QEDAI Assistant
The right-rail AI - chat, background tasks, prompt library, and the page edit model.
AI Assistant
The AI assistant lives in the right rail. It reads your open work - scenes, world pages - and can queue edits for your review. It runs alongside you; nothing is applied without your confirmation.
Curious why it doesn't loop, guess file names, or quietly edit the wrong thing? That discipline is engineered, not hoped for - see How the AI stays on track.
Chat
Chat is the main surface. Ask questions about the open work, request a draft, ask whether a scene contradicts an earlier one. The assistant reads the current work context and answers in place.
Chat can also queue page edits: if you ask it to draft a new scene or revise a character page, it creates a Task rather than editing the file directly. You review and apply the task when you're ready.
Tasks
Tasks are background page edits. They appear in the Tasks panel (right rail, below Chat) while you keep writing. Each task shows what it plans to do; you review it, then apply it in one step.
One Cmd / Ctrl + Z undoes an applied task.
Tasks can come from:
- Chat requests ("write a scene where X confronts Y")
/agentinvoked inside a page
/agent in a page
Type /agent anywhere inside a page to describe a change to that specific page. The instruction is
queued as a Task - it doesn't edit the file inline. The Prompt library is
available as a picker inside /agent, so you can select a saved instruction instead of typing
from scratch.
Prompt library
The Prompt library (right rail, second tab) holds saved, reusable instructions. Two built-in prompts ship with the app:
| Prompt | Kind | What it does |
|---|---|---|
| Reformat | Replace | Rewrites the page to match the canonical NVS structure for that page kind |
| Placeholders | Append | Adds stub sections for any missing optional fields |
You can save your own. Prompts are either Append (new text added to the end of the page) or Replace (full page rewritten).
Append vs Replace
Every prompt - saved or ad-hoc - is one of two kinds:
| Kind | What the AI returns | When to use |
|---|---|---|
| Append | Only new text, added to the end | Drafting, scaffolding, adding sections |
| Replace | The whole page, rewritten | Reformatting, revising existing text |
The distinction matters because applying a finished edit is a single step. One undo reverts it entirely. If you use Replace on a page with existing content, everything changes at once.
Starter prompts
New to the assistant? Open a project - or grab one from the Store - and paste any of these. The AI reads your
open work and answers in place; wherever a prompt says […], drop in a name from your story.
Understand what you've written
- "What are the main plot threads, and which are still open?"
- "Summarize everything that's happened up to chapter 5."
- "Which characters carry the most weight, and who's gone quiet lately?"
- "Trace
[character]'s arc - how do they change across the story?" - "How does the relationship between
[A]and[B]develop?"
Catch problems before a reader does
- "Are there any continuity errors or plot holes in my draft?"
- "Does anything a character says or does contradict an earlier scene?"
- "Who knows about
[the secret], and when did each of them find out?" - the dramatic-irony ledger.
Plan the next move
- "Based on my corkboard, what have I planned but haven't written yet?"
- "Given the open threads, what should probably happen next?"
Get material in
- "Turn this text into NVS scenes and character pages." - for pasting in a draft or transcript.
Anything the assistant proposes to change becomes a Task you review first (see Tasks) - nothing is applied silently. It works with your own model: OpenAI, OpenRouter, an Anthropic key, or local Ollama.