Pokémon Red and Blue

Playing Pokemon Red with Reinforcement Learning

Canonical deep-RL Pokemon Red project (PPO on PyBoy, exploration/badge/level rewards, parallel envs); ~ MIT

Disassembly of Pokémon Red/Blue

Active; status stable

Gen1Recomp - A native Lua / LÖVE2D recreation of Gen 1, 2, and 3 Poke

version v0.3.2 (latest); status beta

A minimal, complete, Pokémon battle simulation engine optimized for performance

Minimal, complete Pokemon battle engine (Zig) aiming at frame-accurate reproduction of the original cartridge battle code (Gen 1 first) plus Showdown semantics; >1000x faster than Showdown; pre-v0.1 heavy development

A reimplementation of the Pokemon Red/Blue engine in Swift and SwiftUI

A reimplementation of the Pokemon Red/Blue engine in Swift and SwiftUI; no GitHub releases; repo created 2026-03-09

Official repository of the paper: Continual Harness: Online Adaptation for Self-Improving Foundation Agents and PokeAgent Speedrun Track 2

game red option in same framework

A free HTML5 remake of the original Pokemon, expanded for modern browsing.

HTML5/TypeScript remake of the original Pokemon (inactive since 2021)

PufferLib-based high-throughput Pokemon Red RL (David Rubinstein et al.); the "PokeRL" line of work that pushed pure-RL agents deep into the game

Minimal starter version of the Anthropic "Claude Plays Pokemon" harness (David Hershey); PyBoy + tool use (button presses, RAM-derived state, navigator)

AI-powered Pokemon gameplay agent with headless emulation, REST API, and live dashboard. Works with any LLM.

2026-06

Pokémon Red/Blue overhauled with full GBC support. Made by FroggestSpirit, Drenn, and dannye. Also check out the crysaudio branch.

OSGC status: playable; development: sporadic; game already in catalog; this project is a different project type for it

2025-06

Pokémon Red, rebuilt from raw pixels by Claude Opus 5.5. Every sprite, tile, battle effect and song is code: zero image files. Plays in the browser.

(2026-09); fidelity to original unverified

Gymnasium environment for Pokemon Red

Inactive since 2024-06

Public version of the "Gemini Plays Pokemon" harness (Joel Z); mGBA-based, multiple Pokemon games incl. Crystal; active Oct 2026

Anniversary Red dissassembly

OSGC status: playable; development: complete; game already in catalog; this project is a different project type for it

LLM Pokemon scaffold on PyBoy (community replication of Claude/Gemini harnesses)

Disassembly of JPN Pokémon Red/Green

disassembly of Japanese edition of Pokémon Red

Partial disassembly of the JP release (pret/pokered now covers several JP builds)

An upgrade to Pokémon Red. Fixes several bugs and adds a lot of new features. Now a standalone repository, so you can actually search it if you want (among other things).

OSGC status: playable; development: complete; An upgrade to Pokémon Red. Fixes several bugs and adds a lot of new features; game already in catalog; this project is a different project type for it

Disassembly of JPN Pokémon Blue

Disassembly of Pokémon Red/Blue with some fixies and improvements (and a little surprise at Vermilion Dock if you talk to Giovanni after the fight in his Gym ;) ). A special thank to Jojobear and Pre…

Pokémon Red & Blue recompiled for the Game Boy using a fork of gb-recompiled · Part of the R.A.I.D. community

recompiled using a fork of gb-recompiled

Playing Pokemon Red using TypeSafe Jev

Pokemon Red played by "TypeSafe Jev" agent on PyBoy (several forks: 2389-research, BrickerP)

Generation I Pokémon (Red/Blue/Yellow) statically recompiled to native C — no emulator at runtime. Plus a headless build and an RL-bot integration that plays and differentially tests the recompiled e…

See static-recomp entry: headless rom_headless.dll with PyBoy-compatible shim, differential tester and PPO training

PSP port of gen1recomp

Pokemon Red with Reinforcement Learning

Modular deep-RL system for early-game Pokemon Red tasks (exit house, reach grass, win first rival battle); paper April 2026

attempt to replicate Claude Plays Pokemon with other LLMs

Attempt to replicate Claude Plays Pokemon with other LLMs

Port of gen1recomp for Nintendo 3DS

Autonomous Pokémon Red agent using local Qwen3-VL and PyBoy. Zero API calls, zero cost, pure local VLM reasoning and computer vision.

Autonomous agent with local Qwen3-VL on PyBoy (no API calls)

Pokemon Red recompilation using gb-recompiled

Pokemon Red recompilation using gb-recompiled; no GitHub releases; repo created 2026-05-10

Gen1Recomp - Lua Native 3ds Port

Local Pokemon Red LLM-agent harness with live gameplay UI, structured traces, save states, replay, and turn-by-turn agent observability.

Local LLM-agent harness with live UI, structured traces, save states, replay and per-turn observability

Pokemon GB/GBC compilation using gb-recompiled

A Lean 4 formalization of the Game Boy's Sharp SM83 CPU, with Pokémon red examples

Lean 4 formalization of the SM83 CPU with Pokemon Red examples

Work in progress: agents playing the real Pokémon Blue ROM on PyBoy, with every badge, catch and evolution verified in cartridge RAM.

WIP multi-agent LLM + PPO agents on PyBoy with badges/catches/evolutions verified in cartridge RAM

Trains a deep Q-network to play Pokémon Blue from the screen, through PyBoy and PyTorch.

DQN from pixels via PyBoy + PyTorch (Gymnasium)

Pokemon Red and Blue in the browser with every picture regenerated (clean room build on the pret/pokered disassembly)

andrewnakas/decompgames browser build: the pret/pokered disassembly on EmulatorJS (gambatte); verification: {"status": "smoke-tested", "date": null, "browsers":, "scope": "Headless Edge test of the live build: boots, picture, sound, input.…

A live disassembly of Pokémon Red that read and rewrite the original game's memory in real time to power a custom web UI for online PvP battles

A modified disassembly of Pokémon Red/Blue

repo archived

A fixed agent harness for Pokémon Red where the model is the only variable: same tools, memory, prompt and observation format across 5 models from 4 vendors, turn-matched and eval-integrity gated. Li…

Fixed agent harness for Pokemon Red where the model is the only variable (same tools/memory/prompt across 5 models, turn-matched)

LLMs Play Pokemon

LLMs play Pokemon

LLMs play pokemon red

Early LLMs-play-Pokemon-Red experiment by a pokemonred_puffer contributor

Pokemon Reinforcement Learning with RLlib

Pokemon Red RL with RLlib

Dataset of the original Claude Plays Pokemon Twitch Chat pulled from ~4 days via the API

Dataset of ~4 days of original Claude Plays Pokemon Twitch chat

Reinforcement learning toolkit for training agents to play Pokémon Red. Built on PyBoy, Stable-Baselines3, and Gymnasium. Includes pixel/symbolic/hybrid observation treatments and rliable-based stati…

RL toolkit on PyBoy + SB3 + Gymnasium with pixel/symbolic/hybrid observation treatments

Completed reproducible Pokémon Red study: 8.24M self-generated actions reached Route 1, but frozen evaluation found zero durable skills.

Completed reproducible study: 8.24M self-generated actions reached Route 1, frozen evaluation found no durable skills (useful negative result)

Experimenting with getting Open Source LLMs to play Pokemon Red.

Open-source LLMs playing Pokemon Red

A verified memory atlas for Pokemon Red and Blue: 2,898 evidenced claims covering the whole address space, every one checked against a running cartridge rather than transcribed.

Verified memory atlas for Red/Blue: 2,898 evidenced claims over the whole address space, each checked against a running cartridge

[ICLR 2026] LLM/VLM gaming agents and model evaluation through games.

Part of lmgame-Bench (2025)

Benchmark environment for evaluating vision-language models (VLMs) on popular video games!

Included in VideoGameBench dev split (2025)

One of Orak's 12 games (2025)

Deep Reinforcement Learning Framework done with PyTorch

Code for arXiv 2502.19920 (with Rubinstein and Whidden); baseline reaches Cerulean City

A native Godot 4 reimplementation of the Gen 1 and 2 Game Boy Color games. Not an emulator, bring your own ROM.

version v0.1.36 (latest); completeness: playable ('Gold, Silver, Crystal, Red, Blue and Yellow are playable from the bedroom to the Hall of Fame and their own credits.')

Grokemon: an autonomous Grok agent that plays Pokemon Red on mGBA from emulator state only.

Autonomous Grok agent playing Pokemon Red on mGBA (mgba-http) from emulator state only; archived fork; repo archived

Platform for AI agents and models to compete playing Pokémon Red via an HTTP API. PyBoy emulator + Next.js; agents register, start/load games, send actions, get state and screenText. Watch live; RL a…

Platform where AI agents/models register and compete at Pokemon Red via HTTP API (PyBoy + Next.js leaderboard)

An LLM that plays Pokémon Red by reasoning over structured game state, using an MCP server on the PyBoy emulator with memory readers, a world model built from the pokered disassembly, and pathfinding.

LLM agent over structured game state via an MCP server on PyBoy with memory readers and a world model built from the pokered disassembly