shipwithjev

Catalog / Games & real time

0056GitHub

minecraft-agent

An Astra planner with Jev as the controller, playing Minecraft.

The planner decides what to do next, and Jev turns that into the control decisions the game needs. The repository ships native recording, tested routes and run verification, so a run can be replayed and checked.

rmalde/minecraft-agentREADME ↗
# Astra and JEV Minecraft agent

This project uses GPT-6 Astra or GPT-5.6 Sol to plan and JEV to select player actions in Minecraft Java 1.16.5. It uses the official vanilla server and Mineflayer. A read-only Java sensor can report the exact dragon head position. It does not change game rules or entity state. Each selected action is sent to the game through the normal player protocol.

## Latest verified result

The latest run, `nether-final-08`, completed in **8 minutes 43.300 seconds**. The previous video took 14 minutes 31.800 seconds. The new run was 40% shorter. The End combat stage took 152 seconds instead of 332 seconds. Dragon flight and landing times can vary between runs.

The agent started with an empty inventory in a fresh Survival/Peaceful world. Astra planned and JEV selected the actions. It used the Nether for travel, killed the dragon with six bed explosions on the first landing, and reached the exit with full health and no deaths. There was no restart, source change, or operator repair during the run.

The run used 131 JEV decisions and 35 Astra calls. All 17 run checks, eight route/camera/screen checks, and 29 local tests passed. The full earlier video and action log were reviewed before the local tests and new recording.

Camera control uses continuous turns, with a limit of 240 degrees per second on each axis and acceleration of 960 degrees per second squared. A hidden native Minecraft client renders the game without desktop input. The capture includes health, hunger, hotbar, held items, inventory, crafting, and native animations.

The video is 960 × 540 at 20 frames per second, without audio. A translucent full-width top banner uses Menlo and shows the model names, elapsed time in milliseconds, XYZ coordinates, planner objective, and selected action

Also filed under Games & real time

  1. 0535

    What Jev Probably Is, And Why You Already Have One

    A company called TypeSafe AI launched a model this week called Jev. They told us how fast it is. They told us what it costs. They told us it plays Doom in a tenth of a second.…

    medium.com · Games & real time

  2. 0534

    TypeSafe’s Jev Makes AI Decisions Fast Enough to Play Doom

    TypeSafe’s Jev Makes ... Decisions Fast Enough to Play Doom | by Kristopher Dunham | Sep, 2026 | Medium ... # TypeSafe’s Jev Makes AI Decisions Fast Enough to Play Doom ... A…

    medium.com · Games & real time

  3. 0533

    🧵 I wanted to understand what makes Jev different from a normal LLM

    🧵 I wanted to understand what makes Jev different from a normal LLM. So I built a tiny real-time game with it. The AI controls the player and has one goal: Don't get hit.

    @abolbuild · Games & real time

  4. 0532

    Searching for Better Game Recommendations with Jev

    To pick that list better, I tried Jev, a decision model from TypeSafe. I did not have Jev write recommendations. I asked it to rate candidate games, then handed the model that…

    dev.to · Games & real time