Quill connects the toolkit to a language model you choose, so it can help polish your messages and summarize your chats. This page explains what it is and how to point it at a model.
Quill has no intelligence of its own. It is only a tool: a pipe that passes your text to a language model you choose and connect, then shows you the reply. The toolkit does not contain, bundle, or run any AI model. Nothing happens until you point Quill at a model you control.
That model can be one you run locally on your own machine (private, free, offline) or a paid API you have an account with. The quality, speed and behaviour of Quill depend entirely on the model you attach — a small local model will be rougher than a large one.
- Improve my message — polishes the text in your chat box, from light grammar fixes (1) up to a full creative rewrite of a rough note (5), using the bot's last message as context.
- Your persona — a per-chat description of who you're playing, so Quill rewrites in character as you.
- Summarize chat — turns recent messages (or the whole chat) into factual bullet points you can save.
Quill never speaks or acts for the bot, never continues the scene for you, and never invents plot you didn't write.
On a bot create/edit page, Quill helps you build — but it stays a sounding board, never a ghost-writer.
- Character Lens — describe what your character is doing wrong and Quill gives an analytical second read of your bot's files. It looks for wording a model might misread (and tells you honestly if it doesn't think the wording is the problem). It will not agree just to please you, and it points you toward testing across models/settings before you rewrite anything.
- Import a character — upload a SillyTavern card (
.png/.json) or a plain.txtbot from anywhere, and Quill reorganises it into DreamJourney's template. You can then download it as a bot.jsonor apply it straight to the open bot page. It rehomes the content faithfully — it doesn't rewrite the character.
What Quill will never do: rewrite your bot's "guts" for you or hand you a finished character. Creating a bot is a human craft — Quill will encourage you to make the changes yourself.
Pick one in the Settings window, fill in its URL/model, then press Test connection. Each option has its own setup docs — follow the links rather than guessing.
Free, private, runs on your machine. Install it, then pull a model (e.g. ollama pull mistral-nemo).
Browser extensions must be allowed to reach it: set the OLLAMA_ORIGINS environment variable so Ollama trusts the toolkit, then restart Ollama.
A desktop app with a built-in OpenAI-compatible server. Start its local server, then use that URL.
Lightweight local server, great for GGUF models. Serves whatever model you launch it with.
Any OpenAI-compatible endpoint works — OpenAI itself, OpenRouter, or your own gateway. You supply a base URL, model name and API key.
→ OpenAI · OpenRouter
- API keys are secrets. Treat them like passwords. They're stored only in your own browser's local storage on this device — never shared with the toolkit's author or any third party. Don't paste them anywhere public or share screenshots of them.
- Costs are yours. Paid APIs charge per use. Summarizing a long chat or running many rewrites uses tokens you pay for. Local models are free.
- Your messages go where you point Quill. Whatever endpoint you configure receives the text you send it. Only use providers you trust. Local models keep everything on your own machine.
- Rotate a leaked key. If you suspect a key is exposed, revoke it in your provider's dashboard and issue a new one.
- You stay in control. Quill only acts when you press a button, and always shows you the result before anything is applied — nothing is sent on your behalf automatically.
