Installation and setup

Installation and setup is the one-time step that installs LangMem 0.0.30 with LangChain's Groq and Google integrations and gives it two free API keys: one for the chat model and one for embeddings.

Last updated: 30 Sep, 2026 · LangMem 0.0.30

LangMem does not call a model provider itself. It hands the work to a LangChain chat model, and the store it writes to can embed text with a LangChain embedding model. So the install is LangMem plus one integration package per provider.

Installing LangMem and the integrations

langmem brings LangGraph and LangChain with it. langchain-groq is the chat model, langchain-google-genai the embeddings, and numpy speeds up the store's vector search (without it the store prints a warning and falls back to plain Python).

pip install "langmem==0.0.30" "langchain-groq==1.1.3" "langchain-google-genai==4.4.0" numpy

from importlib.metadata import version

for package in ["langmem", "langgraph", "langchain", "langchain-groq", "langchain-google-genai"]:
    print(f"{package:24} {version(package)}")

langmem                  0.0.30
langgraph                1.2.12
langchain                1.4.3
langchain-groq           1.1.3
langchain-google-genai   4.4.0

The version is pinned because LangMem 0.0.x may change its API between releases; LangGraph and LangChain came in as its dependencies, at the versions shown.

Getting the two free keys

Both are free plans with daily limits. Set the keys in the shell you run Python from, or in Colab's Secrets panel:

export GROQ_API_KEY="gsk_..."
export GOOGLE_API_KEY="AIza..."

Checking the chat model

init_chat_model turns a "provider:model" string into a chat model object. Every lesson starts with this line; temperature=0 makes the model pick its most likely answer, which keeps runs closer to each other.

from langchain.chat_models import init_chat_model

model = init_chat_model("groq:openai/gpt-oss-120b", temperature=0)

from langchain.chat_models import init_chat_model

model = init_chat_model("groq:openai/gpt-oss-120b", temperature=0)
reply = model.invoke("Reply with one word: ready")
print(reply.text)

ready

Checking the embedding model

init_embeddings does the same for embedding models. gemini-embedding-2 turns any text into a list of 3072 numbers; the store uses those to search by meaning.

from langchain.embeddings import init_embeddings

embeddings = init_embeddings("google_genai:gemini-embedding-2")
vector = embeddings.embed_query("Order A-1001 arrived broken")
print(len(vector))

3072

gpt-oss-120b vs gpt-oss-20b

When you come back to this setup

• Every lesson's code starts with the same imports and the model = ... line; they only work once the keys are set in that shell or notebook.
• When Groq answers 429 with tokens per day in the message, change the model string to groq:openai/gpt-oss-20b for the rest of the day.
• A new Colab session forgets environment variables; run the Secrets cell again.

Related

• Previous: LangMem overview
• Next: Extracting memories
• Reference: langmem on PyPI