AI GO is an AI enablement platform for SMEs, offering ERP AI integration, process automation, and AI customer service. No IT department needed — see results within 30 days.

Do I need to replace my existing ERP?

No. AI GO connects via API, adding AI capabilities on top of your existing ERP system while protecting your current investment.

How much does AI GO cost?

AI GO offers flexible plans based on features and usage, ranging from approximately NT$8,000 to NT$30,000/month — saving over 90% compared to building AI systems in-house. Schedule a free assessment.

Can I use AI GO without a technical background?

Yes. AI GO uses a no-code interface — just describe your needs and the system generates the corresponding AI features. No engineers or technical expertise required.

How long until I see results?

Standard onboarding takes 2–4 weeks, with measurable results typically visible within 30 days, including reduced process time and labor savings.

AI Assistant Capability Integration and Desktop Agent

To allow the AI assistant to not only access internal data but also proactively execute complex daily business tasks, the AI GO platform provides AI Skills configuration and Model Context Protocol (MCP) integration. Furthermore, through the Desktop Agent, the AI assistant can operate your local computer's terminal and software interfaces upon receiving your authorization.

1. AI Assistant Capability Integration (MCP & AI Skills)

Administrators can directly empower the AI assistant to perform external operations and execute specific Standard Operating Procedures (SOPs) in the "Assistant Capability Settings".

Entering Path: Click "Assistant Capability Settings" under the "System and Maintenance" section in the left navigation bar.

MCP Tool Authorization Settings (Model Context Protocol)

What is MCP: MCP is an open standard for assistant tools. You can input the URL and authorization Token of an externally developed MCP server into the system.
Function Expansion: Once connected, the AI assistant automatically acquires all tools registered on the MCP server (e.g., reading/writing Google Calendar, sending Slack notifications, reading/writing GitHub), enabling the AI assistant to collaborate across multiple systems.

AI Skills Management

Definition of Skill: A "Skill" represents a set of structured SOP execution steps. The AI assistant will execute them sequentially when receiving corresponding commands or meeting trigger rules.
Methods of Skill Generation:
- AI Automatic Accumulation: When the AI assistant successfully executes a complex, repetitive workflow in daily collaboration, it automatically accumulates and saves those steps as a skill.
- Manual Creation: Administrators can click "Create Skill Manually" to write the skill name, category, trigger keywords, and specific sequential steps.
Performance Tracking and Management:
- The list page displays the "Usage Count" and "Execution Success Rate" of each skill, assisting administrators in continuous SOP optimization.
- Administrators can toggle any specific skill to enabled or disabled, or edit and delete it.

2. Desktop Agent (@flyidea/ai-go-mcp)

This is one of the advanced features of AI GO: allowing the web-end AI assistant to operate your local computer under your authorization.

Installation and Activation

Local Installation: Run the following command in your local computer's terminal (Node.js environment required locally):
```
npx @flyidea/ai-go-mcp auth login
```
The browser will automatically open the AI GO authorization page, and your personal token will be securely stored locally after logging in.
Start the Agent:
```
npx @flyidea/ai-go-mcp start
```
Once started, the local machine will establish a secure, real-time bidirectional connection with the AI GO server via Server-Sent Events (SSE).

Remote Control Functions and Capabilities

When the Desktop Agent is connected, the web-end AI assistant obtains the following local control permissions when executing complex tasks:

CLI Terminal Control: The AI assistant can run commands directly on your local machine and stream the terminal output back to the web screen (e.g., assisting with dependency installation, running builds).
Screen and Mouse/Keyboard Control: The AI assistant compresses the local screen into WebP images to report to the web, simulating mouse movement, clicks, and keyboard inputs to operate local software GUIs for you.
Local File Operations: Read local files or list directories.

Local Security Protection Mechanisms

To ensure local computer security, the Desktop Agent sets strict defense mechanisms on the local side:

Dangerous Command Blocking: Blocks destructive commands directly (e.g., rm -rf /, format C:) and detects indirect execution (such as executing encrypted strings via Python/PowerShell).
Directory Traversal Prevention: Strictly prohibits reading system core directories (e.g., System32, /etc) and sensitive credentials files (e.g., .ssh/).
Secondary Confirmation for High-Risk Operations: When executing high-risk commands such as sudo, pip install, or rm -r, the local terminal pauses and pops up a prompt. The user must manually approve it in front of the computer before the AI assistant can continue execution.