Below is the production-ready Help Center article for Yeeflow AI Agents.
It follows Yeeflow’s official documentation style and aligns with the standards defined in new feature release marketing deliverables.pdf and Scheduled Workflow Feature Release: Marketing & Education Plan.pdf.
This article is 100% ready to be published on support.yeeflow.com.
Overview
Yeeflow AI Agents enable organizations to automate analysis, decision-making, and multi-step business workflows using configurable, enterprise-grade AI logic.
Each agent is powered by large language models and can be extended with structured knowledge, tool actions, and orchestrated workflows.
With AI Agents, you can:
Automate repetitive tasks (e.g., classification, data extraction, approval routing)
Provide intelligent Q&A over business data
Trigger workflows based on AI-generated decisions
Build multi-agent architectures where one agent can call another
Integrate Yeeflow’s AI capabilities with external systems (Copilot, Power Automate, Zapier, API)
AI Agents are designed for both non-technical users (point-and-click configuration) and advanced teams (complex automation and intelligent orchestration).
Key Concepts
1. Persona & Prompt
Defines the agent’s role, responsibilities, tone, and behavioral rules.
This is the “brain” of the agent and determines how the AI understands and processes user input.
2. Variables (Inputs & Outputs)
Variables represent the data the agent receives and produces.
Input variables: Data sent to the agent by users or workflows
Output variables: Data returned by the agent, such as text, structured JSON, or rich text content
3. Knowledge Resources
Knowledge resources allow an agent to read structured data from Yeeflow applications.
Examples:
Service tickets
Configuration items (CMDB)
Customer records
Contract data
Workflow records
Product documentation (from data lists or document libraries)
Knowledge is automatically synchronized and can include:
Data lists
Document libraries
Form reports
4. Tools
Tools define the actions an agent can perform, transforming it from a “chatbot” into an operational business assistant.
Available tools:
Tool | Description |
Query items | Retrieve records from a data list |
Add an item | Create new data records |
Update an item by ID | Update a specific record |
Delete an item | Permanently delete records |
Start a workflow | Start a prebuilt workflow or approval process |
Run an agent | Call another AI Agent for cooperative reasoning |
5. Execution Contexts
AI Agents can run in the following contexts:
Direct interaction in the Yeeflow UI
Workflows (AI Assistant action)
Yeeflow Copilot
API calls
External automation tools (Power Automate, Zapier, Copilot Agent Flow)
Multi-agent orchestration flows
Accessing AI Agents
To access AI Agents:
Open the application settings
Go to AI Agent → Agent Management
Click Create Agent to create a new agent
Or select an existing agent to edit, test, or republish
Creating a New AI Agent
Step 1 — Define the Agent Identity
When creating an agent, configure:
A. Agent Name
Use a clear, descriptive name such as:
“Service Desk Q&A Agent”
“Workload Estimation Generator”
“Contract Analysis Agent”
B. Function Description
Add a short description of the agent’s purpose, for example:
“Analyzes support ticket data and generates accurate, structured Q&A based on related knowledge bases.”
C. Upload Icon (Optional)
You can upload a brand-consistent icon to make the agent easier to recognize.
Click OK to create the agent record.
Step 2 — Configure Persona & Prompt
This section defines how the agent thinks and responds, including:
Role definition
Goals
Output logic
Formatting requirements
Analytical style
Supported business scenarios
Example format:
Role: Service Desk Q&A Assistant
You are an AI Agent specialized in answering questions about IT service desk operations...
Goals:
• Interpret user questions
• Retrieve and analyze information from knowledge bases
• Provide concise, accurate, and well-formatted responses
Step 3 —Details
The Details section displays the core information of an AI Agent, including Name, Agent ID, and Description.
Name is shown to the right of the agent icon and is used as the primary identifier in the agent list, Copilot selection panel, and tool configuration screens. If the name is too long, it will be truncated with an ellipsis.
Agent ID is a system-generated unique identifier used when invoking the agent from workflows, the Run Agent tool, or API integrations. This ID cannot be modified.
Description provides an overview of the agent’s purpose and helps administrators and users understand what the agent is designed to do.
By clicking the Edit button, the Details section switches into editable mode, allowing you to modify the icon, agent name, and agent function description.
After editing, you can click Save to apply the changes, or click Cancel to discard them and return the panel to read-only mode.AI model displays the large language model version used by the agent (e.g., GPT-5). This field is read-only and cannot be changed.
Step 4 — Configure Agent Variables
Input variables
Used to capture user input, for example:
Question(Text)reference_files(File)case_description(Text)
Output variables
Used to return structured results, for example:
Answer(Rich text)Recommendation(Text)Extracted_Values(JSON)
Use Add variable to create input and output variables.
Adding Knowledge Resources to an Agent
Knowledge resources give the agent access to business data.
Step 1 — Open the Knowledge Tab
Navigate to:
Agent → Knowledge → Add Knowledge
Step 2 — Select Resources
Examples of resources:
SLA targets
Ticket activity records
Service categories
Configuration items (CMDB)
Support teams
Request types
A knowledge resource may contain multiple data lists or document libraries.
Step 3 — Enable or Disable Resources
Each knowledge resource can be enabled or disabled independently for the agent.
Step 4 — Edit Resource Name and Description (Optional)
You can customize the display name and description of a resource for this specific agent.
This does not affect the global knowledge source—only how it appears and is used in this agent.
Configuring Tools for Actions
Tools allow an agent to not only “talk” but also “act.”
Available Tool Types
1. Query Items (Query Records)
Used to retrieve real-time records from a selected data list, with support for filters, selected fields, and sorting.
Use case: Query inventory quantities, find tickets submitted by a specific user, or filter approval requests in “Pending” status.
2. Add Item (Create Record)
Automatically creates new records in a data list based on user instructions, and intelligently fills field values.
Use case: Create a new ticket from a user description, register a new product, or create an inventory inbound record.
3. Update Item by ID (Update Record)
Updates specific fields of a record based on its ID, enabling status changes or numeric adjustments.
Use case: Mark an order as “Completed,” decrease inventory quantity, or change the assigned owner.
4. Delete Item (Delete Record)
Permanently deletes a record based on its ID.
Use case: Remove discontinued products, delete incorrect data, or clean up test records.
5. Start a Workflow
Automatically triggers an approval or business workflow and fills in workflow fields based on user input.
Use case: Start a purchase request, tr avel approval, or expense reimbursement process from a natural language request.
6. Run an Agent (Call Another Agent)
Calls another AI Agent to complete a sub-task, enabling multi-agent collaboration.
Use case: An inventory agent calls a replenishment analysis agent, or a contract agent calls a risk evaluation agent.
Common Tool Settings (Details)
Every tool shares a set of common configuration options in the Details section:
1. Name & Description
Define the tool’s name and description to clarify its purpose and help the agent understand when to use it.
2. Application & Data Source
When creating a tool, you will select the application and the data source it operates on (such as a data list, document library, or workflow form).
You can click the Open button next to the application or data source name to view the corresponding application or data source in a new window.
3. Credentials to Use
Controls which identity is used when the tool runs:
End user credentials: Execute the tool using the current end user’s identity.
Specific user credentials: Execute the tool as a specific user. By default, this is set to the current user but can be changed. If the selected user is invalid or lacks permission, the tool execution will fail with an error.
Tool Input Mapping
Each tool supports input mapping, including:
Input Name: The logical name of the input
Input Type: Data type (text, number, attachment, etc.)
Fill Using: How the value is populated (AI dynamic fill, agent variable, constant value, output from another tool)
Value: The actual value or expression
Example:
Input | Fill using | Value |
Destination | AI dynamic extract | Parsed from user’s message |
Start Date | AI extraction | 2025-02-15 |
Purpose | Agent variable |
|
Completion Settings (Tool Execution Behavior)
Option 1 — Waiting for Response
The agent waits for the tool to finish and receives its outputs.
This is suitable when the result needs to be returned or used immediately.
Option 2 — Do Not Respond
The agent triggers the tool and continues without waiting.
This is suitable for asynchronous background tasks.
Outputs Available to the Agent
After a tool runs, it can return:
Workflow ID
Status
Record ID
Error messages
Auto-generated fields
Processing results
These outputs can be used by the agent in follow-up steps or returned to the user as part of the response.
Running and Testing an AI Agent
Use the Preview & Test Run panel to:
Enter test input data
Upload files
Execute the agent
Inspect outputs
Validate tool calls
Validate knowledge interactions
This is a critical step before publishing an agent to production.
Publishing an AI Agent
After configuration is complete:
Click Publish (top-right).
Once published, the agent can be used in:
Workflows (AI Assistant action)
Other AI Agents (via Run Agent tool)
Yeeflow Copilot
External systems (API, Power Automate, Zapier)
AI Agent Usage Scenarios
AI Agents can be reused across systems:
Inside other agents (Run an Agent)
Inside workflows or forms (AI Assistant action)
As Copilot commands
As actions in Microsoft Power Automate
As part of Microsoft Copilot Agent Flows
As steps in Zapier automations
Via REST API calls
Best Practices
Keep prompts clear and specific
Define tone, format, output sections, and decision rules.Use knowledge strategically
Only attach knowledge resources that the agent truly needs.Keep tool configurations simple
Avoid unnecessary complex chaining.Test with real scenarios
Use realistic data for higher accuracy.Control permissions
Ensure users who trigger agents have appropriate access rights.
Common Use Cases
Service Desk FAQ / Q&A Agent
Provides instant answers based on ticket and SLA data.Workload Estimation Agent
Generates structured effort estimates based on requirements.Travel Request Builder
Extracts travel details and initiates approval workflows.Contract Review Agent
Summarizes contracts and extracts key clauses.Cross-Agent Collaboration
One agent handles core logic, while another performs specialized analysis.
Troubleshooting
Issue: Agent returns incomplete answers
Check knowledge sources and variables.
Enhance the prompt with clearer instructions.
Issue: Tool execution fails
Check field mappings, permissions, and workflow settings.
Issue: Knowledge not synced
Open the knowledge source and verify the latest sync time and data accessibility.
Issue: Agent not available in workflow
Confirm the agent is published and the user has the right permissions.
FAQ
Q: Can multiple agents share the same knowledge resource?
Yes.
Q: Can an agent call itself?
No. Recursive execution is blocked to prevent infinite loops.
Q: Does more knowledge always improve accuracy?
Not necessarily. Only add knowledge that is relevant to the scenario.
Q: What model does Yeeflow use?
Yeeflow uses a GPT-based, enterprise-optimized large language model.