Shared state conceptual docs #1958
Conversation
rlancemartin
force-pushed
the
rlm/memory-service-concepts
branch
from
fcbb71f
to
2844d38
Compare
rlancemartin
force-pushed
the
rlm/memory-service-concepts
branch
from
2844d38
to
02533a9
Compare
docs/docs/concepts/memory.md
Outdated
| Persistence is critical sustaining a long-running chat sessions. For example, a chat between a user and an AI assistant may have interruptions. Persistence ensures that a user can continue that particular chat session at any later point in time. However, what happens if a user initiates a new chat session with an assistant? This spawns a new thread, and the information from the previous session (thread) is not retained. This motivates the need for a memory service that can maintain data across chat sessions (threads). | ||
| Persistence is critical sustaining a long-running chat sessions. For example, a chat between a user and an AI assistant may have interruptions. Persistence ensures that a user can continue that particular chat session at any later point in time. However, what happens if a user initiates a new chat session with an assistant? This spawns a new thread, and the information from the previous session (thread) is not retained. This motivates the need for a memory that can maintain data across chat sessions (threads). | ||
|
|
||
| For this, we can use LangGraph's `Store` interface to save and retrieve information across threads. Shared information can be namespaced by, for example, `user_id` to retain user-specific information across threads. See more detail in the [persistence conceptual guide](https://langchain-ai.github.io/langgraph/concepts/persistence/#persistence) and this [how-to guide on shared state](../how-tos/memory/shared-state.ipynb). |
There was a problem hiding this comment.
if you have a code snippet for the checkpointer, i would do the same for here
also, now that we have this, i would add more concrete examples for the two below
also id maybe rename "metaprompting" to "update own instructions" or something?
docs/docs/concepts/memory.md
Outdated
|
|
||
| ## Meta-prompting | ||
| For this, we can use LangGraph's `Store` interface to save and retrieve information across threads. Shared information can be namespaced by, for example, `user_id` to retain user-specific information across threads. Let's show how to use the `Store` interface to save and retrieve information. |
There was a problem hiding this comment.
Do we have an API reference that we can link Store interface to?
docs/docs/concepts/memory.md
Outdated
| namespace_for_memory = (user_id, "memories") | ||
|
|
||
| # Save memories | ||
| memory_id = str(uuid.uuid4()) |
There was a problem hiding this comment.
If there's a single sentence answer to the question why is the key not a tuple? Why has the key been broken into a namespace (tuple) and a string.
docs/docs/concepts/memory.md
Outdated
| in_memory_store.put(namespace_for_memory, memory_id, memory) | ||
|
|
||
| # Retrieve memories | ||
| memories = in_memory_store.search(namespace_for_memory) |
There was a problem hiding this comment.
What does search return without a query?
There was a problem hiding this comment.
it always returns a list of items. Default is listing
| namespace_for_memory = (user_id, "memories") | ||
| ``` | ||
|
|
||
| We use the `store.put` to save memories to our namespace in the store. When we do this, we specify the namespace, as defined above, and a key-value pair for the memory: the key is simply a unique identifier for the memory (`memory_id`) and the value (a dictionary) is the memory itself. |
There was a problem hiding this comment.
The full key words out to be:
(user_id, "memories", "some_random_uuid", "food_preference")Does structuring memories as:
{
'type': "food_preference",
"value": "I like pizza"
}work with respect to the memory store API? (e.g., does search work as expected etc.)
cc @hinthornw
There was a problem hiding this comment.
You can either put food preference in the namespace (so you can search in that directory) or as a key and do filter of {"type", "food_preference"}
More flexible if it's in the body
| in_memory_store.put(namespace_for_memory, memory_id, memory) | ||
| ``` | ||
|
|
||
| We can read out memories in our namespace using `store.search`, which will return all memories for a given user as a list. The most recent memory is the last in the list. |
There was a problem hiding this comment.
- This looks like a
listAPI. Doessearchsupport queries? - Is it all memories or is there a page size?
There was a problem hiding this comment.
not yet - just filtering, limit, offset.
search query & additional things to be added
|
|
||
| # Let's say hi again | ||
| for update in graph.stream( | ||
| {"messages": [{"role": "user", "content": "hi, tell me about my memories"}]}, config, stream_mode="updates" |
There was a problem hiding this comment.
The input into the graph is messages in the format of dicts.
I think that the schema of the graph is MessagesState -- which doesn't support the dict notation probably?
class MessagesState(TypedDict):
messages: Annotated[list[AnyMessage], add_messages]AnyMessage = Annotated[
Union[
Annotated[AIMessage, Tag(tag="ai")],
Annotated[HumanMessage, Tag(tag="human")],
Annotated[ChatMessage, Tag(tag="chat")],
Annotated[SystemMessage, Tag(tag="system")],
Annotated[FunctionMessage, Tag(tag="function")],
Annotated[ToolMessage, Tag(tag="tool")],
Annotated[AIMessageChunk, Tag(tag="AIMessageChunk")],
Annotated[HumanMessageChunk, Tag(tag="HumanMessageChunk")],
Annotated[ChatMessageChunk, Tag(tag="ChatMessageChunk")],
Annotated[SystemMessageChunk, Tag(tag="SystemMessageChunk")],
Annotated[FunctionMessageChunk, Tag(tag="FunctionMessageChunk")],
Annotated[ToolMessageChunk, Tag(tag="ToolMessageChunk")],
],
Field(discriminator=Discriminator(_get_type)),
]There was a problem hiding this comment.
cc @vbarda / @hinthornw IMO this is a problem if we start using dict notation
There was a problem hiding this comment.
I just ran into mypy anger from this type of thing where BaseMessgae isn't included either.
I think we can use from langgraph.graph.mesages import Messages which is a union of that and "messagelikedict" or something
It's frustrating that core doesn't have a single type for this
|
@rlancemartin I think this is looking great! Just a few nits and tagged @hinthornw in a few places for clarification on API |
No description provided.