Run a local model (fully private)
Run chekd on a model that lives entirely on your own machine. No provider per-use bill and no model-provider egress — the model uses your device resources, and nothing about your app leaves your computer. Here’s how to set it up.
How this works
Ollama is an open-source tool that runs language models directly on your computer. Once it’s installed and a model is pulled, chekd can talk to it locally — no provider account, no API key, and no provider per-use bill. Your own hardware, electricity, and model-license terms still apply.
Because the model runs on your machine, this fits chekd’s local-first promise: nothing about the app you’re cheking is uploaded anywhere. It stays between chekd and the model, both on your computer.
Set it up
- Install Ollama Download and install Ollama for your operating system from its official site, ollama.com. Follow its installer for your platform.
-
Make sure it’s running
On most systems Ollama starts automatically after install. If it isn’t already running, start it from a terminal:
$
ollama serve -
Pull a model
Download a model for Ollama to run. Pick any model Ollama supports — replace
<model>with the one you want:$ollama pull <model> - Come back to chekd and pick “a local model” With Ollama running and a model pulled, chekd can find it. Open the model picker, choose the local option, and you’re set.
What chekd does next
chekd checks for a running Ollama with at least one model pulled. If it finds one, the local option shows a ready tick and lists the models it can use. Choose it in one click and optional model-backed actions use that local model — your device resources, no provider per-use bill, and no metered API call.
The model runs on your own machine, so nothing about your app leaves your computer to a model provider. You can instead choose Claude via Claude Code or ChatGPT via Codex; those model-backed actions send the task context they need to the selected provider, whose plan, limits, charges, and terms apply. Antigravity is a separate, explicitly enabled text-only experiment. Paid API keys also work through explicit authorization, a capped connection test, and a separate spend ceiling for every job.