#AI Employee · Admin Configuration Guide

This document helps you quickly understand how to configure and manage AI Employees, guiding you step-by-step through the entire process from model services to task assignment.

#I. Before You Start

#1. System Requirements

Before configuring, please ensure your environment meets the following conditions:

• NocoBase 2.0 or higher is installed
• The AI Employee plugin is enabled
• At least one available Large Language Model (LLM) service (e.g., OpenAI, Claude, DeepSeek, GLM, etc.)

#2. Understanding the Two-Layer Design of AI Employees

AI Employees are divided into two layers: "Role Definition" and "Task Customization".

Layer	Description	Characteristics	Function
Role Definition	The employee's basic personality and core abilities	Stable and unchanging, like a "resume"	Ensures role consistency
Task Customization	Configuration for different business scenarios	Flexible and adjustable	Adapts to specific tasks

To put it simply:

"Role Definition" determines who this employee is, "Task Customization" determines what they are doing right now.

The benefits of this design are:

• The role remains constant, but can handle different scenarios
• Upgrading or replacing tasks does not affect the employee itself
• Background and tasks are independent, making maintenance easier

#II. Configuration Process (in 5 steps)

#Step 1: Configure Model Service

The model service is like the brain of an AI Employee and must be set up first.

💡 For detailed configuration instructions, please refer to: Configure LLM Service

Path: System Settings → AI Employee → LLM Service

Click Add and fill in the following information:

Item	Description	Notes
Provider	e.g., OpenAI, Claude, Gemini, etc.	Compatible with services using the same specification
API Key	The key provided by the service provider	Keep it confidential and change it regularly
Base URL	API Endpoint (optional)	Needs to be modified when using a proxy
Enabled Models	Recommended models / Select models / Manual input	Defines which models are available in chat

After configuration, use Test flight to test the connection. If it fails, please check your network, API key, or model name.

#Step 2: Create an AI Employee

💡 For detailed instructions, please refer to: Create AI Employee

Path: AI Employee Management → Create Employee

Fill in the basic information:

Field	Required	Example
Name	✓	viz, dex, cole
Nickname	✓	Viz, Dex, Cole
Enabled Status	✓	On
Bio	-	"Data Analysis Expert"
Main Prompt	✓	See Prompt Engineering Guide
Welcome Message	-	"Hello, I'm Viz…"

At the employee creation stage, focus on role and skill configuration. The actual model can be selected in chat via Model Switcher.

Prompt Writing Suggestions:

• Clearly state the employee's role, tone, and responsibilities
• Use words like "must" and "never" to emphasize rules
• Include examples whenever possible to avoid abstract descriptions
• Keep it between 500–1000 characters

The clearer the prompt, the more stable the AI's performance. You can refer to the Prompt Engineering Guide.

#Step 3: Configure Skills

Skills determine what an employee "can do".

💡 For detailed instructions, please refer to: Skills

Type	Capability Scope	Example	Risk Level
Frontend	Page interaction	Read block data, fill forms	Low
Data Model	Data query and analysis	Aggregate statistics	Medium
Workflow	Execute business processes	Custom tools	Depends on the workflow
Other	External extensions	Web search, file operations	Varies

Configuration Suggestions:

• 3–5 skills per employee is most appropriate
• It's not recommended to select all skills, as it can cause confusion
• For important operations, prefer Ask over Allow

#Step 4: Configure Knowledge Base (Optional)

If your AI employee needs to remember or reference a large amount of material, such as product manuals, FAQs, etc., you can configure a knowledge base.

💡 For detailed instructions, please refer to:

• AI Knowledge Base Overview
• Vector Database
• Knowledge Base
• RAG (Retrieval-Augmented Generation)

This requires installing the vector database plugin.

Applicable Scenarios:

• To make the AI understand enterprise knowledge
• To support document Q&A and retrieval
• To train domain-specific assistants

#Step 5: Verify the Effect

After completion, you will see the new employee's avatar in the bottom right corner of the page.

Please check each item:

• ✅ Is the icon displayed correctly?
• ✅ Can it conduct a basic conversation?
• ✅ Can skills be called correctly?

If all pass, the configuration is successful 🎉

#III. Task Configuration: Getting the AI to Work

What we've done so far is "creating an employee". Next is to get them "to work".

AI tasks define the employee's behavior on a specific page or block.

💡 For detailed instructions, please refer to: Tasks

#1. Page-level Tasks

Applicable to the entire page scope, such as "Analyze the data on this page".

Configuration Entry: Page Settings → AI Employee → Add Task

Field	Description	Example
Title	Task name	Stage Conversion Analysis
Context	The context of the current page	Leads list page
Default Message	Preset conversation starter	"Please analyze this month's trends"
Default Block	Automatically associate with a collection	leads table
Skills	Available tools	Query data, generate charts

Multi-task Support: A single AI employee can be configured with multiple tasks, which are presented as options for the user to choose from:

Suggestions:

• One task should focus on one goal
• The name should be clear and easy to understand
• Keep the number of tasks within 5–7

#2. Block-level Tasks

Suitable for operating on a specific block, such as "Translate the current form".

Configuration Method:

1. Open the block action configuration
2. Add "AI Employee"

3. Bind the target employee

Comparison	Page-level	Block-level
Data Scope	Entire page	Current block
Granularity	Global analysis	Detailed processing
Typical Use	Trend analysis	Form translation, field extraction

#IV. Best Practices

#1. Configuration Suggestions

Item	Suggestion	Reason
Number of Skills	3–5	High accuracy, fast response
Permission Mode (Ask / Allow)	Prefer Ask for data changes	Prevent accidental operations
Prompt Length	500–1000 characters	Balances speed and quality
Task Goal	Single and clear	Avoids confusing the AI
Workflow	Use after encapsulating complex tasks	Higher success rate

#2. Practical Suggestions

Start small, optimize gradually:

1. First, create basic employees (e.g., Viz, Dex)
2. Enable 1–2 core skills for testing
3. Confirm that tasks can be executed normally
4. Then, gradually expand with more skills and tasks

Continuous optimization process:

1. Get the initial version running
2. Collect user feedback
3. Optimize prompts and task configurations
4. Test and iterate

#V. FAQ

#1. Configuration Stage

Q: What if saving fails? A: Check if all required fields are filled in, especially the model service and prompt.

Q: Which model should I choose?

• Code-related → Claude, GPT-4
• Analysis-related → Claude, DeepSeek
• Cost-sensitive → Qwen, GLM
• Long text → Gemini, Claude

#2. Usage Stage

Q: AI response is too slow?

• Reduce the number of skills
• Optimize the prompt
• Check the model service latency
• Consider changing the model

Q: Task execution is inaccurate?

• The prompt is not clear enough
• Too many skills are causing