Python · Transformers · TRL · Apache-2.0 base

Models

Compact local agents that improve through a measured recursive self-improvement loop: better traces, sharper rewards, stronger checkpoints.

Karti-Small-RSI-3B is the first family: a private-weight local agent built from SmolLM3-3B for Veronica, Hermes, reliable tool calling, and offline operation. Its public Hugging Face repository records the open-base lineage and adaptation recipe while each monthly cycle turns reviewed traces and held-out evaluation into the next checkpoint.

Read about Lumbridge Models