Chat

What Is the AI Assistant

The AI assistant is Qritor's creative command center. In Qritor, virtually all creative operations are accomplished through conversation with AI — describe what you need in natural language, and AI understands and executes it directly.

It's not a simple chatbot, but an intelligent agent with tool-calling capabilities: it can read your novel data, create and modify modules, write and polish chapters, and manage plot lines and relation graphs. Just say "create a morally ambiguous sword master character," and AI will automatically call the appropriate tools to complete the task.

Panel Location

The AI assistant panel is on the right side of the editor, open by default. You can adjust its width by dragging the left edge.

The panel is divided into three areas from top to bottom:

• Header bar: Title, conversation history button, new chat button
• Message area: Scrollable conversation content
• Input area: Text input field, toolbar, mode selector, model selector

What Agent Can Do

In the default Agent mode, AI has full creative capabilities. Here are the operations AI can perform:

Capability	Description	Example Commands
Chapter Writing	Continue, rewrite, or generate entire chapters from scratch	"Continue writing Chapter 3", "Make this dialogue more tense"
Chapter Management	Create, rename, delete chapters	"Create Chapter 4 titled 'Undercurrents'"
Characters/Scenes/Props/Organizations	Create, edit, delete all module types	"Create a morally ambiguous sword master character"
Relationship Management	Create and modify relationships between entities	"Set Zhang San and Li Si as master and disciple"
Plot Management	Create plot events, manage storylines and foreshadowing	"Mark a turning point event in Chapter 5"
Dynamic Info	Record entity state changes as the plot progresses	"Record that the protagonist awakened their bloodline in this chapter"
Information Queries	Search characters, chapters, relationships, and more	"List all characters who have appeared"

Tool Call Visualization

When AI performs operations, the message area displays tool call cards, so you can clearly see what AI is doing:

• Each tool call appears as an expandable card
• A status indicator on the left: amber (executing), green (success), red (failed)
• Click to expand and view detailed results
• Multiple tool calls are displayed in a timeline sequence

Confirmation Mechanism

For operations that modify your work, AI doesn't execute directly — it shows the result first and waits for your confirmation:

• Creating/modifying modules: After AI generates content, the tool card auto-expands showing a preview with "Confirm" and "Reject" buttons
• Writing to chapters: The editor enters writing mode where you can preview before confirming or canceling (see Smart Writing Tools)
• Modifying chapters: The editor enters modify mode with a side-by-side comparison of changes (see Smart Writing Tools)

Query operations (like searching characters or viewing chapter content) execute automatically without confirmation.

Context Awareness

AI doesn't just read the messages you send — with each conversation, it automatically receives rich background information:

• Currently selected item: The highlighted chapter or module in the left module tree — AI automatically understands your current focus
• @mentioned modules: Characters, scenes, etc. you actively reference in your message (see @Mention)
• Novel-wide information: Settings, module structure, knowledge graph, and more

This means you don't need to re-explain background every time — AI always knows your novel.

Context Window and Auto-Compression

AI models have a context window — the total amount of information they can "see" at once. Your chat history, novel data, and @mentioned module content all consume this window.

A pie chart icon in the bottom-left of the input area shows real-time context window usage. Hover to see exact token counts and window size.

When a conversation accumulates enough messages to approach the window limit, the system automatically triggers context compression — older messages are compressed into a summary while recent messages are kept in full. After compression, AI still "remembers" the key points discussed, just without retaining every detail of all historical messages. This process is automatic and requires no manual action.

If AI's recall becomes fuzzy after multiple compressions, consider starting a new conversation and using @mention to reference the content you need. See Conversations for details.

Input Area Features

Text Input

• Type directly to chat with AI
• Enter to send, Shift+Enter for line breaks
• Input field height adjusts automatically with content

@Mention

Type @ to reference modules from your novel as context, helping AI understand your needs more precisely. See the @Mention page for detailed usage.

/Skills

Type / to trigger skill commands — these are preset professional workflows such as outline generation, character extraction, etc. Skills support fuzzy search; press Enter or click to select. See the Skills page for details.

Quick Action Buttons

The left side of the input field provides these quick buttons:

• + (Add Reference): Manually add module references or text quotes
• AI Generate: One-click AI content generation for the currently selected module (only shown for AI-enabled module types)

Message Types

Messages in the AI assistant come in several content forms:

• Text replies: AI's regular text responses, rendered in Markdown format
• Tool call cards: Visual display of AI operations (see above)
• Thinking process: If the model supports it, AI's reasoning process appears as collapsible cards, helping you understand AI's decision-making logic
• Streaming JSON preview: Real-time preview of structured data during entity extraction and similar operations

Model Selection

The model selector in the bottom-right of the input area lets you switch AI models at any time:

• Click to show available models
• Different models vary in capability, speed, and cost
• You can switch models mid-conversation
• If a model encounters issues, the error message provides a quick "switch model" shortcut

See the AI Models page for details.

Mode Switching

The mode selector in the bottom-left of the input area switches AI's operating mode. See the AI Modes page for details.

Tips

• Just state your needs: No special command syntax required — describe what you want in natural language and AI will understand and execute
• Use the confirmation mechanism: Every modification by AI requires your approval — reject and ask for adjustments if unsatisfied
• Check tool cards: Expanding tool call cards shows exactly what AI did, helping you verify results
• Pair with @mentions: When discussing specific characters or scenes, reference them with @ for more precise AI responses
• Watch context usage: Keep an eye on the pie chart indicator — the system auto-compresses long conversations, but if AI's recall becomes fuzzy, start a new conversation