Stateful Agents with Letta.ai
Source: https://ludovicobessi.com/stateful-agents-letta Author: Ludovico Bessi Date: 2025-06-29
Summary
Tutorial on building stateful agents with Letta.ai (formerly MemGPT). Practical guide to using Letta’s memory architecture for persistent agent state.
Key Claims
- Letta’s model: agents have named memory “slots” that persist across sessions. Developer defines the schema; the agent writes to and reads from them.
- Key capability: agents can have multi-turn conversations where each turn builds on previous context, even across separate API calls.
- Use case: customer service agents that remember customer preferences, history, and previous issues.
- Technical implementation: memory is stored in a database; at each turn, relevant memory is retrieved and injected into the prompt.
- Comparison: simpler than Context Repositories (file-based) — better for conversational applications.
Entities
Concepts
- Agent Memory — Letta’s slot-based persistent memory
- Context Engineering — memory retrieval and injection