Skip to main content

Creating and Managing a Copilot

This article explains how application administrators can create, configure, publish, test, deploy, and manage a Copilot in Yeeflow.

Updated in the last 15 minutes

What Is a Copilot

A Copilot is an application-level AI assistant in Yeeflow.

It helps users work more efficiently by answering questions, guiding processes, and performing actions based on configured instructions, knowledge, and tools.

Unlike standalone AI Agents, a Copilot is embedded directly into applications, making AI assistance available within users’ daily workflows.

AI Agent vs. Copilot: Differences and Relationship

In Yeeflow, AI Agents and Copilots are both core AI components, but they serve different purposes and operate at different layers.

What Is an AI Agent

An AI Agent is an independently runnable intelligent execution unit. It is typically used to:

  • Perform specific tasks such as data processing, decision-making, or automation

  • Be triggered by workflows or other AI components

  • Run in the background without direct interaction with end users

AI Agents function primarily as capability modules or automated executors.

What Is a Copilot

A Copilot is an application-level AI assistant that:

  • Is designed for end users

  • Is embedded directly within application interfaces

  • Interacts with users through natural language conversations

A Copilot does not usually perform complex tasks by itself. Instead, it interprets user intent and orchestrates instructions, knowledge, tools, and AI Agents to fulfill requests.

Relationship Between AI Agent and Copilot

  • A Copilot can invoke one or more AI Agents

  • AI Agents provide execution capabilities for Copilots

  • Copilot = user-facing interaction and decision layer

  • AI Agent = system-facing execution layer

In simple terms:

AI Agents do the work, while Copilot understands the user, decides what to do, and coordinates how the work gets done.

Copilot Lifecycle Overview

A Copilot follows a clear lifecycle in Yeeflow:

  1. Create a Copilot

  2. Configure its behavior and capabilities

  3. Publish the Copilot

  4. Test the Copilot (Test Run)

  5. Deploy the Copilot to applications

  6. Manage and update it as business needs change

Understanding this lifecycle helps ensure your Copilot is visible, reliable, and usable for end users.

Creating a Copilot

Only application administrators (or users with sufficient permissions) can create a Copilot.

Steps to Create a Copilot：

  1. Go to Application Management

  2. Open the AI Agent section

  3. Select Copilot Management

  4. Click Create Copilot

Basic Copilot Information

When creating a Copilot, you must provide:

  • Icon – Used to visually identify the Copilot in applications

  • Name – Displayed to users during interaction

  • Description – Describes the Copilot’s purpose (administrator-facing)

Click OK to create the Copilot and enter the configuration screen.

Configuring a Copilot

After creation, you will be taken to the Copilot configuration interface.
The configuration is organized into the following sections.

Overview

The Overview tab provides a high-level summary of the Copilot configuration, including:

  • Detail – Copilot identity (icon, name, description)

  • Instructions – Behavior and response guidelines

  • Knowledge – Enabled knowledge sources

  • Tools – Enabled action tools

This page helps administrators quickly understand what the Copilot can do.

Detail

The Detail section defines the Copilot’s identity:

  • Icon

  • Name

  • Description

These fields help both administrators and users recognize the Copilot’s role and purpose.By default, it is in read-only mode. Click Edit to modify the details.

Instructions

Instructions define how the Copilot behaves, including:

  • The role the Copilot plays

  • How it responds to users

  • What it should or should not do

Instructions are critical to Copilot behavior.

Knowledge

Knowledge provides background information that the Copilot can use when responding to users.

In this section, you can:

  • Add knowledge resources

  • Enable or disable specific knowledge sources

  • Customize knowledge names and descriptions for better AI understanding

Well-configured knowledge significantly improves response accuracy and contextual relevance.

Web Search

In addition to internal knowledge, Copilot supports Web Search.

When Web Search is enabled:

  • Copilot can retrieve up-to-date public information from the internet

  • It is useful for policies, industry trends, or general knowledge not covered by internal resources

  • Web Search is typically used as a supplement to internal Knowledge

Web Search is recommended when real-time or external information is required.

Tools

Tools allow Copilot to perform actions, not just answer questions.

In the Tools section, administrators can:

  • Add or remove tools

  • Enable or disable tools

  • Configure post-execution behavior and responses

Knowledge Tab

While the Overview tab shows a summary, the Knowledge tab is used for detailed knowledge management.

In the Knowledge tab, administrators can:

  • Add, edit, or remove knowledge resources

  • Enable or disable individual knowledge sources

  • Customize knowledge names and descriptions to improve AI understanding. This does not affect the global knowledge source—only how it appears and is used in this agent.

Changes made here directly affect how Copilot retrieves background information.This article provides only an overview. For detailed guidance, see Managing Knowledge for Copilot

Tools Tab

The Tools tab provides full control over Copilot’s action capabilities.

Six Core Tools Supported by Copilot：

1. Query Items

Used to retrieve data records from applications.

  • Supports filtering and field selection

  • Commonly used for lookup and reporting scenarios

2. Create Item

Used to create new data records.

  • Copilot can populate fields automatically based on conversation

  • Suitable for form creation and quick data entry

3. Update Item

Used to update existing records.

  • Records can be identified by conditions

  • Supports partial field updates

4. Start Workflow

Used to start a specific workflow.

  • Copilot can trigger workflows based on user intent or context

  • Commonly used for approval and request processes

5. Run AI Agent

Used to invoke other AI Agents.

  • Complex or specialized tasks can be delegated to AI Agents

  • Execution results can be returned to Copilot

6. Respond to User

Used to control how Copilot responds after tool execution.

  • Return fixed messages

  • Return execution results

  • Guide users to next actions

By combining tools effectively, a Copilot can evolve from a conversational assistant into an actionable intelligent work assistant.

Publishing a Copilot

Creating and configuring a Copilot does not automatically make it available.

Publishing Behavior

  • Publish – Publishes the Copilot and makes it available for testing and deployment

  • Published – Indicates that the Copilot has already been published and is ready for testing or deployment

Only published Copilots can proceed to testing and deployment.

Test Run (Testing a Copilot)

After publishing and before deployment, administrators can use Test Run to validate the Copilot.

Purpose of Test Run

Test Run allows administrators to:

  • Validate overall conversational behavior

  • Verify that Instructions are applied correctly

  • Confirm Knowledge and Web Search usage

  • Ensure Tools are triggered and executed as expected

Test Run is available only to administrators and does not affect live applications or user data.

Using Test Run

  1. Ensure the Copilot is published

  2. Click Test Run

  3. Interact with the Copilot in the test window

  4. Review responses and tool execution results

It is recommended to run Test Run after every configuration change.

Deploying Copilot to Applications

Once published and tested, the Copilot can be deployed to applications.

Deployment Steps：

  1. Open the Channels section

  2. Click Deploy to applications

  3. Select the workspace and application

  4. Configure access permissions

  5. Confirm deployment

Deployed Applications

After deployment, you can view:

  • Which applications use the Copilot

  • Deployment status

  • Who deployed it and when

Deployments can be removed if the Copilot is no longer needed.

Managing Copilot Availability

A Copilot may not appear to users if:

  • It is not published

  • It is published but not deployed

  • The user lacks permission

  • The Copilot is disabled in the application

Checking publication status and deployment configuration usually resolves visibility issues.

Common Management Scenarios

Copilot Is Published but Not Visible

  • Verify it is deployed to the application

  • Check access permissions

  • Ensure the Copilot is enabled

Copilot Behavior Needs Adjustment

  • Update Instructions

  • Update Knowledge or Web Search

  • Adjust Tools configuration

Related Articles
Version Management for Applications
AI Agent
Knowledge Sources
AI Agent Integration Guide
Using Copilot in an Application
Did this answer your question?