langgraph checkpoints vs stores.stub

---

id: langgraph-checkpoints-vs-stores name: LangGraph Checkpoints vs Stores repo_path: ~/Documents/sw30labs/repos/langgraph-checkpoints-vs-stores remote_url: https://github.com/sw30labs/langgraph-checkpoints-vs-stores.git primary_language: Python framework: LangGraph category: agentic-frameworks stacks: [langgraph, agentic] ingested: 2026-07-09 last_commit_date: 2026-07-09

LangGraph Checkpoints vs Stores

Purpose

Runnable, deterministic reference explaining the difference between LangGraph checkpoints (thread-scoped graph state — resume, time travel, interrupts, fault tolerance) and stores (cross-thread, long-term memory — user facts, preferences, shared knowledge). Ships a source-verified use-case guide mapping real scenarios and production case studies (Replit, Klarna, Uber, LinkedIn) to the right persistence mechanism.

Architecture

Real StateGraph, InMemorySaver, and InMemoryStore examples with tests and CI. No LLM calls and no API keys — everything runs offline. Every quoted output is captured from a real run by scripts/generate_artifacts.py (only volatile values like checkpoint ids/timestamps redacted). Includes Mermaid decision flowcharts, concept/experiment/production-notes docs, and a comparison matrix CSV. Three chapters: ch1 — in-memory concepts (InMemorySaver/InMemoryStore); ch2 — production persistence: kill-and-resume across PIDs from a SQLite file, plain-SQL peek at both layers in one DB, offline semantic store.search, and a backend matrix running one graph unchanged on memory/SQLite/Postgres/Redis (docker-compose on non-clashing ports 5442/6390, graceful unreachable reporting); ch3 — human-in-the-loop interrupts and time travel.

Dependencies

langgraph (>=1.2.8), langgraph-checkpoint-sqlite, rich, typing-extensions. Docker Compose + Makefile provided.

CI/CD

GitHub Actions ci.yml; Ruff style; pytest; backend-integration job runs chapter-2 tests against real Postgres/Redis service containers (auto-skip locally). Apache-2.0.

Entities and topics

• agent-stack — the layered agentic mental model this persistence question sits inside
• wiki-vs-rag — sibling deterministic, offline-verifiable LangGraph teaching repo
• Tags: [python, langgraph, cli]