novyx-langchain 0.1.1

Persistent semantic memory for LangChain agents using Novyx RAM

novyx-langchain

Persistent semantic memory for LangChain agents.

Your agents forget. Novyx remembers.

from novyx_langchain import NovyxMemory
from langchain.chains import ConversationChain
from langchain_openai import ChatOpenAI

memory = NovyxMemory(api_key="nram_xxx", session_id="user-123")
chain = ConversationChain(llm=ChatOpenAI(), memory=memory)

response = chain.invoke({"input": "My name is Alice"})
# Memory persists across restarts, deployments, and servers

Features

• Semantic retrieval - Finds relevant context, not just recent messages
• LangGraph native - First-class checkpointer support
• Multi-tenant - Built for production SaaS applications
• Tamper-proof - Optional integrity verification (unique to Novyx)
• < 100ms latency - Fast semantic search via Novyx RAM
• CRDT conflict resolution - Handles concurrent writes gracefully

Installation

pip install novyx-langchain

For LangGraph support:

pip install "novyx-langchain[langgraph]"

Quick Start

1. Get an API Key

Sign up at novyxlabs.com to get your API key.

2. Basic Memory

from novyx_langchain import NovyxMemory
from langchain.chains import ConversationChain
from langchain_openai import ChatOpenAI

# Create persistent memory
memory = NovyxMemory(
    api_key="nram_tenant_xxx",
    session_id="user-123",
    k=10,  # Retrieve top 10 relevant memories
)

# Use with any LangChain chain
chain = ConversationChain(
    llm=ChatOpenAI(model="gpt-4"),
    memory=memory,
)

# Conversations persist automatically
response = chain.invoke({"input": "Remember that I prefer dark mode"})
# ... restart your app ...
response = chain.invoke({"input": "What are my preferences?"})
# Agent remembers dark mode preference!

3. LangGraph Checkpointer

from langgraph.graph import StateGraph, END
from novyx_langchain import NovyxCheckpointer

# Create checkpointer
checkpointer = NovyxCheckpointer(api_key="nram_tenant_xxx")

# Build your graph
builder = StateGraph(AgentState)
builder.add_node("agent", agent_node)
builder.add_node("tools", tool_node)
builder.set_entry_point("agent")
builder.add_edge("agent", END)

# Compile with persistence
graph = builder.compile(checkpointer=checkpointer)

# Run with thread_id
config = {"configurable": {"thread_id": "conv-123"}}
result = graph.invoke({"messages": [HumanMessage("Hello!")]}, config)

# State persists across invocations

4. Integrity-Protected Memory (Tamper-Proof)

from novyx_langchain import NovyxIntegrityMemory

# Every write is cryptographically verified
memory = NovyxIntegrityMemory(
    api_key="nram_tenant_xxx",
    integrity_api_key="int_xxx",
    integrity_secret="your_shared_secret",
    session_id="user-123",
)

chain = ConversationChain(llm=llm, memory=memory)

# All writes verified with challenge-response authentication
# Tamper detection, audit trails, forensic rollback
response = chain.invoke({"input": "Store sensitive context"})

API Reference

NovyxMemory

Main memory class, compatible with ConversationBufferMemory.

NovyxMemory(
    api_key: str,              # Novyx RAM API key
    session_id: str,           # Unique session identifier
    k: int = 10,               # Number of memories to retrieve
    semantic_search: bool = True,  # Use semantic (vs recency) retrieval
    return_messages: bool = False,  # Return Message objects vs string
    min_relevance: float = 0.3,    # Minimum relevance score
)

Methods:

• load_memory_variables(inputs) - Load relevant memories
• save_context(inputs, outputs) - Save conversation turn
• clear() - Clear session memory
• search(query, limit) - Semantic search
• store_memory(content, tags, importance) - Store custom memory

NovyxChatMessageHistory

For use with RunnableWithMessageHistory:

from langchain_core.runnables.history import RunnableWithMessageHistory
from novyx_langchain import NovyxChatMessageHistory

def get_session_history(session_id: str):
    return NovyxChatMessageHistory(
        api_key="nram_xxx",
        session_id=session_id,
    )

chain_with_history = RunnableWithMessageHistory(
    chain,
    get_session_history,
)

NovyxCheckpointer

LangGraph checkpoint persistence:

NovyxCheckpointer(
    api_key: str,              # Novyx RAM API key
    namespace: str = "langgraph",  # Namespace for isolation
)

Methods:

• put(config, checkpoint, metadata) - Store checkpoint
• get_tuple(config) - Get checkpoint
• list(config) - List checkpoints
• clear_thread(thread_id) - Clear thread checkpoints

NovyxIntegrityMemory

Tamper-proof memory with cryptographic verification:

NovyxIntegrityMemory(
    api_key: str,              # Novyx RAM API key
    integrity_api_key: str,    # Novyx Integrity API key
    integrity_secret: str,     # Shared secret for challenges
    session_id: str,
    # ... same options as NovyxMemory
)

NovyxBufferedMemory

High-throughput memory with local buffering:

NovyxBufferedMemory(
    api_key: str,
    session_id: str,
    buffer_size: int = 50,     # Flush after N messages
    flush_interval: float = 30.0,  # Or flush every N seconds
)

Examples

Multi-Tenant Setup

def get_memory_for_user(user_id: str) -> NovyxMemory:
    return NovyxMemory(
        api_key=f"nram_{tenant_id}_xxx",  # Tenant-specific key
        session_id=f"user-{user_id}",
        k=20,
    )

# Each user has isolated memory
alice_memory = get_memory_for_user("alice")
bob_memory = get_memory_for_user("bob")

Custom Tags and Importance

memory = NovyxMemory(api_key="nram_xxx", session_id="user-123")

# Store with custom tags and importance
memory.store_memory(
    content="User is a premium subscriber",
    tags=["subscription", "important"],
    importance=9,  # High importance (1-10)
)

# Search by tags
results = memory.search("subscription status", limit=5)

AgentExecutor Integration

from langchain.agents import AgentExecutor, create_openai_functions_agent

memory = NovyxMemory(
    api_key="nram_xxx",
    session_id="agent-session",
    return_messages=True,
)

agent_executor = AgentExecutor(
    agent=agent,
    tools=tools,
    memory=memory,
    verbose=True,
)

result = agent_executor.invoke({"input": "What do you remember about me?"})

Environment Variables

export NOVYX_RAM_API_KEY="nram_tenant_xxx"
export NOVYX_RAM_URL="https://novyx-ram-api.fly.dev/v1"  # Optional
export NOVYX_INTEGRITY_API_KEY="int_xxx"  # For integrity features
export NOVYX_INTEGRITY_SECRET="your_secret"

Comparison

Feature	novyx-langchain	ConversationBufferMemory	Zep	Mem0
Semantic search	✅	❌	✅	✅
Tamper-proof	✅	❌	❌	❌
LangGraph native	✅	❌	❌	❌
Multi-tenant	✅	❌	✅	✅
Self-hostable	✅	N/A	✅	✅
Conflict resolution	✅ CRDT	❌	❌	❌
< 100ms latency	✅	✅	✅	✅

License

MIT License - see LICENSE for details.

Links

• Documentation
• GitHub
• Novyx RAM API
• Novyx Integrity
• Discord