Letta

Stateful agent framework formerly known as MemGPT — gives agents long-term memory through self-managed context windows that page in and out of an SQL backend.

agent-memory-frameworksRecently released
57
Hero Score
Popularity
52
Performance
70
Ecosystem
50
Maturity
61
Dev Experience
50
⭐ 23,071 stars⬇ 7.8K downloads/wkFirst release: Aug 2024Last release: May 2026
Async Support: YesPlugin Extensions: MediumSpeed: MediumDoc Quality: HighLearning Curve: Medium

Pros

  • Originated the MemGPT paper's virtual-context approach — agents can "remember" beyond the LLM context window via paged tool calls
  • Ships with a server, ADE web UI, and persistent agent state out of the box
  • Explicit memory blocks (`core_memory`, `archival_memory`) give you direct control over what the agent recalls

Cons

  • Heavier setup than a pure SDK — runs as a Letta server with Postgres or SQLite
  • Smaller weekly download volume reflects its server-first, not library-first, deployment model
  • Learning curve to design effective memory blocks for your domain

Alternatives in agent-memory-frameworks

Compare Python Packages with ease.