Enabling state persistence for LangGraph agents

Edit online

Enable state persistence with checkpoints for LangGraph agents. Configure persistence in your agent's YAML file by using the checkpointer field.

checkpointer field section

The checkpointer field section is optional and enables state persistence across conversation turns. If omitted, your agent operates in stateless mode.

Checkpointer types

When you specify the checkpointer field section, you can configure the checkpointer with the following options:

  • checkpointer.type

    The type of checkpointer to use:

    • memory - In-memory storage for development and testing. State is lost on restart.
      • Use case: Development, testing, temporary sessions
      • Requirements: None
      • Persistence: Lost on agent restart
    • sqlite - File-based persistence using SQLite. State is lost on agent pod restart.
      • Use case: Single-instance deployments, testing
      • Requirements: Add langgraph-checkpoint-sqlite to requirements.txt
      • Persistence: Lost on pod restart
    • postgres - Database persistence using PostgreSQL for production use.
      • Use case: Production deployments requiring durable state
      • Requirements: Add langgraph-checkpoint-postgres to requirements.txt
      • Persistence: Durable across restarts
      • Additional configuration: Requires connection_string_key
  • checkpointer.connection_string_key

    Field required if you use the postgres type. The app ID of the connection containing the PostgreSQL database connection string.

Setup steps for PostgreSQL database persistence

  1. Create a connection with credential type api_key
  2. Provide your PostgreSQL connection string as the API key value
  3. Use the connection's app ID as the value for connection_string_key
  4. Select this connection during the agent import process

The system automatically manages database connections and schema setup at runtime.

Example configuration

Agent with postgres checkpointer

spec_version: v1
kind: agent
name: stateful_agent
title: Stateful Support Agent
description: Agent with persistent state using PostgreSQL
deployment:
  code_bundle:
    entrypoint: agent:create_agent
checkpointer:
  type: postgres
  connection_string_key: postgres_db

requirements.txt:

langgraph>=0.2.0
langchain-core>=0.3.0
langgraph-checkpoint-postgres

In this example, postgres_db is the app ID of a connection you must create before importing. This connection must have credential type api_key and contain your PostgreSQL database connection string. You select this connection during the import process.

Agent with SQLite checkpointer

spec_version: v1
kind: agent
name: sqlite_agent
title: SQLite Stateful Agent
description: Agent with SQLite state persistence
deployment:
  code_bundle:
    entrypoint: agent:create_agent
checkpointer:
  type: sqlite

requirements.txt:

langgraph>=0.2.0
langchain-core>=0.3.0
langgraph-checkpoint-sqlite

Agent with memory checkpointer

spec_version: v1
kind: agent
name: dev_agent
title: Development Agent
description: Agent with in-memory state for testing
deployment:
  code_bundle:
    entrypoint: agent:create_agent
checkpointer:
  type: memory

No additional dependencies required for memory checkpointer beyond the base LangGraph requirements.

What to do next