Local AI

On your computer, or not at all.

QuillArk is a writing application first. The AI in it is a tool for the dull parts — summarising, extracting, organising — and it runs where your files are.

Two engines

Built in

Works the moment you install QuillArk, with nothing to download and no model to choose. It is extractive: it works with the words already in your manuscript rather than inventing new ones.

Summarise a chapter or a scene
Pull out likely characters, places and objects
Organise a brain dump into things you can act on
Suggest titles drawn from the text
Flag inconsistencies between a record and the chapters that mention it

A local model

Point QuillArk at Ollama or llama.cpp running on your machine and you get a language model as well: rephrasing, continuation, and questions answered in prose.

You choose and install the model; QuillArk does not bundle one
Model size and speed depend on your hardware, not on a subscription
If the engine is not running, QuillArk falls back to the built-in one rather than failing

How “local” is enforced

Not by policy, by code. The endpoint you configure is parsed and checked before any request is made. It must be a loopback address — localhost, 127.0.0.1 or ::1. Anything else is refused, and the request never happens.

Prompts travel that same path. There is no separate route by which a prompt could go somewhere your manuscript could not.

What it will not do

It will not write your book. The built-in engine does not generate prose at all, and QuillArk does not present AI as the reason to use it.
It will not send anything anywhere. Including to us. There is no “us” with a server.
It will not require setup. Every feature of QuillArk works with the AI switched off entirely.

Downloading a model, if you want one, needs the internet — that is between you and whoever distributes the model. Using it afterwards does not.

Try it with nothing installed.

Download QuillArk free How privacy works