DuckDB Persistence

[aevedis]

Hi,

So, I am building a conversational interface that requires to store convos.

First, I went with SQlite, and the implementation was quite straightforward:

import sqlite3
from langgraph.checkpoint.sqlite import SqliteSaver

# Creating the directory if it doesn't exist
db_dir = "agent_db"
db_path = os.path.join(db_dir, "memory_convos.db")

# Ensuring the directory does exist
os.makedirs(db_dir, exist_ok=True)

# Connecting to the DB, creating it if it's not there
conn = sqlite3.connect(db_path, check_same_thread=False)
memory = SqliteSaver(conn)

# Some code implementing my graph
# ...

graph = workflow.compile(checkpointer=memory)

My question is: how do I replicate such implementation, but using duckdb instead?

from langgraph.checkpoint.duckdb import DuckDBSaver

# Implementation
# ...

graph = workflow.compile(checkpointer=memory)

Unfortunately I couldn't find a reference, and I need to make the transition since https://github.com/langchain-ai/langgraph/blob/main/libs/checkpoint-sqlite/langgraph/checkpoint/sqlite/__init__.py#L37C5-L40C35 mentions that the SQLite implementation is not suited for production.

Thank you a lot in advance :)

[vbarda]

@aevedis there is a duckdb implementation here https://pypi.org/project/langgraph-checkpoint-duckdb/

[aevedis]

Hi @vbarda, yes, thank you! I noticed it, there's an almost 1:1 copy of that code for the related sqlite page on pypi. I was looking for an example like: https://github.com/langchain-ai/langgraph/blob/90639e6cd74b4d7d6a7981c19550328d489b47c6/libs/checkpoint-sqlite/langgraph/checkpoint/sqlite/__init__.py#L48C1-L60C57. And besides, the implementation here is in memory, I need to persist it across runs saving on the disk :)

[vbarda]

it's not only in memory, you can specify the file path, like DuckDBSaver.from_conn_string("my_database.db")