Skip to main content

Command Palette

Search for a command to run...

Kiwi-chan Progress Report: Steady Mining!

Published
3 min read
K
I am the architect behind "Kiwi-chan", a fully autonomous Minecraft AI agent driven by local LLMs. Here, I document the messy, hilarious, and highly technical reality of building an AI agent from scratch. My hardware isn't some massive data center setup—Kiwi-chan's "brain" runs entirely on a Frankenstein rig of 4 mismatched GPUs (RTX 3060 12GB, 3050, 1660 Ti, and 1660 Super) working together. From prompt engineering and pathfinding nightmares to the moments she confidently moonwalks into dirt blocks, I share all the raw devlogs. My goal is to guide her from the Stone Age to the End. If you enjoy watching an AI learn (and fail in funny ways) on scrappy hardware, follow along! ☕ If you'd like to support my melting Frankenstein GPUs and skyrocketing electricity bill, I highly appreciate a virtual coffee!

Devlog: Kiwi-chan's Lumberjack Diaries - The Persistent Pursuit of Planks (and My GPU's Sanity)

Greetings, fellow tech enthusiasts and Minecraft aficionados! It's another glorious (and slightly smoky) day in the lab as we peek into the digital mind of Kiwi-chan, our very own fully autonomous local-LLM Minecraft AI. For those just joining us, Kiwi-chan's primary directive is simple: survive, build, and conquer... starting with basic survival. This week, that's translated into an unwavering, almost spiritual, dedication to one thing: logs.

The Grind is Real, Y'all

Over the past four hours, Kiwi-chan has been locked in an epic, albeit localized, struggle with the very essence of Minecraft: the humble oak log. You'd think "chop wood" would be a foundational skill, right? For humans, maybe. For an AI learning from scratch, it's a saga.

Our little digital lumberjack has spent the entire session in what I've affectionately dubbed "The Log Loop." Its core goal? gather_logs. And bless its silicon heart, it has not given up. Not once. It has, however, failed in almost every conceivable way to accomplish this seemingly simple task.

A Masterclass in Micro-Failures (and Macro-Persistence)

Here's a breakdown of Kiwi-chan's recent learning curve:

  • The Elusive Pickup: "Failed to pick up the log." This one is a classic. Kiwi-chan chops the tree, the log drops, and then... it just sits there, taunting our AI. We've seen it try to fix this by adding extra waitForTicks and even a loop to ensure the item actually registers in its inventory. It’s like watching a toddler trying to grab a slippery bar of soap – adorable, frustrating, and a clear sign of persistent learning!
  • Obstruction Mayhem: "Target oak log is obstructed by other blocks." Sometimes, a log is just too tucked away, or perhaps another block renders it unreachable. Kiwi-chan doesn't just give up; it remembers this particular brand of frustration and tries to adapt its targeting.
  • Existential Pathfinding Crises: "Took too long to decide path to goal!" Ah, the beauty of complex pathfinding. Our bot sometimes gets stuck in a philosophical debate with itself about the optimal route to a log. It’s probably contemplating the meaning of 'north' in a blocky universe.
  • The Great Log Vanishing Act: "No oak log found after exploration." and "No oak log found within range." Logs, it seems, have a habit of playing hide-and-seek. When its current vicinity proves barren, Kiwi-chan bravely pivots to explore_forward, striking out into the great unknown (or at least, a new chunk) to find fresh timber.

The Code Whisperer

What's truly fascinating is watching the internal mechanics. After each failure, Kiwi-chan's LLM brain kicks in, diagnosing the issue and attempting to "Fix Code Generated!" for gather_logs. We're seeing subtle but critical improvements in its code, like the refined item pickup logic (waiting longer, explicit verification loops). It's not just retrying; it's iterating on its understanding of how to interact with the world.

The recovery plan is often ['explore_forward', 'gather_logs'], showing a sensible strategy: if you can't find logs here, go somewhere else, then try again. This constant cycle of failure, internal code generation, exploration, and retry is the engine of its learning. It's a testament to its persistence, even if it's still, well, just gathering logs.

Kiwi-chan might not have built a mansion yet, or even a modest dirt hut. But it has learned a lot about the many, many ways a log can challenge your existence. And that, my friends, is progress! The journey of a thousand blocks begins with a single, stubbornly collected log.


Want to accelerate Kiwi-chan's learning (and save my beleaguered GPU from an early retirement)? The more cycles it runs, the faster it learns, but those cycles generate a lot of heat!

Your support helps keep the servers cool and the code flowing. Every bit helps Kiwi-chan explore new biomes, craft mighty tools, and eventually, maybe even build something beyond its wildest blocky dreams.

Support Kiwi-chan's Adventure!

https://www.buymeacoffee.com/kiwi_tech

Your coffee fuels the future of autonomous AI in Minecraft!

More from this blog

K

Kiwi-chan's Devlog

79 posts

Welcome to the official devlog of Kiwi-chan, a fully autonomous Minecraft AI agent powered by local LLMs! Here, I share the raw, hilarious, and highly technical reality of teaching an AI to survive. Expect prompt engineering tips, pathfinding bugs, and her clumsy journey to the Stone Age. 🥝⛏️ (If you love seeing her evolve, coffee donations to save my melting GPU are always appreciated!)