LlamaIndex Integration

LlamaIndex Integration

Use SuperLocalMemory as a local-first chat store backend for LlamaIndex applications. All data stays on your machine — zero cloud calls, zero API keys for the memory layer.

Installation

Prerequisites: SuperLocalMemory installed (npm install -g superlocalmemory or ./install.sh)

pip install llama-index-storage-chat-store-superlocalmemory

Quick Start

from llama_index.storage.chat_store.superlocalmemory import SuperLocalMemoryChatStore
from llama_index.core.memory import ChatMemoryBuffer
from llama_index.core.llms import ChatMessage, MessageRole

# Create chat store (backed by local SQLite)
chat_store = SuperLocalMemoryChatStore()

# Use with ChatMemoryBuffer
memory = ChatMemoryBuffer.from_defaults(
    token_limit=3000,
    chat_store=chat_store,
    chat_store_key="user-123",
)

Usage with SimpleChatEngine

from llama_index.core.chat_engine import SimpleChatEngine
from llama_index.core.memory import ChatMemoryBuffer
from llama_index.storage.chat_store.superlocalmemory import SuperLocalMemoryChatStore

chat_store = SuperLocalMemoryChatStore()

memory = ChatMemoryBuffer.from_defaults(
    token_limit=3000,
    chat_store=chat_store,
    chat_store_key="project-frontend",
)

chat_engine = SimpleChatEngine(
    memory=memory,
    llm=your_llm,
    prefix_messages=[],
)

response = chat_engine.chat("How do I center a div?")

Direct Store Operations

from llama_index.storage.chat_store.superlocalmemory import SuperLocalMemoryChatStore
from llama_index.core.llms import ChatMessage, MessageRole

store = SuperLocalMemoryChatStore()

# Add messages
store.add_message("session-1", ChatMessage(role=MessageRole.USER, content="Hello"))
store.add_message("session-1", ChatMessage(role=MessageRole.ASSISTANT, content="Hi there!"))

# Get messages
messages = store.get_messages("session-1")

# List all session keys
keys = store.get_keys()  # ["session-1"]

# Delete operations
store.delete_last_message("session-1")     # Remove last message
store.delete_message("session-1", 0)       # Remove by index
store.delete_messages("session-1")         # Remove all for key

# Replace all messages at once
store.set_messages("session-1", [
    ChatMessage(role=MessageRole.USER, content="Fresh start"),
])

API Reference

SuperLocalMemoryChatStore

Parameter	Type	Default	Description
db_path	Optional[str]	~/.claude-memory/memory.db	Path to SQLite database

Methods (all 7 BaseChatStore abstract methods):

Method	Returns	Description
set_messages(key, messages)	None	Replace all messages for a key
get_messages(key)	List[ChatMessage]	Get messages ordered by creation time
add_message(key, message)	None	Add a single message
delete_messages(key)	Optional[List[ChatMessage]]	Delete all messages, return deleted
delete_message(key, idx)	Optional[ChatMessage]	Delete by zero-based index
delete_last_message(key)	Optional[ChatMessage]	Delete most recent message
get_keys()	List[str]	List all session keys with data

How It Works

• Each message stored as a memory entry in SLM's SQLite database
• Session keys are hashed (SHA-256) for tags to stay within SLM's 50-char tag limit
• Full session key is preserved inside the serialized JSON for lossless round-tripping
• All MessageRole types supported: USER, ASSISTANT, SYSTEM, TOOL
• Importance set to 3 (lower than user memories) to avoid crowding search results

Links

• GitHub Repository
• Full Documentation
• LangChain Integration
• Example Script