Why this exists

[DomLynch]

Most agent frameworks became heavier than the thing they were supposed to help build.

NanoLetta is the opposite approach: take the useful cognitive loop from Letta and make it small enough to understand, run, and modify directly.

No server. No Postgres. No Docker. No platform layer.

Just the core loop:

message -> model -> tool call -> memory edit -> persist -> repeat

This is meant for people building their own agents who want the reasoning and memory mechanics without adopting a whole infrastructure stack.

Repo: https://github.com/DomLynch/NanoLetta