Technology · Agentic → Memory · Open Source · wiki:deep
LangMem
LangMem helps agents learn and adapt from interactions over time. It extracts important information from conversations, supports prompt refinement, and maintains long-term memory via functional primitives (any storage) plus native integration with LangGraph’s Long-term Memory Store (default on LangGraph Platform).
Core pieces: a storage-agnostic memory API; hot-path tools (create_manage_memory_tool, create_search_memory_tool) agents call during chat; and a background memory manager that extracts, consolidates, and updates knowledge. Install: pip install -U langmem (MIT).
Why it matters here
An agentic OS that forgets preferences and prior case context forces humans to re-type truth. LangMem is the LangGraph-native memory layer: agents record/search in-session, and background jobs consolidate — without inventing a second memory product. Prefer it when the runtime is already LangGraph; prefer Mem0/Zep/Letta when you need a standalone memory service or graph timeline.
Memory here is assistant state, not the company ledger.
How it works
- Attach a LangGraph
BaseStore(dev:InMemoryStore; prod: Postgres/similar). - Give the agent manage/search memory tools (hot path) and/or enable background memory manager.
- Conversations trigger store/search; embeddings index memories for retrieval.
- Persist across restarts only if the store is durable — in-memory is lost on process exit.
Related: langgraph · mem0 · zep · topics/31-memory-architecture
When to reach for it
- Use when: agents run on LangGraph and need hot-path + background long-term memory.
- Skip when: you need a multi-tenant hosted memory SaaS first (mem0 / Zep Cloud), MemGPT-style OS agents (letta), or only short-term thread state.
- Prefer instead: Mem0 for drop-in memory API across frameworks; Graphiti/Zep for temporal knowledge graphs.
Limits
- Store durability:
InMemoryStoreis for demos — production needs DB-backed store. - LLM dependency: extraction/tools need a provider key; quality tracks the model.
- LangGraph gravity: primitives work elsewhere, but native path assumes LangGraph store semantics.
- Not system of record: preferences ≠ declared company configuration or authority.
What we checked
Claims below are backed by science sources on disk.
Product / LangGraph memory tooling
langchain-ai/langmem · MODERATE
“LangMem helps agents learn and adapt from their interactions over time.”
Contrast vs Letta
OS-style memory contrast · MODERATE
“To enable using context beyond limited context windows, we propose virtual context management, a technique drawing inspiration from hierarchical memory systems in traditional operating systems which provide the illusion of an extended virtual memory via paging between physical memory and disk.”
Contrast vs Mem0
Mem0 contrast · MODERATE
“We introduce Mem0, a scalable memory-centric architecture that addresses this issue by dynamically extracting, consolidating, and retrieving salient information from ongoing conversations.”
Research inventory
9 tags · 11 out · 12 in · 3 artifacts · 1 gaps · 32 corpus docs
Catalog tags
- landscape.layer
- Agentic
- landscape.subcategory
- Memory
- license_tag
- Open Source
- maps.dm
- absent
- maps.features
- 8
- one_liner
- Agentic
- review.depth
- science
- slug
- langmem
- title
- LangMem
Artifacts
- dm_map · absent
- features_map · present · technologies/langmem/features.md
- readme · present · technologies/langmem/README.md
Out · alternative_to
Out · maps_to
- learning-adaptationfeature
- memory-state-persistencefeature
- os-company-model-memoryfeature
- r2-10-sleep-improvementfeature
- r2-31-memory-architecturefeature
- r2-35-context-engineeringfeature
- r2-36-importance-consolidationfeature
- wt-data-contextfeature
In · alternative_to
In · in_stack
- agentic-memorystack · primary
In · maps_to
- learning-adaptationfeature
- memory-state-persistencefeature
- os-company-model-memoryfeature
- r2-10-sleep-improvementfeature
- r2-31-memory-architecturefeature
- r2-35-context-engineeringfeature
- r2-36-importance-consolidationfeature
- wt-data-contextfeature
Corpus tags
- category
- Agentic → Memory
- dedication
- open-source
- feature
- learning-adaptationmemory-state-persistenceos-company-model-memoryr2-10-sleep-improvementr2-31-memory-architecturer2-35-context-engineeringr2-36-importance-consolidationwt-data-context
- kind
- map_edgetech_featurestech_quotetech_readmetech_science_sourcetech_section
- needs_deepen
- false
- quality
- ok
- section
- EvidenceFeatures mapGitHub mapHow it worksLimits & failure modesLinksScientific notesWhat it isWhen to use / skipWhy it matters here
- slug
- langmem
- source_id
- langmem-githubletta-contrastmem0-contrast
- technology
- langmem
Corpus documents (32)
map_edge · 8
- langmem → learning-adaptation
- langmem → memory-state-persistence
- langmem → os-company-model-memory
- langmem → r2-10-sleep-improvement
- langmem → r2-31-memory-architecture
- langmem → r2-35-context-engineering
- langmem → r2-36-importance-consolidation
- langmem → wt-data-context
tech_features · 1
- LangMem · features
tech_quote · 9
- LangMem · langmem-github
- LangMem · langmem-github
- LangMem · langmem-github
- LangMem · letta-contrast
- LangMem · letta-contrast
- LangMem · letta-contrast
- LangMem · mem0-contrast
- LangMem · mem0-contrast
- … +1 more
tech_readme · 1
- LangMem
tech_science_source · 3
- LangMem · langmem-github
- LangMem · letta-contrast
- LangMem · mem0-contrast
tech_section · 10
- LangMem · Evidence
- LangMem · Features map
- LangMem · GitHub map
- LangMem · How it works
- LangMem · Limits & failure modes
- LangMem · Links
- LangMem · Scientific notes
- LangMem · What it is
- … +2 more