Show HN: AI agents play SimCity through a REST API

Converts strategy documents (PDF, Markdown, web pages) into executable AI agents that play games like SimCity, Civilization, or Factorio.
Enables users to see AI follow human‑written tactics, benchmark LLMs on strategy execution, and share time‑lapse videos.

Key	Value
Target Audience	Game developers, AI researchers, hobbyists who want to test LLMs on strategy games
Core Feature	Document parsing → LLM prompt generation → game‑control API → real‑time play & recording
Tech Stack	Python, FastAPI, LangChain, OpenAI/Claude APIs, Micropolis/SimCityJS, WebSocket, Docker
Difficulty	Medium
Monetization	Revenue‑ready: $5/month per user for API access and premium recording features

FrustratedMonky said, “I’d like to see AI agents battle to build resources and to fight.” Doc2Agent turns that wish into a product.
Users can upload a Civilization strategy guide and instantly see an AI play the game, addressing the frustration that “the civ 5/6 bots can play pretty decent strategy and that’s without ‘AI.’”
The platform supports time‑lapse export, social sharing, and a leaderboard for best strategy execution, sparking discussion on how LLMs handle spatial reasoning.

A lightweight, offline tool that bundles the Micropolis engine with an LLM interface, runnable on Raspberry Pi or a local laptop.
Allows users to run evolutionary or scripted AI agents without cloud costs, satisfying the desire for “offline” or “local” play.

Key	Value
Target Audience	Hobbyists, educators, researchers with limited GPU resources
Core Feature	Docker‑free Micropolis binary + Python API for LLM control + optional evolutionary loop
Tech Stack	Micropolis (C++), Python, Flask, Docker (optional), OpenAI API, Prolog for constraints
Difficulty	Medium
Monetization	Hobby

FrustratedMonky and others expressed a need for “offline” play: “What would it take to run this locally / offline?”
The tool includes a simple CLI to launch a city, feed LLM prompts, and record snapshots, directly addressing the “time‑lapse view” request.
By providing a local, low‑resource setup, it removes the barrier of “hemorrhaging finite resources to play these games badly.”

A public benchmarking service for LLMs on spatial tasks, featuring standardized APIs for games like SimCity, Factorio, and Civilization.
Offers a leaderboard, dataset, and integration with OpenRouter, filling the gap of “hard‑to‑game benchmarks” for spatial reasoning.

Key	Value
Target Audience	AI researchers, LLM developers, game AI community
Core Feature	Unified game‑control API, benchmark suites, automated scoring, leaderboard
Tech Stack	Node.js, Express, Docker, OpenRouter, PostgreSQL, Grafana
Difficulty	High
Monetization	Revenue‑ready: $10/month for commercial API access, free tier for research

Users like aed and rglullis want to “see AI play factorio” and “benchmark LLMs on spatial tasks.” SpatialBench provides a structured way to do that.
The service supports “hidden selection criteria” and evolutionary agents, enabling community experiments like “breed evolutionary agents to compete with one another.”
By standardizing the interface, it encourages discussion on how LLMs handle spatial reasoning and fosters reproducible research.

🚀 Project Ideas