X
Letta's next phase
@Letta_AIarticlex
TL;DR. Letta is pivoting to "Letta Code," an open, model-agnostic agent harness built around persistent memory in git-backed files (MemFS) and computer use, while sunsetting its older server-side memory abstractions.
Takeaways
- Memory moves from bespoke DB-backed tools to generalized filesystem ops over "context repositories" — agents edit memory with bash, not custom APIs.
- Positioning bet: while OpenAI, Anthropic, and Google bundle agent harnesses with their own models, Letta wants to be the open runtime where you own the memory and pick the model.
- Casualties of the pivot: Letta Filesystem, legacy memory tools (
core_memory_replace), Templates, and server-side Identities all get replaced by client-side / SDK equivalents.