> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wazoo.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# LangMem

> Worlds vs LangMem: framework-native memory for LangGraph agents against a shared fact graph.

## At a glance

|                 | Worlds                        | [LangMem](https://langchain-ai.github.io/langmem/) |
| :-------------- | :---------------------------- | :------------------------------------------------- |
| Data model      | RDF triples in a ledger       | Vector embeddings                                  |
| How facts enter | Import, patches, agent writes | Agent tools write memory                           |
| Retrieval       | Hybrid search + SPARQL        | Similarity search                                  |
| Provenance      | Fact-level ledger             | None                                               |
| Deployment      | Hosted beta or embedded       | SDK inside LangGraph                               |

## The philosophical difference

[LangMem](https://langchain-ai.github.io/langmem/) is the memory SDK for the
LangGraph ecosystem. It stores embeddings in a vector index and hands the agent
tools to write, search, and update its own memory at runtime. Memory is
self-managed: the model decides what is worth keeping.

Worlds is storage-side. Facts enter through import and patches, relationships
are explicit, and the ledger preserves the history of every change. Agents can
write new facts through the update API, and each write lands in the append-only
ledger as a verifiable patch ([update](/worlds/update)). The difference is
custody. LangMem trusts the model to curate its memory; Worlds curates at the
data layer.

## When Worlds fits

* Facts must survive outside a single framework or agent lifetime.
* Answers need to be verifiable rather than similar.
* You want the same knowledge queryable from more than one framework.

## When Worlds does not fit

* You want memory that lives inside your LangGraph application with no separate
  graph backend to configure.
* You want the model to decide what memory is worth keeping, with no ledger and
  no data-layer custody.
* Embeddings-only recall meets your requirements.

## Coexistence

Use LangMem for the agent's working memory inside a LangGraph application, and
Worlds for the durable facts the agent is allowed to act on. The
[LangGraph integration](/integrations/langgraph) shows how a LangGraph agent
calls Worlds for verified context.