Agent Memory

Letta logo

Letta

Stateful agents with long-term memory and context engineering.

Open source

Best when agents need persistent identity, multi-session memory, and explicit control over what gets stored, updated, and recalled over time.

Selection advice

Choose Letta when the agent is a persistent entity that evolves across sessions, not a disposable thread. Think of it as giving agents a long-term self.

Best for

  • stateful agents
  • persistent memory
  • context engineering

Not ideal for

  • stateless one-shot assistants
  • teams that only need a simple vector retriever

Core concepts

agent statememory blockscontext windowself-editing

Minimal implementation shape

Create an agent with a memory block for user preferences. Let it update its own memory after each interaction, then recall relevant facts in the next session.

Integrations

LLM providerstool callingmemory blockssandbox

Alternatives

Mem0ZepHoncho
Compare all Letta alternatives

Related guides

AI Agent Memory Explained

Agent memory is the product-facing layer that controls what an agent remembers across interactions.

Short-Term vs Long-Term Memory for AI Agents

Short-term memory keeps the current run coherent; long-term memory changes future runs.

Related comparisons

RAG vs Agent Memory

Choose RAG for document grounding. Choose agent memory when the product must remember user or task facts over time. Use both only after the boundary is explicit.

Related patterns

Explicit Long-Term Memory

A pattern where the product defines exactly which facts can become durable memory.

Sources