LangMem

Warnings

• breaking API stability is not guaranteed below version 1.0.0. The library is in early development (0.0.x), and breaking changes may occur frequently between minor versions without extensive prior notice.
  Fix: Refer to the GitHub repository and commit history for recent changes. Pin your dependency versions to specific patch releases (e.g., `langmem==0.0.30`) to mitigate unexpected breaks.
• gotcha Many useful components (e.g., specific embedding models, vector stores) are optional dependencies. You must explicitly install them (e.g., `pip install openai chromadb`) to use them with `langmem`.
  Fix: Consult the `pyproject.toml` or `setup.py` on GitHub for available `extras_require` groups, or manually install individual packages as needed for your chosen components.
• gotcha When initializing vector stores like `Chroma`, ensure you pass the embedding model instance using the correct parameter name, which is `embedding_function`, not `embeddings`.
  Fix: Use `vectorstore = Chroma(embedding_function=embeddings_instance)` to correctly configure your vector store with an embedding model.
• gotcha For cloud-based embedding services (e.g., OpenAIEmbeddings), the necessary API key (e.g., `OPENAI_API_KEY`) must be set in your environment variables or passed directly during initialization. Forgetting this will lead to authentication errors.
  Fix: Set the API key as an environment variable (e.g., `export OPENAI_API_KEY='your_key'`) or pass it directly to the constructor (e.g., `OpenAIEmbeddings(openai_api_key='your_key')`).

Install

• pip install langmem Install core library
• pip install langmem[openai,chromadb] Install with common extras

Imports

• MemoryManager

  from langmem import MemoryManager

• OpenAIEmbeddings

  from langmem.embeddings import OpenAIEmbeddings

• Chroma

  from langmem.vectorstores import Chroma

  The 'vectorstores' submodule is plural.

Quickstart

This quickstart demonstrates how to set up `MemoryManager` with `OpenAIEmbeddings` and `Chroma` to add and recall memories. It assumes `openai` and `chromadb` are installed (e.g., via `pip install langmem[openai,chromadb]`) and `OPENAI_API_KEY` is configured in your environment. Note the use of `embedding_function` when initializing `Chroma`.

import os
from langmem import MemoryManager
from langmem.embeddings import OpenAIEmbeddings
from langmem.vectorstores import Chroma

# NOTE: For this example, you need to install 'openai' and 'chromadb'
# pip install langmem openai chromadb

# Ensure OPENAI_API_KEY is set in your environment variables.
# For testing, you can uncomment and set it directly:
# os.environ['OPENAI_API_KEY'] = 'YOUR_OPENAI_API_KEY'

openai_api_key = os.environ.get('OPENAI_API_KEY', '')
if not openai_api_key:
    print("Warning: OPENAI_API_KEY environment variable not set. OpenAIEmbeddings might fail.")

# Initialize embeddings and vector store
embeddings = OpenAIEmbeddings(openai_api_key=openai_api_key)
vectorstore = Chroma(embedding_function=embeddings) # Chroma requires embedding_function

# Initialize MemoryManager
memory_manager = MemoryManager(vectorstore=vectorstore)

# Add memories
memory_manager.add_memory("The quick brown fox jumps over the lazy dog.")
memory_manager.add_memory("The quick red bird flies high in the sky.")
memory_manager.add_memory("The lazy dog often naps under the oak tree.")

# Recall information
results = memory_manager.recall("What is the fox doing?")
print("\nRecall results for 'What is the fox doing?':", results)

results = memory_manager.recall("Where does the dog nap?")
print("\nRecall results for 'Where does the dog nap?':", results)