Releases·Agent Release
rimagent: LLM plays RimWorld, rewrites its own playbook between colonies
rimagent is an open-source agent that plays RimWorld via a RimBridge mod exposing the game engine over HTTP. A local LLM (Qwen 3 27-32B on vLLM) sees the world as structured objects rather than pixels, and after each colony writes a reflection that edits its skills, tools, and journal before starting the next run. It matters to agent developers as a worked example of persistent, self-improving game-playing agents with all memory stored on disk as editable files.
34 engagement1 sourceWed, Sep 16, 2026, 03:14 PM
AI Overview
- GitHub project uses Qwen 3 27–32B on vLLM via a RimBridge mod exposing the game engine over HTTP
- Agent sees the world as objects, trends, diffs; controls orders, designators, trade; stores skills and tools as editable markdown and hot-loaded Python
- Each colony ends with a reflection that edits learned files and commits them before starting the next run
Related
Products·building
Harness letting Claude Code play RimWorld autonomously and narrate on Twitch
A hobbyist evening project that connects Claude Code to RimWorld via the RimBridge mod and an MCP server, so the game appears as tools. A Python director script orchestrates headless Claude Code runs, each prompted with the current game state, giving the AI a personality as it plays and narrates live on Twitch. It solves the problem of wiring an LLM agent into a complex simulation game for entertainment and experimentation.
A hobbyist evening project that connects Claude Code to RimWorld via the RimBridge mod and an MCP server, so the game appears as tools. A Python director script orchestrates headless Claude Code runs, each prompted with the current game state, giving the AI a personality as it plays and narrates live on Twitch. It solves the problem of wiring an LLM agent into a complex simulation game for entertainment and experimentation.
11 engagement1 sourceredditTue, Oct 13, 2026, 10:16 PM
Products·launched
IronBee Gamer: bot that plays browser games by reading the page, with LLM-written rules and a local model for moves
A bot that opens a browser game with no API or hooks, observes what the page renders (canvas draw calls, engine scene dumps, or raw pixels), and plays it. An LLM acting as trainer writes the game-understanding code and rules; a small local model then executes each move in about 25 ms. It lets people automate or study arbitrary browser games without game cooperation.
A bot that opens a browser game with no API or hooks, observes what the page renders (canvas draw calls, engine scene dumps, or raw pixels), and plays it. An LLM acting as trainer writes the game-understanding code and rules; a small local model then executes each move in about 25 ms. It lets people automate or study arbitrary browser games without game cooperation.
32 engagement1 sourceredditThu, Oct 8, 2026, 07:49 PM
Social·Community
Redditor proposes AI agents that write and inject mods into running games like GTA 5 and Skyrim in real time
Reddit user proposes AI agent that writes and injects game mods in real time
Reddit user proposes AI agent that writes and injects game mods in real time
4 engagement1 sourceredditSun, Oct 11, 2026, 10:25 PM
Products·concept
AstraBridge — adapter letting LLM agents play Morrowind via OpenMW
AstraBridge is a specialized adapter/interface for a modified OpenMW build that lets LLM agents play Morrowind. Agents receive screenshots plus a semantic/action interface with helpers for movement, dialogue, inventory, journal, combat, UI interaction, and persistent memory. It gives researchers and hobbyists a way to run LLM agents through a complex RPG without granting hidden game-state access.
AstraBridge is a specialized adapter/interface for a modified OpenMW build that lets LLM agents play Morrowind. Agents receive screenshots plus a semantic/action interface with helpers for movement, dialogue, inventory, journal, combat, UI interaction, and persistent memory. It gives researchers and hobbyists a way to run LLM agents through a complex RPG without granting hidden game-state access.
1 engagement1 sourceredditFri, Oct 2, 2026, 05:33 AM
Products·building
400+ LLM agents living in a 2004-era MMO server, all local on Qwen3.8-4B
A multi-agent simulation running 400+ LLM-driven characters inside a vanilla-era MMO private server emulator. Each agent has its own backstory, goals, and memory, perceiving the world and issuing in-game actions. Built as a fun way to visualize and stress-test a multi-agent ecosystem, with a live view for inspecting any agent's reasoning in real time.
A multi-agent simulation running 400+ LLM-driven characters inside a vanilla-era MMO private server emulator. Each agent has its own backstory, goals, and memory, perceiving the world and issuing in-game actions. Built as a fun way to visualize and stress-test a multi-agent ecosystem, with a live view for inspecting any agent's reasoning in real time.
2 engagement1 sourceredditTue, Sep 29, 2026, 03:55 AM