langgraph-checkpoint-hana

LangGraph checkpoint saver for SAP HANA Cloud — persist AI agent state where your enterprise data already lives.

This package implements the LangGraph BaseCheckpointSaver interface for SAP HANA Cloud, enabling conversation memory, human-in-the-loop workflows, time travel, and fault recovery for LangGraph agents running on SAP BTP.

Why HANA?

LangGraph provides official checkpointers for PostgreSQL, SQLite, and Redis. But if your enterprise runs on SAP, your business data is in HANA — and your AI agents should persist their state there too.

Benefits of keeping agent state in HANA Cloud:

• Co-located with business data — agent state lives alongside financial documents, master data, and transactional records. One database to manage, back up, and secure.
• Enterprise-grade security — leverage HANA's native audit logging, role-based access control, and encryption at rest.
• BTP-native deployment — deploy LangGraph agents on SAP BTP (Kyma / Cloud Foundry) with HANA Cloud as the persistence layer. No external database dependencies.
• Unified querying — join agent decision traces with business data using standard SQL.

Installation

pip install langgraph-checkpoint-hana

Requirements:

• Python 3.10+
• SAP HANA Cloud instance (or HANA on-premise 2.0 SPS 05+)
• hdbcli driver (installed automatically)

Quick Start

from langgraph_checkpoint_hana import HANASaver
from langgraph.graph import StateGraph, MessagesState, START, END

# Create checkpointer from connection parameters
with HANASaver.from_conn_info(
    address="your-instance.hanacloud.ondemand.com",
    port=443,
    user="DBADMIN",
    password="your-password",
) as checkpointer:
    
    # Build your graph
    workflow = StateGraph(MessagesState)
    workflow.add_node("chatbot", chatbot_node)
    workflow.add_edge(START, "chatbot")
    workflow.add_edge("chatbot", END)
    
    # Compile with HANA persistence
    graph = workflow.compile(checkpointer=checkpointer)
    
    # Invoke — state is automatically persisted to HANA
    config = {"configurable": {"thread_id": "user-session-42"}}
    result = graph.invoke(
        {"messages": [("human", "What's our AP aging?")]},
        config,
    )
    
    # State survives restarts — resume anytime
    state = graph.get_state(config)

Usage with Environment Variables

Convenient for containerised deployments on Kyma or Cloud Foundry:

import os
os.environ["HANA_HOST"] = "your-instance.hanacloud.ondemand.com"
os.environ["HANA_PORT"] = "443"
os.environ["HANA_USER"] = "DBADMIN"
os.environ["HANA_PASSWORD"] = "your-password"

checkpointer = HANASaver.from_env()
checkpointer.setup()

Usage with Existing Connection

If you already manage HANA connections in your application (e.g. via a connection pool or SAP CAP service bindings):

from hdbcli import dbapi
from langgraph_checkpoint_hana import HANASaver

conn = dbapi.connect(address="...", port=443, user="...", password="...")
checkpointer = HANASaver(conn=conn)
checkpointer.setup()  # creates tables if needed

# Use with your graph
graph = workflow.compile(checkpointer=checkpointer)

Tables Created

HANASaver.setup() creates two tables (if they don't already exist):

Table	Purpose
LANGGRAPH_CHECKPOINTS	Stores graph state snapshots (one row per checkpoint)
LANGGRAPH_CHECKPOINT_WRITES	Stores pending writes for fault recovery

Both tables use NCLOB columns for serialised data and are keyed by (thread_id, checkpoint_ns, checkpoint_id).

Thread Management

# Delete all state for a thread (cleanup, GDPR, error recovery)
checkpointer.delete_thread("user-session-42")

# List checkpoint history
for cp in checkpointer.list(config, limit=10):
    print(cp.checkpoint["id"], cp.metadata)

# Time travel — get a specific checkpoint
historical_config = {
    "configurable": {
        "thread_id": "user-session-42",
        "checkpoint_id": "1ef4f797-8335-6428-8001-8a1503f9b875",
    }
}
past_state = graph.get_state(historical_config)

Async Support

The hdbcli driver is synchronous. Async methods (aget_tuple, alist, aput, aput_writes, adelete_thread) delegate to their sync counterparts. For high-concurrency async deployments, consider wrapping calls with asyncio.to_thread().

Development

git clone https://github.com/stubborncoder/langgraph-checkpoint-hana.git
cd langgraph-checkpoint-hana
pip install -e ".[dev]"
pytest

Compatibility

Component	Tested Versions
LangGraph	0.2.x, 0.3.x
langgraph-checkpoint	2.x
SAP HANA Cloud	2024.x, 2025.x
Python	3.10, 3.11, 3.12

License

MIT

Contributing

Contributions welcome. Please open an issue to discuss before submitting a PR.