FoxQuill
Download

AI setup

AI in FoxQuill is off until you turn on Enable AI features in Settings. With it off, FoxQuill has no AI features at all, and nothing here applies. The AI group in Settings shows once Show more tools is on, under Settings > This app > Tools.

The quick way#

FoxQuill’s AI features need a language model: a program that reads text and writes more of it. FoxQuill does not come with one. You can run one on your own computer for free, and then nothing you write leaves it.

The easiest program for this is LM Studio.

  1. Install LM Studio from lmstudio.ai. It is free.
  2. Open it and download a model. LM Studio suggests one when it first opens, and marks the ones that fit your computer. A model of 4 to 8 billion parameters (the size is in its name, such as “8B”) suits most computers.
  3. Turn on LM Studio’s local server. Open the Developer tab and switch the server on. Leave LM Studio open while you write.
  4. In FoxQuill, open Settings, then AI, and turn on Enable AI features. FoxQuill looks for LM Studio by itself and shows what it found, such as “Found LM Studio with Qwen3 8B.”
  5. Press Use. That fills in everything for you.
  6. Press Test connection if you want to be sure it works.

If FoxQuill found nothing, check that LM Studio’s server is on and press Find local AI.

With Ollama instead#

Ollama does the same job. Install it from ollama.com, open it, and download a model. You can pick one in the Ollama app, or type this in a terminal:

ollama pull qwen3:8b

Then press Find local AI in FoxQuill, and Use.

When it finds more than one#

If you have LM Studio and Ollama, or several models, FoxQuill lists them all. Pick one and press Use. A model that is already loaded is offered first, because it answers without a wait.

What Find local AI does#

It asks two fixed addresses on your computer whether LM Studio or Ollama is running there, and which models they have. It never looks anywhere else, and it never sends your writing.

It only looks when you ask: when you press Find local AI, or once when the AI settings open with AI on and no address saved yet, which includes switching AI on for the first time. It never runs in the background, and with AI off it never runs at all.

Use fills in three things: the address, the model, and the context window. You can still change any of them by hand below.

Models that cannot write are left out, such as the embedding models some apps download alongside. So are Ollama’s cloud models, which run on Ollama’s servers rather than on your computer.

Give the AI more room#

The context window is how much of the book the model can read at once. LM Studio and Ollama both start a model with a short one, usually 4,096 tokens, which is about 2,700 words. FoxQuill always fits what it sends into the window, so the AI still works, but it sees less of your book.

When you press Use, FoxQuill sets the context window to what the server really uses, and never more. Setting it higher than the server allows does not give the model more room: the server cuts the start of what FoxQuill sends, without saying so.

To give the model more room:

  • LM Studio: when you load a model, raise Context Length in its load settings, to 16,384 for example, and load it. Then press Find local AI again.
  • Ollama: raise Context length in the Ollama app’s settings, or set OLLAMA_CONTEXT_LENGTH where Ollama starts. FoxQuill can only see the new size while the model is loaded, so type the same number into Context window in FoxQuill yourself.

A bigger window needs more memory, and the model takes longer to start each answer. FoxQuill never sets more than 32,768 on its own.

Setting it up by hand#

Any server that speaks the OpenAI chat completions API works. FoxQuill talks to the server directly. No account is needed.

Fill in three things in Settings under AI:

  1. Server address, under Use another server, usually ending in /v1.
  2. Model, the name the server uses. List models lists what it offers.
  3. API key, only if the server wants one. Local servers usually do not. A saved key belongs to its server: change the server address to another computer or port and the key is removed, so type it again if the new server needs one.

Test connection tells you whether it worked, and how long it took. Context window, the longest answer, and the temperature are under Advanced.

Ollama#

Server address: http://127.0.0.1:11434/v1. No key.

LM Studio#

Server address: http://127.0.0.1:1234/v1. No key.

llama.cpp and llama-swap#

llama-server speaks the same API.

llama-server -m model.gguf --host 0.0.0.0 --port 8080

Server address: http://localhost:8080/v1. Set Context window to the size you started it with.

llama-swap sits in front of several models and loads whichever one is asked for. Point FoxQuill at llama-swap’s port and use its model names.

An AI on another computer#

Use that computer’s address on your network in place of 127.0.0.1, such as http://192.168.1.50:11434/v1. The AI program has to accept connections from other computers: start Ollama with OLLAMA_HOST=0.0.0.0, or turn on Serve on Local Network in LM Studio’s server settings.

Using a service instead#

Any provider with an OpenAI-compatible endpoint works the same way: their server address, your key, their model name. Understand that your text then leaves your computer and reaches their servers.

The context budget#

A model can only hold so much at once. Context window in Settings is how much yours holds. 32,768 tokens is the default when you fill in the settings by hand, and is enough for the way FoxQuill builds context. Use sets whatever the server really uses, which is often less. The window is at least 2,048 tokens, and Max output tokens is at most a quarter of it, so there is always room for the book.

FoxQuill never sends the whole book. It builds a short brief, in this order:

  1. The book’s description and style guide.
  2. Cards that matter here.
  3. The notes you let the AI read, for this scene, this section, and the book.
  4. Summaries of earlier scenes.
  5. The end of the previous scene.
  6. The scene you are in.

What is sent#

Only what you have marked as visible.

  • A scene is sent unless you have hidden it with the eye icon on the scene.
  • A card is sent always, only when it is mentioned, or never. Set this on the card.
  • A card field you added is not sent unless you switch it on.
  • A note is sent only when The AI can read this is checked in its menu. A new Outline note starts checked. Every other note starts unchecked.
  • The book’s description and style guide are always sent.

Your statistics are never sent. The Threads board is only sent when you press Suggest beats.

Reasoning models#

FoxQuill always asks the server to turn thinking off. A reasoning model left to think will spend its whole output budget on hidden reasoning and answer with nothing. Servers that do not know the setting ignore it.