A Gem is a system prompt with a job and a set of rules. These starters load that prompt from a file in gems/ and send one user message to the Gemini API.
flowchart LR
A[Pick a Gem] --> B[Copy instructions]
B --> C[Set GEMINI_API_KEY]
C --> D[Run the starter]
D --> E[Review the draft]
The same flow is on the gallery page.
Each Gem file has one fenced text block. That block is the whole instruction. The starter uses the first block and ignores the rest of the page. Do not add a second text fence.
Create a key in Google AI Studio. Put it in the environment. Do not write it into a source file.
export GEMINI_API_KEY="your-key"
Optional: GEMINI_MODEL overrides the default model, gemini-2.5-flash.
pip install google-genai
python examples/gem_app.py gems/02-meeting-recap-desk.md "Recap these notes: We will ship the export fix on Thursday. Priya owns the patch."
The script is gem_app.py. Run it from the repository root so the Gem path resolves.
Node 18 or newer.
npm install @google/genai
node examples/gem_app.mjs gems/02-meeting-recap-desk.md "Recap these notes: We will ship the export fix on Thursday. Priya owns the patch."
The script is gem_app.mjs.
You get a draft in the terminal. The model is not connected to your mail, calendar, bank, or repository. If you want a different desk, change the file path. The meeting recap above is a safe sample because the notes are included in the prompt.