# MineDojo/NitroGen

A Foundation Model for Generalist Gaming Agents

Repository: https://github.com/MineDojo/NitroGen
Canonical: https://ross.abutalabs.com/products/nitrogen
Language: Python
License: NOASSERTION
License Family: other
Last push: 2026-01-25T21:05:37+00:00

## Health v2 (maintenance only)
Score: 45/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 64, release rhythm 35, longevity 18
- inputs: {"age_days": 259, "days_push": 220, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2168, forks 239 (observed 2026-08-28T04:06:21.546556+00:00)

## What it is
NitroGen is an open foundation model for generalist gaming agents that takes pixel input from games and predicts gamepad actions, trained via behavior cloning on internet gameplay videos. The repository provides inference serving and agent-running scripts for deploying the 500M parameter model on Windows games.

## Use cases
- run an AI agent that plays Windows games from pixel input
- serve a gameplay foundation model for inference
- post-train a generalist gaming agent on a new game
- research training agents from internet video data
- control a game via predicted gamepad actions

## When to choose
- you want to experiment with a generalist game-playing foundation model
- you need a fast-reacting system-1 sensory agent for Windows games
- you are researching behavior cloning from internet gameplay videos

## When to avoid
- you need long-horizon planning or end-to-end game completion
- your games run on Linux or macOS only
- you need a production-ready game bot rather than a research prototype
- you cannot provide your own copies of the games

## Facets
- artifact type: library
- maturity: experimental
- function: machine-learning, deep-learning, agent-framework, llm-inference
- domain: artificial-intelligence, gaming-tools, machine-learning
- platform: windows, python
- tags: foundation-model, game-agents, behavior-cloning, video-game-ai, research-project, ai-agents, linux, gpu

## Member repositories
- MineDojo/NitroGen (main) score 45

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:21.546556+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:49:29.292491+00:00, confidence not recorded.
  - readme: https://github.com/MineDojo/NitroGen (fetched 2026-08-28T04:06:21.546556+00:00, sha 3623516ef2ba)
- Data as of 2026-08-30T08:39:29.467469+00:00.