Pace
How to get your AI key
Pacedex's race engineer already talks for free through the cloud (up to 20 questions a day). With your own AI key it goes unlimited — and the key stays on your PC, never touching Pacedex servers. Recommended path: Google's Gemini — free, no credit card, 2 minutes. Already running models at home? Point Pacedex at your own server — no key at all.
Open Google AI Studio
Go to aistudio.google.com/apikey and sign in with your Google account (the same one you use for Gmail/YouTube). It's free and asks for no card.
Create and copy the key
Click Create API key and copy the generated key. Treat it like a password: it's yours.
Paste it into Pacedex and hear the radio 🎙️
In Pacedex, open Settings → the Engineer section. Pick Gemini (Google) — free, paste the key, save and hit Test — the engineer answers instantly, out loud.
Prefer Claude or ChatGPT?
They work too — and they're the sharpest models on the radio. The difference: their APIs are paid, so you need to add credit first (US$ 5 minimum). Don't worry: one radio question costs a fraction of a cent — US$ 5 lasts for months.
Claude (Anthropic)
Create an account at platform.claude.com, go to Billing and buy credits (min. US$ 5). Then, under API Keys, create a key and paste it into Pacedex.
ChatGPT (OpenAI)
Create an account at platform.openai.com, add credit under Billing (min. US$ 5). Then, under API keys, create a key and paste it into Pacedex.
What does it cost to use?
Next to nothing. A radio question is a small chunk of text — on the fast (⚡) models that's a fraction of a cent. You won't feel it even across a full league season.
Run your own models? Point Pacedex at your server
If you already run LLMs at home — Ollama, LM Studio, LiteLLM, vLLM, llama.cpp — the engineer can talk to them directly. Any OpenAI-compatible endpoint works: no key, no quota, no cloud, and nothing leaves your PC.
Start your server and note the address
Ollama serves on http://localhost:11434/v1, LM Studio on http://localhost:1234/v1. The address alone is enough — Pacedex adds the /v1 for you. Server on another machine at home? Use that machine's IP.
Point Pacedex at it
In Settings → Engineer, pick OpenAI-compatible (your server), paste the address and type the model id you loaded (there's no list to pick from — the catalogue is whatever your server holds). The key is optional: fill it in only if your server asks for one. Then hit Test — the engineer confirms over the headset.
What to expect
On the radio, latency rules: a small model answers instantly, a big one leaves the engineer talking late. Your server also powers the AI race analysis — that one reads your database through tool calls, so it wants a model that supports them; without tool calling it still writes a report, just a shallower one.