Aurora AI Assistant

Aurora is Niuton's built-in AI assistant, deeply integrated into the desktop environment. It can chat, execute tools, analyze images, and interact with all Niuton apps.

Overview

• Unified Model — Powered by Qwen 3.5 35B (MoE, 3B active) with 200K context
• 66 tools — File operations, web search, app management, PIM, weather, finance, and more
• Inline Vision — Paste or drop images directly in chat for analysis
• Thinking — Visible chain-of-thought reasoning with Markdown rendering
• Adaptive — Automatically detects task type (coding, reasoning, creative, vision) and adjusts parameters
• Context-aware — Knows your files, apps, settings, date/time, and system state

Accessing Aurora

1. Aurora Orb — Click the floating orb (bottom-right) for a split-view chat
2. Team Chat — @mention Aurora in any chat room for collaborative AI
3. App Builder — Aurora powers the AI code generation pipeline

Model Selection

Users manually select their preferred model in the chat interface. The default model (Qwen 3.5 35B) is a unified model that handles all capabilities in a single model — no need to switch between separate chat, vision, or reasoning models.

Model Capabilities

Each model declares its capabilities. Unified models like Qwen 3.5 support all three:

Capability	Description	Badge
thinking	Extended reasoning / chain-of-thought	Brain
tools	Function calling (Aurora tools)	Wrench
vision	Image analysis and understanding	Eye

When only one model is configured, the model picker simplifies to a static display.

Adaptive Task Detection

Aurora automatically detects the task type from your message and adjusts sampling parameters:

Task Type	Detection Keywords	Temperature	Max Tokens
coding	code, script, programmier, function	0.6	32768
reasoning	warum, erklaere, analysiere, vergleiche	1.0	32768
creative	schreib, dichte, kreativ, story	1.0	16384
vision	bild, foto, screenshot, zeige	0.7	4096
general	(default)	0.7	16384

Desktop Vibe-Coding v3.8

Tool	Description
modify_desktop_theme	Change CSS variables (accent color, backgrounds, etc.)
create_desktop_widget	Place an HTML/CSS/JS widget on the desktop
remove_desktop_widget	Remove a widget
list_desktop_components	List all active customizations
reset_desktop_component	Reset a component to default
get_desktop_state	Get current theme, widgets, hooks state
modify_desktop_component	Write a JS/CSS override for any component
read_desktop_component	Read current override code
preview_desktop_change	10s live preview with auto-rollback
save_desktop_profile	Save entire desktop state as a profile
load_desktop_profile	Load a saved profile
modify_desktop_shortcuts	Customize keyboard shortcuts

See Desktop Vibe-Coding for details.

System Roles

Configure default models for different system functions in Settings > Aurora:

• Chat — Interactive conversations
• Coding — Code generation and analysis
• Vision — Image analysis

Inline Vision

Paste images from your clipboard (Ctrl+V) or drag & drop image files directly into the Aurora chat. The unified model analyzes images inline without requiring a separate vision model call.

• Up to 3 images per message
• Supported formats: JPEG, PNG, GIF, WebP
• Images appear as thumbnails in the attachment bar
• The model receives images as multimodal content alongside your text

Available Tools

Aurora has access to 66 tools for interacting with the system:

File Operations

Tool	Description
list_directory	List files in a directory
read_file	Read file contents
write_file	Create or overwrite a file
create_directory	Create a new directory
delete_item	Delete a file or directory
move_item	Move or rename a file
search_files	Search files by name pattern
get_quota	Get user's storage usage

Knowledge & Data

Tool	Description
list_notes	List all notes
read_note	Read a specific note
create_note	Create a new note
list_finance	List financial entries
create_finance_entry	Add income/expense
list_documents	List wiki documents
read_document	Read a wiki document
create_document	Create a wiki document

Rechner (Calculator)

Tool	Description
calculate	Evaluate a math expression
convert_unit	Convert between units
convert_currency	Convert currencies with live rates

Kontakte (Contacts)

Tool	Description
list_contacts	List contacts (optional: group, search)
search_contacts	Search by name, company, or email
create_contact	Create a new contact
update_contact	Update an existing contact
get_birthdays	Show upcoming birthdays

Kalender (Calendar)

Tool	Description
list_calendar_events	List events in a date range
create_calendar_event	Create a new event
delete_calendar_event	Delete an event
search_calendar_events	Search events by title or location

Aufgaben (Tasks)

Tool	Description
list_tasks	List tasks (filter by status, list)
create_task	Create a task with optional subtasks
complete_task	Mark task completed/uncompleted
get_my_day	Show today's "Mein Tag" tasks
get_overdue_tasks	Show overdue tasks
search_tasks	Search by title, description, or tags

Wetter (Weather)

Tool	Description
get_weather	Current weather for a location
get_forecast	7-day forecast

Sprachmemos (Voice Memos)

Tool	Description
list_voice_memos	List memos (optional: folder)
search_voice_memos	Search by title or transcript

Bildschirmfoto (Screenshots)

Tool	Description
list_screenshots	List recent screenshots

Finance (Portfolio)

Tool	Description
get_portfolio_overview	Portfolio dashboard data
list_portfolio_positions	List holdings
add_portfolio_transaction	Add buy/sell/dividend

Desktop Vibe-Coding v3.8

Tool	Description
modify_desktop_theme	Change CSS variables (accent color, backgrounds, etc.)
create_desktop_widget	Place an HTML/CSS/JS widget on the desktop
remove_desktop_widget	Remove a widget
list_desktop_components	List all active customizations
reset_desktop_component	Reset a component to default
get_desktop_state	Get current theme, widgets, hooks state
modify_desktop_component	Write a JS/CSS override for any component
read_desktop_component	Read current override code
preview_desktop_change	10s live preview with auto-rollback
save_desktop_profile	Save entire desktop state as a profile
load_desktop_profile	Load a saved profile
modify_desktop_shortcuts	Customize keyboard shortcuts

See Desktop Vibe-Coding for details.

System

Tool	Description
list_custom_apps	List user's custom apps
delete_custom_app	Delete a custom app
web_search	Search the web
fetch_webpage	Fetch and read a webpage
get_system_info	Get system information
list_chess_games	List chess games
get_chess_game	Get chess game details

Team Chat Tools

In team chat context, Aurora has additional tools:

Tool	Description
shared_list	List shared room files
shared_read	Read a shared file
shared_create	Create a shared file
shared_delete	Delete a shared file
shared_create_directory	Create shared directory
my_list	List user's private files (read-only)
my_read_files	Read user's private files (read-only)

AI Provider Configuration

Providers are configured in Settings > Aurora > Systemintegration:

┌──────────────────────────────┐
│  AI Provider Configuration   │
├──────────────────────────────┤
│  Anthropic:  [API Key]       │
│  OpenAI:     [API Key]       │
│  Google:     [API Key]       │
│  LM Studio:  [Endpoint URL]  │
└──────────────────────────────┘

Thinking Toggle

Aurora's chain-of-thought reasoning is controlled via the "Denken" button in the chat footer:

Level	Steps	Description
Aus	0	No reasoning, fastest responses
Auto	-1	Model decides depth based on complexity (v3.8)
Standard	3	Light reasoning for simple questions
Gruendlich	7	Medium reasoning for analysis tasks
Tiefgehend	12	Deep reasoning for complex problems
Maximum	20	Full chain-of-thought for hardest tasks

When enabled:

• The model uses <think>...</think> blocks for internal reasoning
• A collapsible "Denkprozess" block appears above the response
• Thinking content is rendered with full Markdown support (code blocks, lists, bold) and shows the line count

Auto-Thinking v3.8

The Auto level (thinkingSteps = -1) lets the model decide how many reasoning steps are needed. The system prompt instructs the model to analyze complexity and think accordingly — from zero steps for simple greetings to deep analysis for complex multi-step tasks. This is the recommended default for most users.

Mobile Support

Aurora is fully responsive for phone/tablet use in portrait mode:

• Fullscreen panel — On screens ≤768px, the Aurora panel opens as a fullscreen overlay
• Mode lock — Sidebar mode is enforced on mobile (split-view disabled)
• Touch targets — All buttons are minimum 44px for comfortable tapping
• iOS zoom prevention — Input fields use 16px font to prevent auto-zoom
• Safe area — Header and input respect notch/home-indicator insets via env(safe-area-inset-*)
• Simplified UI — Mode switch and input hints are hidden on mobile

Architecture

User Input (text + optional images)
    │
    ▼
niuton-orb.js (Frontend)
    │ POST /api/assistant.php
    ▼
assistant.php
    │
    ├── Detect task type (coding/reasoning/creative/vision/general)
    ├── Build system prompt (tools, context, date/time, apps)
    ├── Build multimodal content (text + inline images)
    ├── Get adaptive params for task type
    ├── Call AI provider (ai-providers.php) with adaptive options
    ├── Parse response for tool calls
    ├── Execute tools (tool-executor.php)
    ├── Feed results back to AI (with error recovery)
    ├── Extract thinking blocks from response
    └── Return final response with metadata

The tool execution loop runs up to 10 iterations in Aurora Chat and 8 iterations in Team Chat, allowing complex multi-step tasks. The response includes metadata: model name, response time, iteration count, task type, and thinking content.

v3.12 Updates v3.12

XSS Sanitizer Upgrade

The shared rendering module (chat-shared.js) now includes an upgraded XSS sanitizer applied to all Aurora message rendering:

• Event handler stripping — All on\w+ HTML event attributes (e.g., onclick, onerror, onmouseover) are removed from rendered output
• javascript: URL blocking — Any href or src attributes containing javascript: protocol are stripped
• Applied everywhere — The sanitizer runs in both the Team Chat app and the ChatPanel, ensuring consistent protection across all Aurora rendering surfaces