Quickstart

There are three paths to a running Memory instance:

1. DAB starter (recommended) — one command bootstraps the entire reference architecture
2. Manual setup — wire up a Memory directly in Python
3. LangChain integration — drop into RunnableWithMessageHistory or a retrieval chain

---

DAB starter (recommended)

The Databricks Asset Bundle starter provisions Unity Catalog tables, Vector Search indexes, and a working chat agent notebook in your workspace:

databricks bundle init https://github.com/travis-burmaster/lakehouse-memory \
  --template-dir templates/lakehouse-memory-bundle \
  --output-dir my-memory-demo
cd my-memory-demo
databricks bundle deploy
databricks bundle run setup_job

You will be prompted for:

• Unity Catalog catalog and schema names
• Vector Search endpoint name
• SQL Warehouse HTTP path
• LLM serving endpoint name

After setup_job completes (~15 minutes end-to-end, mostly Vector Search index provisioning), open notebooks/02_chat_agent.ipynb and run all cells.

---

Manual setup

from lakehouse_memory import Memory, MemoryConfig, Scope
import os

mem = Memory.from_databricks(
    catalog="my_catalog",
    schema_name="agent_memory",
    workspace_url=os.environ["DATABRICKS_HOST"],
    access_token=os.environ["DATABRICKS_TOKEN"],
    http_path=os.environ["DATABRICKS_HTTP_PATH"],
    vector_search_endpoint=os.environ["DATABRICKS_VECTOR_SEARCH_ENDPOINT"],
)

# Idempotently create UC tables + Vector Search indexes
mem.provision()

# Scope to a specific user and session
scoped = mem.with_scope(user_id="u_1", session_id="s_1")

# Write a semantic fact
scoped.semantic.upsert(fact="User prefers SQL over Python.")

# Delta Sync indexes are triggered — fire the sync after writes
scoped.semantic._index.trigger_sync()

# Wait for sync (production code uses exponential backoff)
import time; time.sleep(15)

# Retrieve similar facts
facts = scoped.semantic.retrieve("language preferences", k=3)
print(facts)

---

LangChain integration

from langchain_core.runnables.history import RunnableWithMessageHistory

# Wire up the LangChain adapters
chat_history = mem.as_langchain_chat_history(limit=50)
retriever = mem.as_langchain_retriever(k=5)

# Use in a chain
chain_with_history = RunnableWithMessageHistory(
    your_chain,
    lambda session_id: mem.with_scope(session_id=session_id).as_langchain_chat_history(),
    input_messages_key="input",
    history_messages_key="history",
)

The LakehouseChatHistory adapter persists each turn as an episodic event. The LakehouseSemanticRetriever adapter returns Document objects from the semantic store.

Scope per session

Call mem.with_scope(session_id="...") before constructing LangChain adapters to isolate history and retrieval per conversation.