.g., Color vs Shape).
• Latent Reasoning (v0.9.0): Perform concept arithmetic (King - Man + Woman = Queen) directly in vector space.
• Temporal Intelligence (v0.10.0): Track thought trajectories and predict future states.
• Intelligence APIs (v0.11.0): High-level methods like imagine(), predict_future(), and conceptual_search().
• Hybrid Compute: Automatically runs on GPU (PyTorch) if available, gracefully falls back to CPU (NumPy/MMap).

📦 Installation

git clone https://github.com/ethcocoder/paradoxlf.git
cd paradoxlf
pip install .[ai,ui]

⚡ Quick Start: Intelligence Layer

from paradox.engine import LatentMemoryEngine
from paradox.media.clip_module import CLIPEncoder

# 1. Initialize the Brain
encoder = CLIPEncoder() # Loads CLIP Model
engine = LatentMemoryEngine(dimension=encoder.dimension)
engine.set_encoder(encoder)

# 2. Learn Concepts
engine.add("Telephone", {"name": "Telephone"})
engine.add("Computer", {"name": "Computer"})
engine.add("Smartphone", {"name": "Smartphone"})

# 3. Imagine New Concepts (Blending)
# What is half phone, half computer?
new_idea = engine.imagine("Telephone", "Computer", ratio=0.5)

# 4. Search for Meaning
results = engine.conceptual_search(new_idea, k=1)
print(f"Imagined Concept is closest to: {results[0][2]['name']}")

🧠 Advanced Capabilities

1. Temporal Prediction (Forecasting)

Predict where a sequence of thoughts or video frames is heading.

history = [vector_t0, vector_t1, vector_t2]
future_vector = engine.predict_future(history, steps=1)

2. Semantic Search with Attention

Search for "Red Car", but tell the engine that Color is 10x more important than Shape.

weights = [10.0, 1.0, ...] # Heavy weight on first dimensions
results = engine.query(query_vec, weights=weights)

3. Visual Dashboard

Explore your memory space interactively.

streamlit run paradox/ui/dashboard.py

🌍 Innovation Impact

Paradox is a fundamental engine for Massive Scale Simulation:

Domain	Problem	Paradox Solution
Cognitive AI	LLMs are stateless/expensive.	Paradox provides a cheap, evolvable long-term memory.
Scientific Sim	Simulating millions of particles is slow.	Latent physics allows interacting with millions of entities.
Big Data	Searching billions of logs is slow.	Proximity search finds anomalies instantly (O(1) approx).

🌐 Distributed & Networked Memory (v0.12.0+)

Paradox can now scale horizontally across multiple processes or machines.

1. Local Cluster (Multi-Threaded)

Simulate a distributed system on a single machine.

from paradox.distributed import LatentCluster
cluster = LatentCluster(num_shards=4)
cluster.add(vector) # Round-robin distribution
cluster.query(vector) # Map-reduce query

2. Networked Memory (Client-Server)

Run shards on different servers (Cloud/Edge).

Server (Node A):

# Start a shard server on port 8000
python -c "from paradox.distributed import start_server; start_server(port=8000)"

Client (Node B):

from paradox.distributed import RemoteShard
client = RemoteShard(host="192.168.1.5", port=8000)
client.add(vector)

Hybrid Cluster: Combine local and remote shards into one brain.

cluster = LatentCluster(num_shards=0)
cluster.shards.append(RemoteShard(host="node_1"))
cluster.shards.append(RemoteShard(host="node_2"))
# Now queries search the entire network!

🤝 Contributing

Open source contributions are welcome. Please submit a PR for review.

📄 License

MIT License