Skip to main content

LangChain and LangGraph

Use PowerMem as a persistent memory layer for LangChain and LangGraph applications.

Prerequisites​

  • Python 3.11+.
  • PowerMem configured with your LLM provider, API key, and model.
  • LangChain or LangGraph dependencies for your app.

Install​

For LangChain:

pip install powermem langchain langchain-core langchain-openai

For LangGraph:

pip install powermem langgraph langchain-core langchain-openai

For the runnable healthcare support example:

cd examples/langchain
pip install -r requirements.txt

LangChain pattern​

PowerMem is used beside LangChain as a durable retrieval and write-back layer:

  1. Create a PowerMem Memory instance.
  2. Search PowerMem before each response to load relevant context.
  3. Inject the retrieved context into a LangChain prompt or LCEL chain.
  4. Save the user/assistant exchange back into PowerMem.

Minimal shape:

from powermem import Memory, auto_config

memory = Memory(config=auto_config())

context = memory.search(
    query="what does the user prefer",
    user_id="user123",
    limit=5,
)

memory.add(
    "User prefers concise answers with examples.",
    user_id="user123",
)

For a full LCEL implementation, see examples/langchain/README.md and ../examples/scenario_5_custom_integration.md.

LangGraph pattern​

PowerMem works well as a graph node or helper in LangGraph workflows:

  1. Add a load-context node that searches PowerMem.
  2. Add the retrieved memories to graph state.
  3. Generate the response with state-aware context.
  4. Add a save-memory node after the response.

See the LangGraph section in ../guides/0009-integrations.md for a complete example.

Verify​

  1. Add a probe memory for a test user_id.
  2. Search for the probe before invoking the chain or graph.
  3. Confirm the retrieved memory is included in the generated prompt/context.
  4. Save a new exchange and search it back.

Troubleshooting​

  • If search returns no results, confirm the same user_id is used for writes and searches.
  • If extraction creates no memory, check LLM provider configuration and logs.
  • If LangChain imports fail, install the packages listed in examples/langchain/requirements.txt.
  • If embeddings fail, confirm your embedding provider and dimensions match the configured storage backend.

See also​