Go Outside
A web application designed to encourage users to put down their phones and go outdoors. It uses a browser-side local AI model to generate small outdoor micro-tasks. All data is stored locally on the user's device, requiring no backend service.
Timeline (last 90 days)
Go Outside is a web application developed by independent developer jackydu1981 that generates outdoor tasks using a Gemma model running on the browser's local GPU.
Go Outside's default model is gemma-2-2b-it-q4f16_1-MLC-1k, which is the Gemma 2 2B instruction-tuned version with 4-bit quantization and a 1k context window.
If the GPU does not support 16-bit shaders, Go Outside will automatically switch to the q4f32 build of the same model; on mobile there is a lighter option using approximately 600MB Gemma 3 1B (gemma3-1b-it-q4f16_1-MLC).
Go Outside is built with pure HTML, CSS, and ES modules, with no build step, no backend, and is licensed under MIT.
Go Outside's weather and sunset time data come from Open-Meteo, which is free and requires no API key, and is only requested when the user explicitly grants location permission.
Go Outside includes a parser to handle small-model outputs, and when the model output is invalid or loading fails, it conditionally selects one of 22 handwritten tasks as a fallback.
Go Outside's model download is approximately 1.5GB; the first task will first seek user consent and offer handwritten tasks as an alternative.
Go Outside uses the Service Worker's Cache Storage API to persist weights and the application itself locally, enabling full functionality including model inference even offline after the first successful load.
The only network request Go Outside makes during a task is an optional weather request with coordinates rounded to approximately 10km, and user data is not sent to any model provider.
All Facts (9)
Go Outside is a web application developed by independent developer jackydu1981 that generates outdoor tasks using a Gemma model running on the browser's local GPU.
50%UnverifiedGo Outside's default model is gemma-2-2b-it-q4f16_1-MLC-1k, which is the Gemma 2 2B instruction-tuned version with 4-bit quantization and a 1k context window.
50%UnverifiedIf the GPU does not support 16-bit shaders, Go Outside will automatically switch to the q4f32 build of the same model; on mobile there is a lighter option using approximately 600MB Gemma 3 1B (gemma3-1b-it-q4f16_1-MLC).
50%UnverifiedGo Outside is built with pure HTML, CSS, and ES modules, with no build step, no backend, and is licensed under MIT.
50%UnverifiedGo Outside's weather and sunset time data come from Open-Meteo, which is free and requires no API key, and is only requested when the user explicitly grants location permission.
50%UnverifiedGo Outside includes a parser to handle small-model outputs, and when the model output is invalid or loading fails, it conditionally selects one of 22 handwritten tasks as a fallback.
50%UnverifiedGo Outside's model download is approximately 1.5GB; the first task will first seek user consent and offer handwritten tasks as an alternative.
50%UnverifiedGo Outside uses the Service Worker's Cache Storage API to persist weights and the application itself locally, enabling full functionality including model inference even offline after the first successful load.
50%UnverifiedThe only network request Go Outside makes during a task is an optional weather request with coordinates rounded to approximately 10km, and user data is not sent to any model provider.
50%