AI Commander — Documentation

1. Getting Started

AI Commander for Windows is a desktop control center for power users. It lets you chat with AI agents, execute system commands, automate browser workflows, schedule recurring tasks, and manage everything from one local Windows application.

Installation

1. Download the free version from mountaindevs.com or subscribe at €10/month for Pro.
2. Extract the ZIP file to a folder of your choice (e.g., C:\Tools\AICommander).
3. Run CommandTool.exe to launch the application.
4. On first launch you will be prompted to enter a license key or continue with the free tier.

2. First Launch & Initial Setup

The main workspace — chat panel, profile configuration, AI model selector, and floating sticky notes.

When AI Commander starts for the first time, the following sequence executes automatically:

1. License check — The app checks for an existing license at %LocalAppData%\AICommander\license.dat. If none is found, the License dialog opens. You can activate a Pro key or dismiss it to use the free tier.
2. Model loading — The app loads your saved models from models.dat. If no models file exists, a default model (gpt-4.1 via GitHub Copilot) is inserted automatically so you always have at least one model available.
3. Agent initialization — Four default agents are created: Assistant, FileSurfer, Commander, and WebSurfer. These can be customized later.
4. Options loading — Timeout, theme, confirmation settings, and working folder are restored from app-options.json.
5. Welcome message — The chat window displays: "Welcome to AICommander. Chat with the assistant, or ask for system actions."
6. Services start — The job scheduler, Slack integration, and Telegram bot are initialized (if configured).

What if no model is configured?

The app guarantees at least one model is always available through its EnsureDefaultModels mechanism. If your models.dat file is empty or missing, the default gpt-4.1 model (pointing to the GitHub Copilot inference endpoint) is auto-inserted. However, this default model ships with an empty API key. If you attempt to send a message without configuring a valid API key, the request will fail and you will see an error in the chat output.

3. Model Configuration

AI Commander supports multiple AI models simultaneously. You can switch between them at any time using the model dropdown in the toolbar.

The Model Configuration dialog — select from saved models or add new providers.

Adding a Model

1. Click Model in the menu bar, then choose Manage Models.
2. In the Models dialog, fill in the following fields:
   • Model Name — A display name, e.g. gpt-4.1, claude-3-opus, llama-3.
   • URL — The API endpoint. Examples:
     • OpenAI: https://api.openai.com/v1/chat/completions
     • GitHub Copilot: https://models.github.ai/inference/chat/completions
     • Anthropic: https://api.anthropic.com/v1/messages
     • Local (Ollama, LM Studio): http://127.0.0.1:11434/v1/chat/completions
   • API Key — Your secret API key from the provider. Leave empty for local models that don't require authentication.
3. Click Add to save the model.
4. Select the new model from the dropdown in the main window to start using it.

Provider Auto-Detection

The provider is automatically determined from the model name:

• Names containing claude → Anthropic provider
• Names containing gpt with Copilot endpoint → Copilot provider
• Everything else → OpenAI-compatible provider (works with any OpenAI-format API)

Model Settings

Auto-Route Models (Cost Optimization)

When enabled in Options, simple messages (under 200 characters with no commands) are automatically routed to a cheaper model (e.g., gpt-3.5-turbo) to save API costs, while complex tasks continue to use your primary model.

4. Supported AI Providers & Models

OpenAI

All OpenAI chat models are supported, including:

• gpt-4.1, gpt-4o, gpt-4-turbo
• gpt-3.5-turbo (good budget option for simple tasks)
• o1-mini, o1-preview (reasoning models)

Endpoint: https://api.openai.com/v1/chat/completions

API Key: Obtain from platform.openai.com

Anthropic (Claude)

Claude models use a dedicated Anthropic API integration:

• claude-sonnet-4-20250514, claude-3.5-sonnet
• claude-3-opus (most capable)
• claude-3-haiku (fast and affordable)

Endpoint: https://api.anthropic.com/v1/messages

API Key: Obtain from console.anthropic.com

GitHub Copilot Models

If you have a GitHub Copilot subscription, you can use the GitHub AI inference endpoint:

Endpoint: https://models.github.ai/inference/chat/completions

API Key: Your GitHub personal access token with Copilot scope.

Local / Self-Hosted Models

Any model that exposes an OpenAI-compatible API works with AI Commander:

• Ollama — http://127.0.0.1:11434/v1/chat/completions
• LM Studio — http://127.0.0.1:1234/v1/chat/completions
• text-generation-webui — http://127.0.0.1:5000/v1/chat/completions
• vLLM, LocalAI — Any OpenAI-format endpoint

5. Prompting Guide & Best Practices

AI Commander builds a system prompt from your Profile and Personality fields, plus agent-specific instructions. Good prompting dramatically improves results.

Profile & Personality

The Profile field (default: "Windows desktop automation assistant") tells the AI what it is. The Personality field (default: "Direct, reliable, and concise") shapes its tone. These are injected into every system prompt.

Effective Prompt Patterns

Be Specific About the Action

List all .log files in C:\Logs that were modified in the last 7 days and delete any larger than 100MB.

Clean up my logs.

Use Step-by-Step Instructions

1. Search for all .csv files in D:\Reports
2. Read the first file and show me the column headers
3. Create a summary of the file counts by extension

Provide Context When Needed

I'm working on a Node.js project in D:\Projects\myapp. Check if node_modules exists.
If not, run "npm install". Then run "npm test" and show me the results.

Ask for Confirmation Before Destructive Actions

Find all temporary files in C:\Temp older than 30 days. Show me the list first before deleting anything.

Prompt Tips for Command Generation

• The AI responds with JSON commands when it determines an action is needed. You do not need to write JSON yourself.
• If the AI describes what it will do but doesn't produce commands, the app automatically asks it to provide the runnable JSON.
• Mention file paths explicitly when possible — this helps the AI generate accurate commands.
• Use the Working Folder feature to avoid typing full paths repeatedly.
• For multi-step tasks, describe all steps in one message for better planning.

6. Agent Team System

AI Commander uses a team of specialized agents. Each agent has a name, description (used as system prompt context), and set of aliases. The AI can route tasks to different agents based on the nature of the request.

Default Agents

Customizing Agents

Open Agents from the menu to access the Agent Settings dialog. For each agent you can customize:

• Name — The display name shown in the chat.
• Description — The system prompt instructions that define the agent's behavior.
• Aliases — Alternative names the AI can use to route to this agent.

Agent definitions are persisted in agent-settings.json in the application directory.

How Agent Routing Works

When the AI produces a JSON response containing an agent field, the app resolves that agent by checking aliases (case-insensitive), switches the active character, and forwards the task description to the selected agent. This allows the Commander agent to delegate file work to FileSurfer, web tasks to WebSurfer, etc.

{
  "agent": "FileSurfer",
  "description": "Scan the project folder and list all .cs files"
}

7. Command Execution Loop

AI Commander's core power is its ability to execute commands on your system. When the AI determines that an action is needed, it produces JSON commands. The app parses, confirms (if configured), and executes them — then feeds results back to the AI for analysis.

Execution Flow

1. You send a message describing what you want done.
2. The AI analyzes the request and produces one or more JSON commands.
3. If Read/Write Confirmations are enabled, a dialog asks for your approval.
4. Commands execute against your system (files, shell, browser, etc.).
5. Results are captured and sent back to the AI.
6. The AI analyzes the output and either responds with a summary or issues follow-up commands.
7. This loop continues for up to 5 rounds per message.

Command Confirmation Dialogs

When confirmations are enabled, a modal dialog appears before each command executes. The dialog clearly identifies the operation so you always know exactly what will happen before you approve it.

What the Dialog Shows

• Action type badge — READ (data accessed but not modified) or WRITE (data created or modified), color-coded for quick recognition.
• Specific resource details — The dialog always names the exact resource involved:
  • File operations — full file path, e.g. Read file: C:\project\readme.md
  • Registry operations — full key path and value name, e.g. Write registry: HKLM\Software\MyApp\Version = 2.0
  • Shell commands — the complete command line, e.g. Run shell command: npm install
  • Web operations — the target URL, e.g. Read page content: https://example.com
  • System tools — the tool being opened, e.g. Open Device Manager (devmgmt.msc)
  • Process management — process name or PID, e.g. Kill process: notepad
  • Credential operations — the site name, e.g. Save credential for: github.com (user: admin)

Dialog Choices

The confirmation dialog offers three choices:

• Allow Once — Execute only this specific command.
• Allow All This Run — Execute all remaining commands in this execution loop without further prompts.
• Deny — Block this command and stop the execution.

Enabling Confirmations

Open Options from the menu. In the General tab you will find two independent toggles:

• Require Read Confirmation — prompts before any command that reads files, directories, registry keys, system info, credentials, or web content.
• Require Write Confirmation — prompts before any command that creates, modifies, or deletes files, registry values, credentials, or runs shell commands.

Both are off by default for a frictionless experience but are recommended for sensitive environments. Read and write confirmations are tracked separately — choosing "Allow All This Run" for writes does not auto-approve reads, and vice versa.

JSON Command Format

Commands are JSON objects with an action field and parameters. They can be sent individually or as arrays:

{"action": "readfile", "path": "C:\\project\\readme.md"}

[
  {"action": "dirlist", "path": "C:\\project"},
  {"action": "readfile", "path": "C:\\project\\config.json"}
]

8. Complete Command Reference

Below is every command AI Commander can execute. You can ask the AI for any of these operations using natural language.

File & Directory Operations

Search Operations

Shell Command Execution

Web & Browser Automation

Credential Management

System Information & Diagnostics

Windows Registry

System Management

Scheduling & Notes

MCP & External Integrations

9. Browser & Web Automation

AI Commander uses Playwright (Chromium) for all browser automation. The Browser menu button opens a persistent Playwright browser window that the AI can interact with. The WebSurfer agent specializes in web tasks.

Persistent Browser Profile

The Playwright browser uses a persistent profile stored locally. This means cookies, saved passwords, and session data are preserved between launches — you can log in once and stay authenticated for future sessions. Anti-bot-detection measures are applied automatically to reduce the chance of sites flagging the browser as automated.

Opening the Browser

Click Browser in the menu bar to launch a Playwright Chromium window (defaults to Google). If the browser window was previously closed, a fresh instance is created automatically. The AI can also open pages programmatically via the navigate command.

Playwright Automation

Playwright handles JavaScript-heavy sites, login flows, form filling, and more:

• Requirement: Run playwright install once to install browser binaries.
• Playwright handles page rendering, clicking, filling, scrolling, and credential management.
• The AI can chain multiple browser actions in sequence (navigate → fill form → click submit → read result).
• If the Playwright page or browser context is closed, a new instance is created on the next command.

Example Web Automation Prompt

Go to https://example.com/login, fill the username field with "admin"
and the password field with "mypassword", then click the Login button.
After login, read the content of the dashboard page and summarize it.

10. Scheduled Jobs

Pro Feature

Schedule AI tasks to run automatically at specified times. Jobs can be one-time or recurring, and each can use a different model or working directory.

The Scheduled Jobs manager — create jobs with custom prompts, models, recurrence, and folder scope. Track status and view reports.

Creating a Job

1. Open Jobs from the menu or press Ctrl+J.
2. Enter a Title and the Prompt (the message sent to AI when the job runs).
3. Set the Date using the calendar dropdown picker, and the Time using the time spinner.
4. Choose a Recurrence: Once, Daily, Hourly, or Custom interval (minutes).
5. Optionally override the AI model and working directory for this specific job.
6. Click Save.

Recurrence Options

Job Execution

The scheduler checks for due jobs every 10 seconds. When a job fires, it sends the prompt to the AI, executes any resulting commands, captures all output into a report, and marks the job as Finished or Failed. If the main window is minimized, a tray notification appears.

Parallel execution: Multiple jobs can run at the same time — each job gets its own isolated AI conversation and history, so they never interfere with each other or with your active chat session.

Output isolation: Scheduled jobs do not inject messages into the main chat window. All progress and results are captured in the job's report. The main window shows only a brief ▶ Started / ✓ Completed / ✗ Failed banner per job so your conversation is never disrupted.

Live progress: Open a job's report while it is still running (double-click the job or click View Report) to watch progress appear in real time. The report window stays open and updates automatically as each step completes. When the job finishes, the final formatted report replaces the live stream.

Fresh start on every run: Each execution — including re-runs of recurring jobs and manual Run Now — starts with a cleared report so old output never mixes with new results.

Jobs are persisted in scheduled-jobs.json and survive application restarts.

Clearing Finished Jobs

Click the Clear Finished button in the Scheduled Jobs dialog to remove all completed and failed jobs from the list. A confirmation prompt appears before deletion.

Example Scheduled Job Prompts

Check disk space on all drives. If any drive is below 10% free space,
find and list temporary files older than 7 days in C:\Temp and C:\Windows\Temp.

Read the last 50 lines of C:\Logs\app.log. If any line contains "ERROR"
or "CRITICAL", create a sticky note with the error summary.

11. Working Folder Scope

The Working Folder feature confines AI operations to a specific directory. When set, the AI is instructed to prefer paths inside that folder and use it as the base path for any file operations where an explicit path is not given.

Setting the Working Folder

1. Click the Browse button next to the working folder field in the toolbar.
2. Select a directory from the folder picker.
3. The status bar updates to show the current scope.

To clear the scope, click the Clear button. The setting is persisted across sessions in app-options.json.

How Scope Affects Behavior

• The folder path is appended to your prompt as context for the AI.
• Commands like readfile, scanfolder, findfile, and searchinfile default to this folder.
• Each scheduled job can override this with its own working directory.
• Path enforcement: If the AI attempts to use an absolute path outside the working folder, the path is automatically redirected into the working folder. This ensures all file operations stay within the configured scope.

The working folder is saved in app-options.json and restored on next launch.

12. Sticky Notes

AI Commander can create floating sticky note windows. Notes persist across sessions and can be created manually (via Ctrl+N) or by the AI during command execution.

• 6 color themes available for each note.
• Customizable font family, size, and bold settings.
• Notes are saved to %LocalAppData%\AICommander\ and restored on startup.
• Close a note by clicking its close button; it will be removed from disk.

Create a sticky note titled "Meeting Agenda" with the content:
- Review Q1 results
- Plan Q2 sprint
- Assign new team leads

13. Slack & Telegram Integrations

Pro Feature

Slack

Connect AI Commander to your Slack workspace. Messages sent to the AI Commander bot are processed through the AI, and responses are posted back to the channel.

Configure in Options → Slack section:

• Bot Token — Your Slack bot OAuth token.
• Signing Secret — For verifying Slack request signatures.
• Webhook URL — Incoming webhook for posting messages.
• Socket Mode — Enable for real-time connection via App-Level Token.

Telegram

Connect a Telegram bot for mobile AI access:

• Bot Token — Obtain from BotFather.
• Chat ID — The chat ID for receiving messages.

Both integrations route messages through the same AI pipeline and command execution system.

14. MCP Protocol Integration

AI Commander supports the Model Context Protocol (MCP) for connecting to external tool servers. MCP lets you extend AI Commander's capabilities without increasing token costs — context gathering and tool execution are delegated to MCP-compatible servers.

What is MCP?

MCP is an open standard that allows AI assistants to discover and invoke tools exposed by external servers. Instead of feeding large amounts of context into the AI model (which increases token costs), the AI can call MCP tools to retrieve only the specific data it needs.

Configuring MCP

1. Open Options from the menu.
2. Navigate to the Socket tab.
3. Enter the MCP Server URL (e.g., http://localhost:3000/mcp).
4. Check Enable MCP.
5. Click Test MCP to verify the connection and discover available tools.
6. Click Save.

How It Works

When MCP is enabled, AI Commander connects to the MCP server on startup and discovers available tools. The AI can then invoke these tools via the mcptool command:

{"action": "mcptool", "tool": "get_weather", "arguments": {"city": "London"}}

Results from MCP tools are fed back to the AI for analysis, just like any other command output.

15. WebSocket Bridge

AI Commander can connect to a WebSocket server for real-time bidirectional communication with external tools and services.

Configuring WebSocket

1. Open Options from the menu.
2. Navigate to the Socket tab.
3. Enter the WebSocket URL (e.g., wss://your-server.com/ws).
4. Click Test Connection to verify connectivity.
5. Click Save.

How It Works

When a WebSocket URL is configured, AI Commander connects on startup and begins listening for messages. Incoming messages appear in the chat window. The AI can process and respond to WebSocket messages through the same command execution pipeline.

Messages use a JSON envelope format with type and data fields for structured communication.

16. Free vs Pro Comparison

AI Commander works out of the box with a free tier. The free tier provides core functionality with a daily message limit. Pro unlocks unlimited usage and advanced features.

How the Daily Limit Works

The free tier tracks usage locally with tamper-resistant cryptographic verification. Each day resets at midnight UTC. When you approach the limit, the status bar shows remaining messages (e.g., "Free: 2/4 messages used today"). When the limit is reached, the Upgrade to Pro dialog appears. You can subscribe via Stripe (€10/month) or enter a license key to unlock unlimited access. Scheduled tasks are also limited to 2 per 24-hour period on the free tier.

17. Licensing & Activation

AI Commander uses a license key to unlock Pro features. Subscribe at mountaindevs.com for €10/month via Stripe, then activate in-app.

Activating Your License

1. Open the License dialog from the menu bar (or when prompted at first launch).
2. Enter your subscription email or the license key received after checkout.
3. The app validates your subscription against the server and activates Pro features.

Expiration & Renewal

The status bar shows your remaining Pro days. When a license expires, the app reverts to the free tier with daily limits (4 messages, 2 scheduled tasks). Your Stripe subscription renews automatically — simply keep your subscription active to maintain Pro access. Manage or cancel your subscription anytime from the product page.

18. Options Reference

The Options panel — configure your messaging provider, integration tokens, timeout, and command confirmations.

Open Options from the menu to configure application-wide settings. All options are saved to app-options.json.

Slack Configuration

Bot Token, Signing Secret, Webhook URL, Events Port, Socket Mode — configure in the Slack section of the Options dialog.

Telegram Configuration

Bot Token, Chat ID, Webhook URL — configure in the Telegram section of the Options dialog.

Socket & MCP Configuration

Configure external integrations in the Socket tab of the Options dialog:

19. Keyboard Shortcuts

20. System Requirements

21. Troubleshooting

"There was an error with the AI response"

• Check that your model has a valid API key configured (Model → Manage Models).
• Verify the API endpoint URL is correct.
• If using a local model, ensure the server is running (e.g., ollama serve).
• Check your internet connection for cloud models.

Commands are not executing

• Ensure Write/Read Confirmation dialogs are not blocking — check if a confirmation popup is hidden behind other windows.
• Verify the AI response contains valid JSON. If it does not, try rephrasing your request more specifically.
• Check the Working Folder scope — if set to a directory that doesn't exist, commands may fail.

Daily message limit reached unexpectedly

• The counter resets at midnight UTC, not local time.
• Each command-loop round counts as part of the same message (does not increment the counter).
• To check remaining messages, look at the status bar at the bottom of the window.

License won't activate

• Double-check the key you received after subscribing — copy-paste to avoid typos.
• Ensure you have an internet connection for initial activation.
• If the activation server is temporarily unreachable, the app grants offline access while you retry.

Playwright browser automation not working

• Run playwright install from a terminal in the application directory to install required browsers.
• Some websites block automated browsers. Try adding a delay between actions in your prompt.

Application won't start

• Ensure .NET 8.0 Desktop Runtime is installed: Download .NET 8.0
• Try deleting app-options.json and models.dat to reset to defaults.
• Run from an elevated (administrator) command prompt to see error output.

22. FAQ

23. Roadmap & Future Features

AI Commander is under active development. Below are planned features and improvements for upcoming releases, based on user feedback and community input.

Email Integration

Send and receive emails directly from the AI agent. Compose via SMTP (Gmail, Outlook, custom servers), read and summarize inboxes via IMAP, extract attachments, and auto-reply with AI-generated responses.

Cloud Storage Operations

Upload, download, list, and sync files with cloud providers like AWS S3, Azure Blob Storage, and Google Drive. A dedicated CloudSurfer agent for managing remote storage.

Database Querying

Execute SQL queries against SQL Server, MySQL, PostgreSQL, and SQLite databases. Browse schemas, run stored procedures, and export results to CSV or Excel — all through natural language.

Git & Version Control

Clone, pull, push, commit, and branch from within the chat. Generate commit messages, review diffs, and analyze repository history with AI assistance.

Batch Job Workflows

Chain scheduled jobs into pipelines (Job A → Job B → Job C) with conditional branching, parallel execution, and dependency graphs for complex automation sequences.

Voice Assistant

Text-to-speech for AI responses and speech-to-text for hands-free input, using built-in Windows Speech APIs for a fully voice-driven automation experience.

Screenshot & OCR Commands

Capture screenshots of the full desktop or individual windows, and extract text from images using built-in OCR. Useful for monitoring, reporting, and visual data extraction.

Prompt Templates

Save and reuse common prompts with variable placeholders. Build a library of templates for recurring tasks and share them across scheduled jobs and agents.

Custom Command Plugins

A DLL-based plugin system that lets you define your own JSON commands, extending AI Commander's capabilities without modifying the core application.

REST API & Webhook Commands

Send HTTP requests to external REST APIs and webhooks directly from the command loop. Trigger AI workflows from external events and integrate with third-party services. (WebSocket and MCP integrations are already available — see sections 14 and 15.)

24. Community & Support

Need help, want to share a workflow, or just chat with other AI Commander users? Join the official community channels.

Discord

Our primary community hub. Get real-time help, share tips, report bugs, and discuss upcoming features with other users and the development team.

Subscribe

Get Pro access for €10/month via Stripe. Your subscription directly funds development and new features.

Bug Reports & Feature Requests

Found a bug? Drop details in the Discord #support channel with your OS version, model configuration, and steps to reproduce. Feature requests are welcome in #suggestions.