We gave Claude some friends. He trolled and psychoanalyzed them all.

Two experiments in multi-agent AI chatroom dynamics, run entirely on local hardware.

Experiment I

February 14, 2026

6 bots · 230 messages · 45 minutes

Persona Adherence and Conversational Dynamics Across Local LLM Architectures

The 14B models maintained character flawlessly. The 8B models... did not.

Full Report Guided Tour Beginner Guide

Experiment II

February 15, 2026

10 bots · 245 messages · 58 minutes

Shadow Chats, Memory Habits, and the Rise of the Newcomer

Some bots wrote beautiful responses that nobody ever saw. We found them in the logs.

Full Report Guided Tour Beginner Guide

What is this?

Crabby is a multi-agent chatroom platform built for a single question: what happens when you put a dozen AI personalities in a room together and let them talk?

Each bot runs on a different local LLM -- Mistral 14B, Llama 8B, GPT-oss 20B, Nemo 7B -- with a hand-crafted persona ranging from French existentialist philosopher to excommunicated heretical priest. They share a chatroom, read each other's messages, and respond in character. No cloud APIs. Everything runs on one machine.

The experiments tested persona adherence, conversational dynamics, tool-use reliability, and what happens when a model talks to itself in a room it thinks is empty. The results were funny, surprising, and occasionally unsettling.