neurosleepnet 0.3.1

NeuroSleepNet — Sleep-inspired persistent memory for AI agents. import nsn; nsn.init(); agent = nsn.wrap(your_slm)

NeuroSleepNet Python SDK

A sleep-inspired hybrid memory layer for continual AI learning. Give your local LLMs and agents infinite memory with a single line of code.

Installation

pip install neurosleepnet

For local LLM support (HuggingFace, Torch):

pip install "neurosleepnet[local_llm]"

Quick Start

import neurosleepnet as nsn

# 1. Initialize — no API keys required for local use
nsn.init(project="my-agent-v1")

# 2. Wrap your agent (OpenAI, LangChain, HuggingFace, etc.)
# All memory injection and storage becomes transparent.
agent = nsn.wrap(your_agent)

# 3. Use your agent as normal
response = agent("What did we talk about in our last session?")

Core Features

• Project Scoping: Isolate memories by project or agent identity.
• Attention-based Retrieval: Hybrid semantic search + recency weighting.
• Sleep Consolidation: Nightly background pruning and reinforcement of important facts.
• Local-First: Built-in SQLite fallback ensures your agent never crashes, even if the backend is down.
• Zero-Ops: Designed to run entirely on your own infrastructure.

Advanced API

# Manually remember a fact
nsn.remember("User prefers Python", importance=0.9, tags=["pref"])

# Retrieve memories semantically
memories = nsn.recall("coding preferences", top_k=3)

# Diagnostics
nsn.status()

# Explain why a memory was retrieved
nsn.explain_last()

# Export/Import full state
state = nsn.snapshot()
nsn.restore(state)

License

Apache 2.0