StealthAI

StealthAI is an intelligent assistant built into Stealth that can perform real system operations on your Windows computer through natural conversation. Unlike chatbots that only provide information, StealthAI can actually execute tasks like cleaning files, managing processes, organizing folders, downloading software, scheduling reminders, controlling windows, and much more.

Getting Started with StealthAI

Open StealthAI from Settings → Tools, or use the keyboard shortcut. You'll see a chat interface where you can type requests in plain English. StealthAI understands context, so you can have natural conversations about what you want to accomplish.

For example, instead of navigating through Windows settings, you can simply type:

"Clean up my temporary files and tell me how much space was freed"

StealthAI will locate temporary files, calculate their size, remove them safely, and report back with the results.

AI Provider Options

StealthAI supports three different AI providers, each with distinct advantages. Choose the one that best fits your needs:

Stealth AI (Built-in)

Stealth AI is included with your paid Stealth subscription plan. This provider uses a built-in API key managed by Stealth, so you don't need to configure anything. Simply select "Stealth AI" as your provider and start using the assistant immediately.

Key Features:

• No API key configuration required - works out of the box
• Includes a monthly token allowance based on your subscription plan
• Additional token packages available for purchase if you exceed your monthly allowance
• Access to approved models configured by Stealth administrators
• Automatic model updates - new models become available as they're approved
• Seamless integration with your Stealth account

When using Stealth AI, the available models are dynamically loaded from Stealth's servers. This ensures you always have access to the latest approved models without needing to update the application. The model list is refreshed each time you open the AI settings window.

Key provisioning. Stealth AI keys are tied to your license + hardware ID and are fetched on demand from admin.stealth.online. The key never leaves your machine and is regenerated automatically if your license changes.

OpenAI (Your Own Key)

If you prefer to use your own OpenAI API key, select "OpenAI" as your provider. This gives you full control over your API usage and billing directly with OpenAI.

Configuration:

1. Select "OpenAI" as your provider
2. Enter your OpenAI API key in the API Configuration section
3. Choose your preferred model from the full OpenAI model list
4. Save your settings

Advantages:

• Direct control over API costs and usage
• Access to the complete OpenAI model catalog
• No dependency on Stealth's token allocation
• Use your existing OpenAI credits or billing

When using your own OpenAI key, you have access to all OpenAI models including GPT-5, GPT-4o, o1, o3, and GPT-3.5 variants. Your API key is stored securely on your local machine and never transmitted to Stealth servers.

Local LLM (Offline)

For maximum privacy and zero cost, use a local LLM that runs entirely on your computer. This option requires you to have a local AI service running (like Ollama) but ensures no data ever leaves your device.

Supported Local Providers:

Configuration Steps:

1. Install a local LLM service - Download and install Ollama or your preferred local LLM server
2. Select "Local LLM (Offline)" as your provider
3. Use Auto-Detect - Click the "Auto-Detect" button to scan for running local LLM services on your network
4. Or configure manually:
   • Select your provider type (Ollama, LlamaCpp, etc.)
   • Enter the base URL where your local service is running (default: http://localhost:11434 for Ollama)
   • Enter the model name (e.g., "llama3:latest" for Ollama)
5. Test Connection - Click "Test Connection & Load Models" to verify connectivity and see available models
6. Select a model - Click on any detected model from the available models list to use it

Auto-Detection Feature:

The auto-detect feature scans your local network for running LLM services. When services are found, they appear in a list with their base URL, provider type, and available model count. Simply click on a detected service to automatically configure StealthAI to use it.

Advantages of Local LLM:

• Complete privacy - no data leaves your device
• No API costs - runs entirely on your hardware
• Works offline - no internet connection required
• Full control over models and versions
• No rate limits or usage restrictions

Performance Considerations:

Local LLMs require sufficient system resources. For best performance, ensure you have:

• At least 8GB RAM (16GB+ recommended for larger models)
• Adequate CPU power or a compatible GPU for acceleration
• Enough storage space for model files (models can be several GB)

Response times will be slower than cloud-based AI, but you gain complete privacy and control.

AI Behavior Settings

The AI Behavior section controls how StealthAI responds to your requests. These settings affect the quality, length, creativity, and safety of AI responses.

Model Selection

The available models depend on which AI provider you're using. The model dropdown is dynamically populated based on your provider selection and current configuration.

Stealth AI Provider:

When using Stealth AI (built-in provider), the available models are loaded directly from Stealth's admin panel. The model list is fetched from the server each time you open AI settings, ensuring you always see the current approved models.

Stealth administrators configure which models are available through the admin panel. This means:

• The model list can change as new models are approved or removed
• You'll only see models that are currently enabled and approved
• The list updates automatically - no application restart needed
• Models are filtered to exclude voice-only models (TTS/Transcribe) which are for the voice agent only

If no models are configured in the admin panel, the system falls back to default models (GPT-4o, GPT-4o Mini, GPT-3.5 Turbo). The server also provides a default model recommendation, which is automatically selected if your current model is no longer available.

OpenAI Provider:

When using your own OpenAI API key, you have access to a standard list of OpenAI models. This includes GPT-5, GPT-4o, o1, o3, GPT-4, and GPT-3.5 variants. The exact models available may depend on your OpenAI account's API access level and what models OpenAI currently offers.

Local LLM Provider:

When using a local LLM, available models depend entirely on what you've installed in your local LLM service. For example:

• If using Ollama, models are whatever you've downloaded (e.g., llama3:latest, codellama, mistral)
• The "Test Connection & Load Models" button will discover and list all available models from your local service
• You can manually enter any model name that exists in your local LLM installation

Model Selection Tips:

Choose models based on your needs:

• For complex system operations and reasoning tasks, use more capable models
• For simple, quick tasks, faster/lighter models may be sufficient
• Consider token costs if using paid providers - larger models typically cost more per request
• With Stealth AI provider, models are pre-configured for optimal performance and cost balance

Safety Mode

Safety Mode controls how strictly StealthAI enforces safety restrictions when executing system operations:

Strict	Maximum safety - StealthAI will be very cautious about executing potentially destructive operations. It will ask for explicit confirmation more frequently and may refuse some risky operations entirely. Recommended for production environments or when you want maximum protection.
Moderate (Default)	Balanced approach - StealthAI will execute most operations but will still ask for confirmation on significant changes. This is the recommended setting for most users, providing a good balance between safety and functionality.
Permissive	Minimal restrictions - StealthAI will execute operations more freely with fewer confirmation prompts. Use this when you trust the AI and want faster workflow, but be aware that it may perform operations you didn't intend.

Response Length

Controls the maximum length of AI responses. This setting affects how detailed and comprehensive the AI's answers will be:

Short (1000 tokens)	Brief, concise responses. Good for quick answers and simple tasks. Faster and uses fewer tokens.
Medium (1500 tokens) - Default	Balanced response length. Suitable for most tasks with adequate detail without being excessive.
Long (2000 tokens)	Detailed, comprehensive responses. Good for complex explanations, multi-step instructions, or when you need thorough information.
Extended (3000 tokens)	Maximum detail. Use when you need very comprehensive responses, detailed analysis, or extensive explanations. Note: Higher token usage may increase costs with paid providers.

Creativity Level (Temperature)

The temperature slider controls how creative or deterministic the AI's responses are. This is a value between 0.0 and 1.0:

• 0.0 - 0.3 (Conservative) - More deterministic, focused, and consistent responses. The AI will stick closely to facts and be less creative. Recommended for system operations, technical tasks, and when you need precise, repeatable results.
• 0.4 - 0.7 (Balanced) - Moderate creativity. Good balance between consistency and flexibility. Suitable for most general-purpose tasks.
• 0.8 - 1.0 (Creative) - More varied, creative, and unpredictable responses. The AI will be more exploratory and may provide different answers to the same question. Use for brainstorming, creative writing, or when you want diverse perspectives.

Default: 0.3 - This conservative setting ensures StealthAI provides consistent, reliable responses for system operations while still allowing some flexibility.

GPT-5 Advanced Settings

When you select any GPT-5 model (GPT-5, GPT-5 Mini, or GPT-5 Nano), an additional "GPT-5 Advanced Settings" section appears. These settings are specific to GPT-5's advanced capabilities and allow fine-tuned control over its behavior.

Reasoning Effort

GPT-5 includes advanced reasoning capabilities. The Reasoning Effort setting controls how deeply the model thinks through problems before responding:

Auto (Recommended)	The model automatically adjusts reasoning depth based on query complexity. Simple questions get quick answers, complex problems get deep analysis. This provides the best balance of speed and quality.
Minimal	Fastest responses with minimal reasoning. Use for simple, straightforward tasks where speed is more important than deep analysis.
Low	Quick thinking with basic reasoning. Good for most everyday tasks.
Medium	Balanced reasoning depth. Suitable for moderately complex tasks requiring some analysis.
High	Deep analysis and thorough reasoning. Use for complex problems, multi-step operations, or when you need the most thorough response possible. Responses will take longer but will be more comprehensive.

How Auto Mode Works:

In Auto mode, GPT-5 analyzes your request and determines the appropriate reasoning depth. A simple question like "What's the weather?" gets minimal reasoning for a quick answer. A complex request like "Analyze my system performance and optimize it" triggers high reasoning effort for thorough analysis. This ensures you get the right level of thinking for each task without manual adjustment.

Response Verbosity

Controls how detailed and verbose GPT-5's responses are, independent of the Response Length setting:

Low	Concise, to-the-point responses. The AI provides essential information without extra explanation. Best for quick tasks and when you want fast, direct answers.
Medium (Default)	Balanced detail level. Responses include necessary context and explanations without being excessive. Good for most use cases.
High	Detailed, comprehensive responses with extensive explanations, context, and additional information. Use when you want thorough understanding of what the AI is doing and why.

Note: Response Verbosity works in conjunction with Response Length. A high verbosity setting with a short response length may still be limited by the token limit, while low verbosity with extended length allows for longer but more focused responses.

Enable Custom Tools (CFG)

This checkbox enables GPT-5's advanced Context-Free Grammar (CFG) tool calling system. When enabled, GPT-5 can use freeform tool calling with grammar constraints for safer command execution.

What CFG Does:

Traditional tool calling requires predefined function schemas. CFG allows GPT-5 to construct commands more flexibly while still enforcing safety constraints through grammar rules. This enables:

• More natural command construction
• Better handling of edge cases and complex scenarios
• Safer execution through grammar-based validation
• More intelligent tool selection and parameter generation

When to Enable:

Keep this enabled (default) for the best experience. CFG provides more intelligent tool usage while maintaining safety. Disable only if you encounter compatibility issues with specific operations or prefer more traditional, rigid tool calling.

Vision Requests

When an image is attached to a GPT-5 model, Stealth automatically routes the request to GPT-4o (which natively handles vision) and returns the result. You don't need to switch models manually - vision Just Works regardless of the GPT-5 variant you have selected.

Advanced Settings

Enable Detailed Logging

When enabled, StealthAI logs detailed information about all operations, function calls, and responses. This is useful for:

• Debugging issues with AI responses
• Understanding what operations the AI is performing
• Tracking token usage and API calls
• Reviewing conversation history for troubleshooting

Logs are stored locally and can help identify issues or optimize your usage patterns. Enable this if you want visibility into StealthAI's operations, or disable it for better performance and privacy.

Remember Conversation History

When enabled, StealthAI maintains context from previous messages in your conversation. This allows you to:

• Reference earlier parts of the conversation without repeating context
• Build on previous requests naturally
• Have multi-turn conversations about complex tasks
• Get consistent responses that understand the full context

For example, if you ask "Show me large files in Downloads" and then say "Delete the biggest one," StealthAI will remember which files were shown and know which one to delete.

Disable this if you want each message to be treated independently, or if you're concerned about context being maintained across sessions.

Global Right-Click AI Menu

When enabled, adds an "Ask Stealth AI" option to Windows' right-click context menu. This allows you to:

• Right-click on any file or folder and ask StealthAI about it
• Get context-aware help for selected items
• Perform operations on files directly from Windows Explorer
• Access StealthAI quickly from anywhere in Windows

Example Usage:

Right-click a folder → "Ask Stealth AI" → "What's taking up space in this folder?"

The context menu integration requires administrator privileges to register. Stealth will prompt you for elevation when enabling this feature.

Voice Agent

StealthAI includes a powerful voice interface that lets you interact hands-free. Access the voice agent from the StealthAI interface to speak your requests instead of typing.

Voice Features

Multi-Language Support - Choose from multiple languages for voice interaction. The voice agent can understand and respond in your preferred language, making it accessible to users worldwide.

Push-to-Talk Mode (Default) - Hold a key or button to speak, release to process. This prevents accidental activation and gives you control over when the agent listens. Ideal for office environments or when you want precise control.

Continuous Listening - Optional always-on listening mode. The agent continuously listens for your voice commands. Useful for hands-free operation, but requires a quiet environment to avoid false activations.

Visual Feedback - Animated orb responds to your voice with real-time audio visualization. The orb changes color and animation based on your voice input, providing immediate visual confirmation that you're being heard.

Silent Execution - Voice agent executes tools silently and responds with direct answers. Unlike text mode where you might see "I'll search for that," the voice agent simply provides the answer after executing necessary operations. This creates a more natural conversation experience.

Screen Context - Can capture and analyze your screen for visual tasks. When you ask questions about what's on your screen, the voice agent can see and understand visual elements, making it capable of tasks like "What's on my desktop?" or "Help me fill out this form."

Reminder Delivery - Scheduled reminders are delivered via voice when triggered. When a scheduled task or reminder activates, the voice agent will speak the reminder to you, ensuring you don't miss important notifications.

Language Selection - Set your preferred language for voice interaction. The agent will understand your speech and respond in the selected language, even if your speech contains some words from other languages or background noise.

Voice Agent Capabilities

The voice agent uses the same powerful capabilities as text-based StealthAI, including all system operations, web search, downloads, and automation features. Everything you can do via text, you can do via voice.

Voice <-> Chat Conversation Import

After completing a voice session, you can import the voice conversation into your chat history so the transcript is preserved alongside your text conversations. The full back-and-forth - your spoken prompts and the agent's spoken responses - becomes a regular chat thread you can search, reference, and continue in text mode.

Projects (Workspaces)

The agent window supports Projects - folders that group related conversations and carry their own custom instructions. Use them to keep work, personal, code, and one-off chats organized.

• Sidebar shows expandable project folders with a list of conversations inside each
• Drag conversations into a project to group them
• Per-project custom instructions - system-prompt addendum applied to every conversation in that project
• Expand / collapse projects to keep the sidebar tidy
• Delete a project (conversations stay)
• Create new conversations directly inside a project
Note: Projects are just a UI organization layer for conversations. They don't change how StealthAI accesses your filesystem - file operations still operate on the paths you mention or the working directory.

Developer Tools

In addition to the system-management tools, StealthAI ships with a developer-focused tool set so you can use it as a coding assistant. These are exposed automatically - no special mode required.

read_file / open_file	Read or open any file by path
write_file	Create new files with content
edit_file	Exact-match string replacement with optional replace_all, plus quote/whitespace fallbacks for resilient edits
find_files	Glob-aware filesystem search, skips node_modules / .git / build dirs by default, sorts by modification time
search_in_files	Regex content search with globs, context lines, whole-word, binary skip
manage_project	Scaffold new projects (Node, Python, React, HTML, Electron, Express), git init, dependency install via npm/yarn/pnpm/pip
run_cmd	Run CMD commands with persistent session CWD, timeouts, streaming
run_powershell	Run PowerShell with persistent session CWD, timeouts, streaming, optional elevation

Examples:

"Find every TODO in this folder and list the file and line"
"Scaffold a new React app called dashboard in C:\projects\"
"Edit config.json to change port from 3000 to 4000"
"Run npm install in the current project"

The persistent session CWD means follow-up commands run in the directory the previous one moved to - cd src then npm test works as expected.

Complete Capabilities

StealthAI has access to 40+ system functions organized into these categories:

System Maintenance

Keep your computer running smoothly with automated maintenance tasks. StealthAI can clear caches, remove temporary files, empty the recycle bin, clean up Windows update remnants, defragment disks, create system restore points, scan and optimize the Windows registry, and check for Windows updates. It understands which files are safe to remove and which should be preserved.

File Management

Organize your files without manual dragging and dropping. StealthAI can list directory contents (desktop, downloads, documents, etc.), read and open files (including system files like the hosts file), copy, move, and delete files and directories, create directories and manage file permissions, compress and extract archives, find duplicate files, and sort files by type, size, or date.

Process & Service Control

Monitor and manage running applications and Windows services. StealthAI can view what's using CPU or memory, close unresponsive programs, identify unknown processes, start, stop, and restart Windows services, manage service startup types, and create or delete services.

System Information

Get detailed information about your computer without digging through Windows settings. Ask about hardware specifications (CPU, RAM, GPU, drives), disk space usage and analysis, memory usage and performance metrics, installed applications with versions and sizes, network configuration and status, OS version and Windows build information, and event logs and system diagnostics.

Network Operations

Diagnose and fix network issues. StealthAI can run comprehensive network diagnostics (connectivity, DNS, ping, traceroute, speed), flush DNS cache, reset network adapters, check firewall status and configuration, test internet speed, and automatically fix common network issues.

Software Management

Install, uninstall, and manage applications. StealthAI can list all installed applications, uninstall applications using multiple methods (winget, MSI, Control Panel), download software by name using intelligent download that finds official sources, download files from URLs with progress tracking, and manage startup programs.

Window Management & Automation

Control windows and automate GUI interactions. StealthAI can minimize, maximize, and restore windows, move and resize windows, arrange, tile, or cascade windows, scroll pages up or down, focus specific windows by title or process, and open and close applications programmatically.

Advanced Automation

Complex automation sequences for system configuration. StealthAI can perform registry operations (read, write, delete, export, import), manage user accounts (create, delete, change passwords, manage groups), configure system settings (power plans, environment variables, system settings), configure network settings (reset adapters, flush DNS, reset network stack), and execute automation sequences with error handling.

Scheduled Tasks & Reminders

Schedule future actions and reminders. StealthAI can create scheduled tasks to execute at specific times, set reminders that trigger voice notifications, schedule recurring tasks, manage and query existing scheduled tasks, and show Windows toast notifications.

Information Retrieval

Get current information from the web. StealthAI can search the web for latest information, software versions, and technical specs, provide weather information with current conditions and forecasts, retrieve news and current events, and access technical documentation and solutions.

Security & Diagnostics

Protect and analyze your system. StealthAI can run Windows Defender malware scans (quick, full, or custom), check firewall status and rules, scan registry for security issues, check event logs for errors and warnings, and perform comprehensive system health scans.

Command Execution

Run Windows commands and scripts. StealthAI can execute CMD commands, run PowerShell scripts (with optional elevation), create interactive console sessions with visible input/output, and maintain persistent console sessions.

Visual Features

Screen interaction and visual assistance. StealthAI can capture screenshots, point out elements on screen, show highlights on screen, and analyze screen content for context-aware responses.

Example Conversations

Task	What to say
Free up disk space	"Clean up my system and free as much disk space as possible"
Find large files	"Show me the largest files in my Downloads folder"
Check performance	"What's using the most CPU right now?"
Organize files	"Sort my desktop files into folders by type"
Download software	"Download OpenRGB for me"
Schedule reminder	"Remind me to check email in 30 minutes"
Window control	"Scroll down on this page" or "Minimize all windows"
Network diagnostics	"Why is my internet connection slow?"
Weather info	"What's the weather in 44124?"
System info	"Tell me about my computer's hardware"

Safety and Permissions

StealthAI is designed with safety as a priority. Before performing any action that modifies your system, it will:

1. Explain what it intends to do
2. Show you which files or settings will be affected
3. Ask for your confirmation before proceeding

For significant changes, StealthAI may create a restore point first, allowing you to undo the operation if needed. You're always in control. Nothing happens without your explicit approval.

The Safety Mode setting (Strict, Moderate, or Permissive) controls how frequently confirmations are required and how cautious the AI is about potentially destructive operations.

Tips for Best Results

Be specific. "Clean my temp files" works, but "Clean temporary files from Chrome and Windows, then empty the recycle bin" gives better results.

Ask for explanations. If you're unsure about an action, ask StealthAI to explain what it will do before confirming.

Use follow-ups. StealthAI remembers context within a conversation (when "Remember conversation history" is enabled). After asking about disk space, you can say "delete the largest ones" without repeating the context.

Start simple. Begin with straightforward tasks to learn how StealthAI works before attempting complex operations.

Voice mode tips. In voice mode, speak naturally and clearly. The agent executes tools silently and responds with direct answers. Use push-to-talk for better control over when the agent listens. Ensure your microphone has proper permissions and is working correctly.

Model selection. Use GPT-5 for complex tasks requiring reasoning. Use GPT-5 Mini or GPT-4o Mini for faster, everyday tasks. Use GPT-3.5 Turbo for simple, quick operations when speed is more important than capability.

Temperature settings. Keep temperature at 0.3 (conservative) for system operations to ensure consistent, reliable results. Increase to 0.5-0.7 for creative tasks or when you want more varied responses.

Response length. Use Medium (1500 tokens) for most tasks. Increase to Long or Extended for complex explanations. Use Short for quick, simple queries to save tokens.