Touch grass at your own pace

Step outside, your way.

Most "go outside" apps assume you can hike. Porchlight is for everyone else: older adults, people with limited mobility or low vision, anyone with five minutes and a doorstep. An open AI model runs on your own computer, so your details stay with you.

Try the safety guard
This is a demo site. The real Porchlight runs on your own computer with a local open-weight model (Gemma through Ollama). It is not running live here. The guard below uses the same rules as the app, but the suggestions are hand-written samples, not a live AI.

How it works

The model suggests. The code decides.

1. You set the limits

How far you can go, stairs, heat, time and what you have at home. Asked once, saved only on your computer.

2. The model suggests

A local open-weight model writes one tiny mission for today: a window, balcony, doorstep, yard or short walk.

3. The code decides

Every idea is checked against your limits. If it doesn't fit, it is rejected and the model tries again, or a safe backup mission is used.

A mission card

One of the built-in backup missions, shown the way the app displays it.

Example mission

Doorstep deep breath

doorstepabout 5 minfocus: smellshade
  1. 1Open the door and sit or stand on the threshold.
  2. 2Breathe slowly and notice any smell in the air.
  3. 3Stay until you can describe the air in three words.

Try the guard

Pick a person and a suggestion, then see whether the guard accepts it. The same suggestion can be fine for one person and rejected for another.

Choose a person and a suggestion, then press the button.

The checks are the same rules the app applies: allowed places, minutes, distance, stairs, direct sun on a hot day, and a blocklist of strenuous words. In the real app a rejected answer is sent back to the model for another try.

Run the real thing

You need Python 3 and Ollama. No account, no API key, no extra packages.

ollama pull gemma3:4b
git clone https://github.com/Karishma-Das/porchlight
cd porchlight
python porchlight.py

Source code: github.com/Karishma-Das/porchlight

Why open and local

Mobility and health details never leave the computer. It works offline once the model is downloaded, the model can be swapped with one setting, and it costs nothing to run.