AI on your terms
An AI assistant that answers to you, not the other way around
Bring your own key to the provider you trust, or run it fully local with Ollama so nothing leaves your machine. Either way, it's off until you turn it on, and it never touches your manuscript without asking first.
Download Writer Studio — freeWriter Studio's AI assistant is an ai writing assistant you control: you choose the provider, bring your own key or run it fully local with Ollama, and nothing is sent anywhere until you say so. It's a single agent, not a set of AI-only buttons, and it's off by default — the app works completely without it.
Which AI providers can I bring my own key to?
Among others, Writer Studio connects to Anthropic, OpenAI, DeepSeek, GigaChat, YandexGPT, and a custom OpenAI-compatible endpoint of your own — plus Ollama, running locally. The first time you turn AI on, you pick a provider explicitly; there's no default, because where your manuscript text goes is your call to make.
For cloud providers, you paste the key that provider gave you. A custom endpoint needs its URL, and YandexGPT needs a folder ID before it's ready to use. The choice screen states plainly where your text will go before you commit to it, and you can switch providers or update your key later in Settings.
| Provider | Needs your own key? | Where it runs |
|---|---|---|
| Anthropic | Yes | Provider's servers |
| OpenAI | Yes | Provider's servers |
| DeepSeek | Yes | Provider's servers |
| GigaChat | Yes | Provider's servers |
| YandexGPT | Yes, plus a folder ID | Provider's servers |
| Custom OpenAI-compatible endpoint | Yes, plus the endpoint URL | Wherever you point it |
| Ollama | No key needed | Your own device — nothing leaves it |
What does the assistant actually do once it's on?
It's one agent with the same reach you have: it can search your project, open scenes, read your Book Wiki, propose an edit, and read your proofreading diagnostics — because you ask it in plain language, not because it has private commands you don't. Ask it where a character first appears, to open every scene where a location shows up, or to summarize a plotline, and it answers using the same read and search tools you have, citing what it found.
AI instructions carry this further: you can write standing guidance at the project, book, part, chapter, or scene level, and a deeper instruction supplements the ones above it rather than replacing them — so a scene can add detail without overriding your book-wide style notes. What it can't do is act on its own: every AI-only shortcut is really a shortcut into this one confirmable agent.
Will the AI ever change my manuscript without asking first?
No — every change is a suggestion until you approve it. When the assistant wants to edit a scene, its output shows up as a diff you review, with Apply and Reject waiting for your decision; nothing lands until you click Apply, and an applied edit is a normal, undoable change like any other. Edit shapes are limited to replacing text, appending to the end, or inserting at the start — no silent rewrites.
The same holds for your Book Wiki: a suggested character or location entry is marked "suggested" and never becomes canonical on its own. This human-in-the-loop ai for novelists approach is the whole point of bringing your own key — control over where your text goes should come with control over what changes it.
Where does my key and my manuscript text actually go?
Nowhere you haven't chosen. Your provider key is a device-level secret — it's never written into the project folder, never exported, and never included in a backup, so sharing a project folder or a backup file never leaks it. Anything in the Trash is invisible to the assistant too, so a deleted scene never shows up in its context or its answers.
Below the input, you'll see tokens sent and received with a rough cost estimate for the current conversation. A long conversation eventually fills its context window; when that happens, older context compacts automatically to keep going, with no warning and no undo — start a new conversation if you'd rather keep a clean one. Each book keeps its own separate chat history.
Prefer not to manage a key or a provider account at all? Authr AI is the managed alternative built on the same agent — no key, subscription-based. Read the full mechanics in the AI assistant docs, or download Writer Studio — free and pick your own provider from Settings whenever you're ready.
Frequently asked questions
- Do I need to buy a subscription to use AI in Writer Studio?
- No. AI is optional and off by default — the app is fully useful without it. When you do turn it on, you choose a provider yourself: bring your own API key to a service like Anthropic or OpenAI, point it at a custom endpoint, or run Ollama locally with no key and no subscription at all.
- Can I use Writer Studio's AI assistant without sending my manuscript anywhere?
- Yes, if you choose Ollama. It runs on your own machine, so nothing about your book leaves it — no key needed either. Every other provider states plainly where your text goes before you pick it, so the choice is always yours to make.
- Will the AI edit my scenes or Book Wiki on its own?
- No. Every suggested edit — to a scene or to a Book Wiki entry — arrives as a proposal with Apply and Reject buttons, and text edits show a diff first. Nothing is written into your manuscript or your confirmed story knowledge until you approve it.
- Which AI providers can I bring my own key to?
- Anthropic, OpenAI, DeepSeek, GigaChat, YandexGPT, and any custom OpenAI-compatible endpoint, among others — plus local Ollama, which needs no key. You pick one explicitly the first time you enable AI; none is chosen for you.