Vector Memory¶

The vector memory module uses vector search (backed by a vector db) to retrieve relevant conversation items given a user input.

This notebook shows you how to use the VectorMemory class. We show you how to use its individual functions. A typical usecase for vector memory is as a long-term memory storage of chat messages. You can

Initialize and Experiment with Memory Module¶

Here we initialize a raw memory module and demonstrate its functions - to put and retrieve from ChatMessage objects.

• Note that retriever_kwargs is the same args you'd specify on the VectorIndexRetriever or from index.as_retriever(..).

In [ ]:

from llama_index.core.memory import VectorMemory
from llama_index.embeddings.openai import OpenAIEmbedding


vector_memory = VectorMemory.from_defaults(
    vector_store=None,  # leave as None to use default in-memory vector store
    embed_model=OpenAIEmbedding(),
    retriever_kwargs={"similarity_top_k": 1},
)

from llama_index.core.memory import VectorMemory from llama_index.embeddings.openai import OpenAIEmbedding vector_memory = VectorMemory.from_defaults( vector_store=None, # leave as None to use default in-memory vector store embed_model=OpenAIEmbedding(), retriever_kwargs={"similarity_top_k": 1}, )

In [ ]:

from llama_index.core.llms import ChatMessage

msgs = [
    ChatMessage.from_str("Jerry likes juice.", "user"),
    ChatMessage.from_str("Bob likes burgers.", "user"),
    ChatMessage.from_str("Alice likes apples.", "user"),
]

from llama_index.core.llms import ChatMessage msgs = [ ChatMessage.from_str("Jerry likes juice.", "user"), ChatMessage.from_str("Bob likes burgers.", "user"), ChatMessage.from_str("Alice likes apples.", "user"), ]

In [ ]:

# load into memory
for m in msgs:
    vector_memory.put(m)

# load into memory for m in msgs: vector_memory.put(m)

In [ ]:

# retrieve from memory
msgs = vector_memory.get("What does Jerry like?")
msgs

# retrieve from memory msgs = vector_memory.get("What does Jerry like?") msgs

Out[ ]:

[ChatMessage(role=<MessageRole.USER: 'user'>, content='Jerry likes juice.', additional_kwargs={})]

In [ ]:

vector_memory.reset()

vector_memory.reset()

Now let's try resetting and trying again. This time, we'll add an assistant message. Note that user/assistant messages are bundled by default.

In [ ]:

msgs = [
    ChatMessage.from_str("Jerry likes burgers.", "user"),
    ChatMessage.from_str("Bob likes apples.", "user"),
    ChatMessage.from_str("Indeed, Bob likes apples.", "assistant"),
    ChatMessage.from_str("Alice likes juice.", "user"),
]
vector_memory.set(msgs)

msgs = [ ChatMessage.from_str("Jerry likes burgers.", "user"), ChatMessage.from_str("Bob likes apples.", "user"), ChatMessage.from_str("Indeed, Bob likes apples.", "assistant"), ChatMessage.from_str("Alice likes juice.", "user"), ] vector_memory.set(msgs)

In [ ]:

msgs = vector_memory.get("What does Bob like?")
msgs

msgs = vector_memory.get("What does Bob like?") msgs

Out[ ]:

[ChatMessage(role=<MessageRole.USER: 'user'>, content='Bob likes apples.', additional_kwargs={}),
 ChatMessage(role=<MessageRole.ASSISTANT: 'assistant'>, content='Indeed, Bob likes apples.', additional_kwargs={})]