Technology · Agentic → Memory · Open Source · wiki:deep

Letta

Taxonomy

Letta (formerly MemGPT) builds stateful agents with memory that can learn and improve over time. Active development has moved: current source lives in letta-ai/letta-code (agent harness, terminal UI, App Server, channels, desktop/web runtime). The letta-ai/letta repo README points there; an archive branch holds the retired Letta V1 API server.

Install/run today via npm install -g @letta-ai/letta-code → letta (TUI) / letta server, plus desktop, chat.letta.com, Slack/Telegram/Discord channels, Agent SDK, and Letta Cloud for cross-device memory/identity.

GitHub · letta-ai/lettaDocsWatch · officialStack: agentic-memory

Why it matters here

Memory for an agentic OS is not only “store embeddings” — it is agents that keep identity and memory across sessions while tools and channels change. Letta is the research pack for MemGPT-lineage stateful agents and self-hosted/cloud runtimes. Prefer LangMem when you only need LangGraph store tools; prefer Mem0/Zep when you want a memory service plugged into many frameworks.

How it works

  1. Install @letta-ai/letta-code (or use desktop / chat.letta.com / Cloud).
  2. Run interactive letta or letta server for local/self-hosted agents.
  3. Connect channels (Slack, Telegram, Discord, custom) or embed via Agent SDK.
  4. Memory/identity persist per Letta’s runtime (Cloud keeps them across machines).
  5. Do not build new work on the archived V1 API in letta’s archive branch.

Related: mem0 · langmem · zep · topics/31-memory-architecture

Flow

When to reach for it

  • Use when: you want stateful MemGPT-style agents with a first-party harness, channels, and optional Cloud.
  • Skip when: you only need a memory library behind an existing LangGraph/Crew agent (langmem / mem0), or a temporal knowledge graph (graphiti / Zep).
  • Prefer instead: letta-code over historical letta tags for new installs; Mem0 for framework-agnostic memory API.

Limits

  • Repo split: starring/cloning letta-ai/letta alone misses the active harness — follow letta-code.
  • V1 API retired: reproducibility tags remain; active projects must migrate.
  • Cloud vs self-host: cross-computer memory nudges toward Cloud — clarify tenancy before production.
  • Not the company ledger: agent memory ≠ declared process/authority.

What we checked

Claims below are backed by science sources on disk.

Platform / stateful agents

letta-ai/letta repository · MODERATE

“The platform for building stateful agents”

MemGPT research lineage

MemGPT: Towards LLMs as Operating Systems · STRONG

“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.”

Research site / positioning

MemGPT research site · MODERATE

“Use Letta to create coding agents, personal assistants, AI coworkers, and stateful agents embedded in your own applications.”
Research inventory

9 tags · 12 out · 13 in · 3 artifacts · 1 gaps · 34 corpus docs

Catalog tags

landscape.layer
Agentic
landscape.subcategory
Memory
license_tag
Open Source
maps.dm
absent
maps.features
9
one_liner
Agentic
review.depth
science
slug
letta
title
Letta

Artifacts

  • dm_map · absent
  • features_map · present · technologies/letta/features.md
  • readme · present · technologies/letta/README.md

Out · alternative_to

Out · maps_to

In · alternative_to

In · in_stack

In · maps_to

Corpus tags

category
Agentic → Memory
dedication
open-source
feature
learning-adaptation
memory-state-persistence
os-company-model-memory
r2-10-sleep-improvement
r2-31-memory-architecture
r2-35-context-engineering
r2-36-importance-consolidation
r2-master-synthesis
wt-data-context
kind
company_review
map_edge
tech_features
tech_quote
tech_readme
tech_science_source
tech_section
needs_deepen
false
quality
ok
section
Evidence
Features map
GitHub map
How it works
Limits & failure modes
Links
Scientific notes
What it is
When to use / skip
Why it matters here
slug
letta
source_id
arxiv-2310-08560
letta-github
memgpt-research-site
surface
interface-review
technology
letta

Corpus documents (34)

company_review · 2

  • Letta · interface review
  • Letta

map_edge · 9

  • letta → learning-adaptation
  • letta → memory-state-persistence
  • letta → os-company-model-memory
  • letta → r2-10-sleep-improvement
  • letta → r2-31-memory-architecture
  • letta → r2-35-context-engineering
  • letta → r2-36-importance-consolidation
  • letta → r2-master-synthesis
  • … +1 more

tech_features · 1

  • Letta · features

tech_quote · 8

  • Letta · arxiv-2310-08560
  • Letta · arxiv-2310-08560
  • Letta · arxiv-2310-08560
  • Letta · letta-github
  • Letta · letta-github
  • Letta · letta-github
  • Letta · memgpt-research-site
  • Letta · memgpt-research-site

tech_readme · 1

  • Letta

tech_science_source · 3

  • Letta · arxiv-2310-08560
  • Letta · letta-github
  • Letta · memgpt-research-site

tech_section · 10

  • Letta · Evidence
  • Letta · Features map
  • Letta · GitHub map
  • Letta · How it works
  • Letta · Limits & failure modes
  • Letta · Links
  • Letta · Scientific notes
  • Letta · What it is
  • … +2 more