Go to the left function menu "AI Features” under “AI”, click the “+Create AI".
2. Name your AI
Select the “Basic Settings” tab, in the “AI Name” field, enter the AI’s name. You can name it based on the AI’s main task, e.g., XX AI Customer Service, XX Regulations Query Helper, XX Project Intelligent Assistant.
One account can create multiple AIs (the quantity is limited by the purchased plan)
3. Choose RAG to make the AI smarter and answers more precise
What is RAG
You can think of RAG as a combination of “a chat-savvy assistant + a librarian who’s great at research.”
A typical AI is like someone with excellent memory and storytelling skills, but can only tell what it learned before. When the AI is paired with RAG, it’s as if before answering, itfirst goes to the library to find the latest informationand then organizes what it finds into its own words and replies clearly to you.
On the MaiAgent platform, this “library” is our Knowledge. The AI will use RAG to retrieve relevant data from the knowledge base, making answers more accurate, timely, and aligned with needs.
The setup for the knowledge base will be explained in detail in the next section.
In addition to the RAG technology mentioned at the OpenAI developer conference, MaiAgent RAG also integrates various classic NLP algorithms and proprietary retrieval techniques. Comparing the accuracy of responses using internal datasets versus OpenAI RAG, both achieve 95% response precision.
RAG setup
Select the “RAG Settings” tab, in the “RAG” dropdown menu, choose different RAG (Retrieval-Augmented Generation). If there are no special requirements,the default is MaiAgent RAG.
Compared to OpenAI RAG, MaiAgent RAG provides more additional features, can flexibly meet different deployment needs, handle more diverse data formats, and offer a more powerful retrieval and generation experience.
4. Choose a model to pick a smart brain for your AI!
Purpose of selecting a language model
An AI’s performance largely depends on its brain—the language model (LLM) it uses. In this step, you can choose different models according to your needs. Response speed, comprehension, depth, and naturalness of answers will all be affected.
Choosing the right model is like turning on efficiency mode for your AI, tailoring the best experience for your use case!
Select the “RAG Settings” tab, in the “LLM model” dropdown menu, you can pick different large language models. If there are no special requirements,the default is Claude 3.7 Sonnet.
5. Create role instructions for your scenarios
To better fit different application needs, you can set “role instructions” so the AI’s style and content align with the scenario.
Hallucination-free generative AI response mechanism
MaiAgent’s “hallucination-free generative AI response mechanism” ensures high accuracy when the AI answers questions. When faced with uncertain or out-of-scope questions, it will candidly state its limitations rather than fabricate answers, providing users with a more reliable and trustworthy AI interaction experience. Its importance for industries and the public sector is as follows:
Financial services:
When handling investment advice and risk assessment, AI must base analyses on accurate data to avoid misinformation leading to wrong investment decisions. When information is insufficient or uncertain, the system clearly informs the user, ensuring reliability of decisions.
Healthcare:
When assisting with medical diagnosis or drug consultation, AI must strictly follow established medical knowledge and must not fabricate suggestions that could mislead patients. For novel or unverified medical information, the system will clearly indicate the need for further professional consultation.
Manufacturing:
In applications such as production process optimization and quality control, AI must provide recommendations based on actual production data and validated methods to avoid losses caused by inaccurate predictions.
Education:
When assisting teaching and answering students’ questions, AI needs to provide accurate knowledge rather than misinformation that could mislead learning. For complex or ambiguous concepts, the system will be upfront about its understanding limits.
Legal:
When providing legal information and advice, AI must rely on current laws and precedents, not speculative suggestions that may pose legal risks. The system will clearly highlight matters requiring confirmation by a professional lawyer.
Customer support:
When handling customer inquiries, AI must provide accurate product and service information. For uncertain issues, it will immediately refer to relevant professionals to avoid misleading customers.
Government policy consultation:
When providing the public with policy information and service guidance, AI must answer based on the latest and correct regulations and procedures, avoiding outdated or incorrect information. When encountering complex issues or those requiring professional judgment, the system will clearly advise seeking assistance from relevant departments.
Public service decision-making:
When helping governments evaluate public works, social welfare, and other decisions, AI must analyze based on real data and studies, and clearly state uncertainties in predictions to ensure policy reliability.
Emergency response management:
When handling emergencies such as natural disasters and public health events, AI must provide accurate information and guidance, and must not generate false messages that could mislead the public and affect response effectiveness.
Choose the appropriate answer mode and create instructions
Select “Answer Mode” tab. From free-form conversation to highly structured responses, meet various business needs. Each mode has its own characteristics and scope. Choose flexibly based on actual usage scenarios.
Applicable scenarios
Answer freely based on the question. The AI will use context and knowledge base content to generate the most suitable response. Suitable for most Q&A situations.
Operation flow
Select the “Answer mode settings” tab, choose the answer mode “General (default)”, in the “Role instructions” field, enter the role instructions you set for this AI. For output format, you can choose plain text or JSON.
Use case: Website customer service assistant
If you want to create a website customer service assistant for “MaiAgent - AI Development Platform,” you can enter the AI’s response settings in the “Role Instructions” field to clearly define its response style and scope of duties.
When you select “Answer Mode: Template”, your AI’s hallucination rate drops to zero, because the AI will strictly answer based on your knowledge base and FAQs, switching to category-based replies rather than having the LLM generate content. Using the template system to generate answers guarantees 100% hallucination-free responses.
Applicable scenarios
Scenarios requiring unified response formats, such as standard inquiry flows, reports, etc.
Example operation
Suppose you want to build a “Tainan City 1999” customer service AI to respond to municipal questions in real time.
2. Fill in the reply template (use the template system to generate responses)
If you haven’t created a knowledge base and FAQ yet, the default instructions for the reply template are as follows.
We recommend creating and adding a knowledge base or FAQ first to use the reply template, and the formatmust be table-type files such as Excel, CSV, json, or jsonl.
In this scenario, I have an AI “Tainan City 1999” and added “Tainan 1999 FAQ” to the knowledge base.
Fields in the FAQ include question, answer, agency, category, and release time.
We recommend uploading Excel, CSV, json, jsonl and other table-type files, with fields that must includetitles and corresponding content.
Now go back to edit the reply template and click the “Initialize Reply Template” in the top right. You will see the system generate sample responses based on the documents you just uploaded to the knowledge base.
You can now edit and format the reply text.
The [] {} parts are system instructions, so just handle the text parts.
📍 Loop: Fill in the corresponding document filename (specify which document to use for generating replies)
I modified the following parts:
When we return to the AI Q&A interface and ask related questions, the AI’s reply will strictly follow the editedtemplate formatand FAQ file contentto respond.
3. Fill in the “Unable to answer” template
If the AI determines there is no relevant data, it will reply according to the “Unable to Answer Template”.
Finally click the “Save” button to finish.
Applicable scenarios
Combines general answers and templates. Some content uses the template format, while the rest is free-form. Suitable for scenarios requiring partially structured responses.
Operational suggestions
At this time, the role instructions may conflict with “Answer Mode: Template”, so when choosing “Answer Mode: Hybrid”, it’s recommended to rewrite role instructions as high-level principles.
Such as the role’s tasks, interaction principles, communication tone, etc.
Applicable scenarios
Suitable for specific task scenarios such as knowledge management, summarization, and ideation
Operation flow
Please go to the “Answer Mode Settings” tab and select “Workflow” as the reply mode.
This mode suits task-oriented applications such as:
Knowledge management: Assist in consolidating, retrieving, or maintaining internal knowledge
Information summarization: Quickly organize document highlights or meeting notes
Proposal writing: Assist with ideation and drafting proposals, etc.
Please select the corresponding workflow module based on business needs to maximize the AI’s effectiveness in specific tasks.
Use case: Copywriting assistant
Suppose you are a marketer at a food company writing promotional copy for a newly launched healthy snack.
First, in “Workflow”, choose “Writing Assistant”.
Next, in the AI Q&A interface, submit your writing request. The AI will guide you step by step to fill in key information needed for the copy, such as:
Copy theme
Writing style
Target audience
Word count
Based on your settings, the AI will generate copy options that fit the scenario and communication needs, helping you ideate quickly and improve writing efficiency.
Applicable scenarios
In daily business operations, employees often need to compile data, analyze it, and create reports and analyses in response to routine needs or tasks assigned by managers. Common queries include:
“Which product had the highest sales last month?”
“List all customers with sales over 100,000.”
“What is the revenue trend over the past three months?”
If handled by non-technical staff, such questions often require help from the data team to write SQL queries, making the process time-consuming and inefficient.
Now, with MaiAgent’s Agent mode, the system can automatically convert natural language questions into SQL, query the database in real time, and quickly return analysis results.
This feature is especially suitable for scenarios requiring real-time queries and data insights such as report analysis, KPI tracking, and data lookups. Non-technical users can easily access data, enabling a more intuitive and efficient data-driven decision process.
Operation flow
Go to the “Answer Mode Settings” tab and select “Agent” as the reply mode
Choose database URL
MaiAgent
If you haven’t created a database, you can upload an Excel file directly. The system will automatically convert it into a queryable database format.
Note: If no data is uploaded, the Text-to-SQL Q&A feature cannot be used.
How to operate: Upload Excel or CSV to the knowledge base, and the system will automatically create a database.
Table format notes
Note: If you switch from an existing AI Q&A mode (e.g., from “General” to Text-to-SQL mode)be sure to reupload the Excel or CSV files to be converted into a database. The system will build a queryable database based on the newly uploaded files to support SQL queries.
MySQL
Connect to an existing MySQL database and paste the MySQL connection string
Note: Ensure the database URL can be accessed by the MaiAgent service
PostgreSQL
Connect to an existing PostgreSQL database and paste the PostgreSQL connection string
Note: Ensure the database URL can be accessed by the MaiAgent service
Use case: E-commerce product sales data query
Suppose you are a marketer for an e-commerce platform and want to quickly query product sales and other data.
First, in the “Answer Mode Settings” tab, select “Agent” as the reply mode
Next, you can choose to upload an Excel file to the knowledge base, and the system will automatically convert it into a queryable database format.
You can also ask your company’s technical staff to provide a MySQL or PostgreSQL connection string directly.
Here we assume you have obtained a PostgreSQL connection string. In the database URL dropdown, select PostgreSQL, paste the connection string, and click Save.
After setup, go to the AI Q&A interface and enter a question, for example:
“What are the top three items by revenue on the official website, excluding shipping?”
6. Pre-assign assistant permissions
Permission management architecture and explanation based on RABC
Select the “Permissions” tab, you can configure which members are pre-assigned access to the assistant. The default is to select all roles. You can modify settings based on your use case.
7. Finish creating your AI
After completing the above steps, click the blue “Confirm” button at the bottom right of the dialog to finish the setup 🎉.
