# gamestar **Repository Path**: mindyleelyyAI/gamestar ## Basic Information - **Project Name**: gamestar - **Description**: No description available - **Primary Language**: Unknown - **License**: Apache-2.0 - **Default Branch**: main - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 0 - **Created**: 2026-10-18 - **Last Updated**: 2026-10-23 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README > **GameStar** is a fork of [OhMyGame](https://github.com/WhiteTowerAI/ohmygame) by WhiteTowerAI, > licensed under [Apache-2.0](LICENSE) — © the original authors. This fork removes cloud telemetry, > disables auto-update, localizes publishing/login, and bundles extra skill plugins (Godot, game design > library, web engines, insurance stories).

OhMyGame — the open-source studio for making games with AI

**Your own one-person game studio.** Everything you need to make games with AI, in one app. Bring your own key.

English · 简体中文

https://github.com/user-attachments/assets/4d5805fe-92a2-40b4-a232-c7452f9a796a

> [!NOTE] > 🚧 OhMyGame is in beta. Expect rough edges. Join us and let's build it together! ## What it is Making a game takes design docs, 2D art, 3D models, video, code, and somewhere to put it when it's done. Today those live in five or six different tools. OhMyGame puts them in one app: describe your game, and the agent designs it, makes the assets, builds it, plays it to check it works, and publishes it to a link you can share. Two core design ideas shape it: 1. **Everything lives on a canvas.** Design docs, concept art, 3D models, and video sit side by side on one board. One character sketch can grow into more art, a video, or a rigged 3D model, and everything is ready to use in your game. 2. **The agent does the work.** It drives the canvas the way you would: drafting the design doc, turning a character sketch into a rigged 3D model, writing the code, and playtesting the result. It's open source and works with 40+ model providers through Pi, from Anthropic and OpenAI to DeepSeek and Qwen. Your keys stay on your machine, there are no per-run credits, and what you make is yours. ## Features ### 🤖 Agent: drives the canvas, builds and plays the game The agent works the canvas the way you would: it drafts the design doc, lays out boards, and generates assets on them. Then it reads the design doc, writes the code, and runs the game in a live preview. Then it opens the game in a window you can watch and plays it, including real-time canvas and WebGL games. When it finds a problem, it tries a fix and plays again. Along the way it can generate images, video, and 3D models, search the web, and use MCP servers or the Codex and Claude Code plugins you already have. **▶ The agent playtesting a new item feature on desktop and mobile, then reporting what it found (sped up 4×).** https://github.com/user-attachments/assets/3cb982e8-a4a8-4715-bf3c-df4e41e7855b ### 🎨 Canvas: design docs, art, video, and 3D on one board Every game project has a design space: a main design doc plus boards where the doc, images, video, and 3D models sit side by side. Nodes can use each other as references, so one character design can lead to more images, a video, or a rigged and animated 3D model. Results go into a shared Asset Library that your games can use. https://github.com/user-attachments/assets/5a35964d-87b3-4153-8b1b-fc6deced5bb3 ### 🔗 Publish: one link, playable anywhere Sign in and publish: OhMyGame builds your game and gives it a link anyone can play in the browser. Republishing updates the same link. Published games are listed in the in-app Community, where you can play what others have made. ## What you can make Three games our team made with OhMyGame:


Interactive film	Island builder	Deckbuilding roguelike

### 🎮 Web games **The best place to start.** 2D and 3D games for the browser, built with Three.js, React Three Fiber, or Phaser. The agent follows a complete web game workflow: a design doc, key-screen mockups, production art, a playable core loop, and playtesting in the live preview. Games fit any window size, support touch, and publish to a shareable link. The project is a normal web project you can open in any editor. ### 🎬 Interactive stories Interactive films, visual novels, and story games. Each screen is a free web page with its own images, video, and code: a main menu, a case archive, a dialogue, a puzzle. You connect them on a graph, and the runtime handles state, saves, and navigation. The architecture and authoring model are documented in [Playable Nodes](docs/playable-nodes/README.md). ### 🎲 General games: any engine or stack **General Game** doesn't assume an engine or stack. Start from an idea, bring a project in any engine, or make tools and content for your game. The Game Studio plugin coordinates design, production art, implementation, and verification across engines. When the output is a web page, preview, playtesting, and publishing work as they do for Web games; native games use their engine's own tools. See [General Game details](docs/runtime.md#agent-game-use). **Making a browser game? Choose Web Game**: it is tuned for the browser. ## How it's built The idea that holds this together: one agent drives every part of the studio. **1 · A full Pi agent with a GUI.** OhMyGame runs [Pi](https://www.npmjs.com/package/@earendil-works/pi-coding-agent) with its full toolset and persistent sessions, in a GUI built for watching results rather than reading diffs. Close the app, come back, and the project, conversation, and context are still there. You describe, review, and steer; editing by hand is always possible, but it is not the main path. **2 · Every workspace is files.** A web game is a normal web project. A design doc is Markdown, and boards and the Asset Canvas share one JSON node model. An interactive story is a node graph on disk. Each comes with an `AGENTS.md` and JSON schemas, so the agent edits them the same way it edits code, and the visual editors read and write the same files. **3 · Games are tool calls.** Playing the game (`game_use`), generating media, and checking story nodes are ordinary Pi tools, next to its coding tools. That is why the agent can chain them: draft a design, make the art, write the code, and play the result in one conversation. ## Any model, your keys OhMyGame does not lock you into one model or charge per run. - **Coding models from 40+ providers** through Pi, including Anthropic, OpenAI, Google, DeepSeek, Qwen, Kimi, GLM, MiniMax, xAI, Mistral, Groq, OpenRouter, Amazon Bedrock, and Vercel AI Gateway. - **Sign in with a subscription you already have**, such as ChatGPT Plus/Pro or GitHub Copilot, or paste an API key. - **Media models** for images, video, and 3D through OpenAI, OpenRouter, Volcengine Ark / BytePlus ModelArk, and Meshy. - **Custom providers and model aliases** with selectable language, image, video and 3D uses, protocol templates and generation defaults. See [custom model configuration](docs/custom-models.md). Keys stay in the local daemon. They never reach the agent's workspace, your game, or the renderer. ## Quick start ### Download See [the changelog](CHANGELOG.md) for changes in each release, or open **Settings → About → Changelog** inside the app. Grab the installer from the buttons at the top or from [Releases](https://github.com/WhiteTowerAI/ohmygame/releases). Builds are available for macOS (Apple silicon) and Windows (x64). The Windows build is not code-signed yet, so Windows may show a SmartScreen warning during install. ### First run ![First-run welcome and provider connection screens](docs/assets/first-run-onboarding.png) OhMyGame ships without a model. On first launch, a small game world forms and transforms into the OhMyGame logo. The wordmark, **Your own game studio.**, and **Next** appear in sequence over about five seconds. Click **Next** to connect a model before your first prompt: 1. **Connect a language model.** **Choose your provider** lists available providers with their sign-in or API key options. Use the **+** button beside search to add a custom or local compatible endpoint in the same window. **Start creating** becomes available when a language model is ready. You can also choose **Skip and start** to enter without setup and configure a model later in **Settings → Providers & Models**. Saved connections show **Connected**. Setup resumes where you left off; existing installations go straight to the studio. 2. **Add media providers if you want them.** The agent can design and build a game with only a language model. Connect image, video, and 3D providers and choose default models in **Settings → Providers & Models** whenever you need them. Generating art, video, or 3D needs its own provider: | To make | Connect | | ----------------------------- | ------------------------------------------------------------------- | | Design docs, code, playtests | Any language model provider | | Images and sprites | OpenAI (API key), OpenRouter, or Volcengine Ark / BytePlus ModelArk | | Video | OpenRouter, or Volcengine Ark / BytePlus ModelArk | | 3D models, rigging, animation | Meshy, Tripo, or Hyper3D (capabilities vary by provider) | Signing in with ChatGPT does not cover image generation. In the OpenAI provider's **Image models** section, configure a separate API key and Base URL, fetch or manually add image models, enable them and save. This works alongside browser sign-in and supports compatible GPT Image relays. API-key chat connections can also share their key and endpoint with image generation. 3. **Pick a model and describe your game.** Choose the model in the prompt box on the home screen, then describe what you want to make. Web search works without setup. Sign in under **Account** only when you want to publish a game; making games locally needs no account. Reopen the welcome from **Settings → About → View getting started**. ### Run from source Requires Bun 1.4.2 or newer for the toolchain and Node.js 22.19+ for the daemon. ```bash git clone --branch main https://github.com/WhiteTowerAI/ohmygame.git cd ohmygame bun install --frozen-lockfile bun run dev # web renderer + local daemon # or bun run dev:desktop # Electron app ``` This uses the release branch. For development contributions, start from `dev` using [CONTRIBUTING.md](CONTRIBUTING.md#fork-and-create-a-branch). Local development can start without cloud configuration; see [CONTRIBUTING.md](CONTRIBUTING.md#configuration) for optional integrations. Design and Asset Canvas share boards, Markdown documents, and local media under `canvas/`. See [Canvas workspace](docs/canvas-agent.md) for the file contract and offline migration instructions. ## Privacy Your projects and keys stay on your machine. Prompts and generation requests go directly to the model providers you choose. Publishing requires an OhMyGame account and uploads only your game's static build. Release builds check for updates and send anonymous usage events (PostHog, no session recording) and crash reports (Sentry, no personal data). Builds from source send no telemetry. ## Roadmap We're working on plain-language playtest reports, remixing games from the Community, and Intel Mac / Linux builds. See [Issues](https://github.com/WhiteTowerAI/ohmygame/issues) for what's next, and vote with 👍 on what matters to you. Want to help? Look for [`help wanted`](https://github.com/WhiteTowerAI/ohmygame/labels/help%20wanted) issues. ## Contributing Issues and PRs are welcome. Start contribution branches from `dev` and target ordinary PRs at `dev`; `main` receives releases. See [CONTRIBUTING.md](CONTRIBUTING.md) for setup, checks, and review, [中文快速指引](docs/contributing.zh-CN.md) for a short Chinese introduction, and [the documentation index](docs/README.md) for module guides. Report security issues using [SECURITY.md](SECURITY.md). ## Star History

## License [Apache-2.0](LICENSE). The OhMyGame name and logo are not covered by the license.