TUI Settlers of Catan built with Llamafile and Bonsai PrismML Models

Name: TUI Settlers of Catan built with Llamafile and Bonsai PrismML Models
Availability: InStock
Author: river_otter

by river_otter·Apr 6, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidCozyRabbit Hole

Runs completely offline with llamafile but LLMs make unreliable Catan negotiation partners.

Strengths

•Entirely offline with llamafile integration requiring zero API keys or setup.
•Agent of Empires hook system tracks game state alongside coding agent sessions.
•Rust implementation with 1B and 8B Bonsai models is technically competent.

Weaknesses

•Novelty project with no staying power beyond initial curiosity factor.
•LLM hallucination makes AI opponents unpredictable for strategy game logic.

Post Description

Hey all, Nathan here from mozilla.ai. I run a lot of agents at once with the agent-of-empires tool, and thought it would be fun if I could also play my favorite board game and have its status tracked just like with Claude Code sessions.

So I built settl: it's a Settlers of Catan style game that runs completely as a TUI, and uses a local LLM (mozilla.ai llamafile!) as the AI agent.

Built in Rust, using the cool new 1B and 8B Bonsai Models by PrismML that my colleague Davide helped turn into a llamafile for dead simple usage of LLMs in the app.

Install: brew install --cask mozilla-ai/tap/settl Source: https://github.com/mozilla-ai/settl Docs: https://mozilla-ai.github.io/settl/

Enjoy and let me know if you got feature requests! I'd like to add the ability to argue with agents to try to get them to accept your trades, and would also consider some sort of framework for faster iteration on agent abilities (maybe using Andrej Karpathy's autoresearch)