Copilot Chat

The chat interface is the primary way to interact with BalkanID Copilot's AI-powered access governance assistant. This comprehensive guide covers all aspects of the chat experience, from basic messaging to advanced features like query transparency and tool calls.

Overview

The chat interface provides a sophisticated conversational experience that goes far beyond simple text messaging. Through real-time AI conversations, you can engage in interactive discussions with the AI assistant about your access governance needs. The interface offers complete query transparency, allowing you to see the actual Cypher queries that power your results, ensuring you understand exactly how your data is being accessed and analyzed.

The system includes powerful tool calls that enable AI-powered actions and data retrieval, smart entity mentions with autocomplete for users, groups, and resources, and a comprehensive feedback system to rate and improve AI responses. You can choose from different AI models depending on your needs and access your complete message history from previous conversations.

Interface Components

Chat Input Area

Located at the bottom of the screen, the chat input area is where you compose and send your questions to the AI assistant. The text input field accepts natural language queries, allowing you to ask questions just as you would to a human colleague. The model selector, positioned at the bottom-right of the input area, lets you choose which AI model to use for your query - different models may provide varying response styles and capabilities based on how your administrator has configured them.

The send button, marked with a "Send" icon, submits your query to the AI. The input area also supports entity mentions using the @ symbol, which triggers autocomplete functionality to help you reference specific users, groups, or resources in your organization.

Message Types

The chat interface displays several distinct types of messages (or dialogues), each serving a specific purpose in your conversation with the AI assistant.

User Messages appear with your user icon on the left and display your text with any highlighted entity mentions. When you reference specific entities in your messages, these appear with annotations showing the recognized items, such as (User: john.doe) or (Group: administrators), confirming that the AI has correctly understood your references.

Assistant Responses feature the bot icon on the left and contain natural language explanations in response to your queries. These responses include well-formatted content with headers, lists, and emphasis to make information easy to scan and understand. The AI often embeds interactive data tables directly in responses and includes code blocks when providing technical details or examples.

Tool Call Messages appear when the AI needs to query your data systems to answer your questions. These messages show query execution progress, display data tables with results, include detailed query information and metadata, and provide insight badges highlighting important findings in the data.

Alternate Tool Calls represent secondary or follow-up queries that the AI performs to gather additional context or verify information. For external integrations:

• Shows tool execution status

• Displays tool arguments and results

• Indicates success or error states

Chat Features

Entity Mentions and Autocompletion

Using @ Mentions:

1. Type @ in your message

2. See suggestions for users, applications, roles, and other entities

3. Select entities to include them precisely in your query

4. Mentions are highlighted to show entity type and name

Query Understanding and Execution

The AI assistant employs sophisticated natural language processing to understand and respond to a wide variety of question types. Whether you're asking informational questions like "What is the current access policy for the finance group?" or analytical questions such as "Which users have administrative privileges?", the system can interpret your intent and provide accurate responses.

When you submit a question, the AI works behind the scenes through a sophisticated process. It first parses your natural language question to understand the intent and identify key entities. Then it generates appropriate database queries, typically in Cypher graph query language, which are executed against your identity and access management data. Finally, the results are formatted and presented in an easy-to-understand format, complete with explanations and actionable insights.

Data queried is NOT sent to the AI, and only displayed to the user querying the data. The AI only receives table metadata (like table headings) to generate a summary with. This may cause the AI to be unable to reference specific data cells in the table visible.

Data Tables and Results

The interactive tables that appear in chat responses provide powerful ways to explore and work with your data. These tables include pagination controls to help you navigate through large result sets, sortable columns that respond to clicks on column headers, and built-in filtering capabilities through search functionality.

You can manage which columns are visible and expand tables to full-screen view for better visibility when working with complex data sets. Each table also provides practical action options including the ability to download data as CSV files for external analysis and copy functionality for individual cells or entire tables to use in other applications.

• Query Information: View the queries that generated the data

Insight Badges

Tables may include colored badges indicating:

• Security Findings: Potential security issues

• Insights: Notable patterns or anomalies

• Compliance Issues: Policy violations or risks

Badge Interactions:

• Click badges to get detailed explanations

• Generate playbooks from insight descriptions

• Get AI-powered analysis of findings

Query Transparency

One of BalkanID Copilot's most powerful features is complete query transparency. Every data table displayed in the chat interface provides full access to the underlying queries that generated the results. This transparency ensures you understand exactly how your data is being accessed and can verify that the results match your expectations.

For each table, you can access your natural language question, the AI's initial query interpretation, the actual query executed against your data, and an AI-generated explanation of what the query accomplishes. The query dialog includes practical features like the ability to copy query text to your clipboard, understand differences between generated and executed queries, download complete query documentation packages, and request detailed AI explanations of query logic.

The platform sometimes enhances AI-generated queries to include additional security-relevant data, optimize performance for your specific environment, ensure comprehensive results that cover all relevant aspects, or add compliance-related information required by your organization's policies.

Interaction Patterns

Follow-up Questions and Contextual Suggestions

The AI assistant provides intelligent suggestions to help you dive deeper into your access governance analysis. After each response, you'll often see drill-down questions that let you explore specific aspects of the results in greater detail, related queries that address similar or complementary questions, and use case suggestions that represent common next steps for your particular scenario.

The system excels at building on previous context within your conversation. You can reference previous results naturally by saying things like "Show me more details about those users," combine insights across different queries with requests like "Cross-reference this with application access," or iterate and refine your analysis by asking to "Filter those results by last login date." This contextual awareness makes each conversation increasingly valuable as you explore related topics.

Use Cases and Templates

The homepage presents suggested starting points to help you begin your analysis. These predefined use cases cover common scenarios like security assessments, access reviews, compliance audits, and user lifecycle management tasks. Template questions are also available to click, which help you learn effective query patterns, explore the platform's full capabilities, and get started quickly with common analytical tasks.

Annotation and Entity Recognition

The system automatically recognizes and annotates various entity types within your conversations, including individual user accounts, business applications and systems, user roles and permission groups, organizational departments and units, and digital resources like files and databases.

When entities are recognized, they appear in a standardized annotation format like (EntityType: EntityName), which provides clear context for AI processing, ensures accurate query generation, and enables precise data retrieval. This annotation system helps both you and the AI maintain clarity about exactly which entities are being discussed throughout your conversation.

Chat History and navigation

Chat History

The sidebar navigation organizes your conversation history in a chronological structure that makes it easy to find and return to previous discussions. Recent conversations appear under "Today," while "Yesterday" contains the previous day's chats. For broader historical access, "This Week" provides your week's conversation history, and "This Month" serves as a monthly archive of all your interactions.

The platform automatically organizes chats by generating titles based on the content of your conversations, making it easier to identify specific discussions at a glance. A built-in search functionality allows you to quickly locate specific conversations by searching through titles and content. To resume any previous conversation, simply click on the chat entry in the sidebar, and you'll be taken back to that discussion with full context preserved.

New Chat Creation

Starting a fresh conversation is straightforward and offers several pathways to begin your analysis. Click the BalkanID Copilot icon to initiate a new chat session with a clean slate. The interface often presents suggested starting questions to help you get oriented and begin exploring your access governance data effectively. You can also begin with use case templates that provide structured starting points for common analytical scenarios like security assessments or compliance reviews.

Chat Management Options

Managing your conversations effectively helps you maintain organized records of your access governance analysis work. The options menu, accessible through the ⋯ button on chat items, provides essential conversation management capabilities.

You can rename chats to give them meaningful titles that reflect their content, delete conversations that are no longer needed to keep your workspace clean, share conversations through generated links when collaboration features are enabled in your environment, and export conversation history to download complete records of your analysis sessions for documentation or compliance purposes.

Feedback and Improvement

Response Feedback

The platform includes a comprehensive feedback system that helps improve AI performance over time. You can rate responses using thumbs up for helpful responses and thumbs down for unhelpful ones. Your feedback directly contributes to improving the AI's performance and accuracy for future interactions.

Provide thumbs up feedback for responses that are accurate and useful, contain the information you needed, or help you accomplish your access governance tasks effectively. Use thumbs down feedback for incorrect or unhelpful responses, when important information is missing, or when results are unclear.

Model Selection for Different Tasks

Understanding when to use different AI models can significantly improve your experience with the platform. General purpose models work well for most everyday questions and provide balanced performance across various query types. Specialized models may be available for particular use cases or domains within your organization's setup.

Performance models prioritize faster response times and work well for simple queries where speed is more important than detailed analysis. Accuracy models focus on providing more thorough analysis and detailed responses for complex questions that require comprehensive investigation.

Tips for Effective Chat Usage

Query Best Practices

Crafting effective questions significantly improves the quality of responses you'll receive. The most successful queries are specific and actionable, clearly stating what information you're seeking. Examples of effective question formats include "Show me all users who have admin access to the HR application," "Which users haven't logged in for more than 90 days?" and "Find applications with the most privileged users."

Less effective approaches tend to be too vague, overly broad, or don't actually request information. Avoid queries like "Users" without additional context, "Security problems" which covers too much ground to provide actionable results, or "Fix access issues" which requests action rather than information analysis.

Iteration Strategies

The most effective analysis sessions often involve progressive refinement of your questions. Start with broad questions to get an overview, then use the AI's suggestions to narrow your focus, build on previous results to dive deeper, and combine multiple data points for comprehensive analysis.

A typical progression might look like this: begin by asking "Show me all applications" to understand your application landscape, then ask "Which of these have the most users?" to identify high-impact systems. Next, focus on a specific application with "Show me admin users for [specific application]" and finally explore activity patterns with "When did these admin users last access the system?" This iterative approach builds a comprehensive understanding of your access governance landscape.

Troubleshooting Chat Issues

When you encounter issues with the chat interface, several common problems have straightforward solutions. If no results are returned, check your tenant selection to ensure you're looking at the right data set, verify you have access to the requested data, try broader or different query terms, and ensure you're using proper entity mentions with the @ symbol.

For slow response times, try using simpler and more focused questions, check your network connectivity, experiment with different AI models if available, and consider breaking complex queries into smaller, more manageable parts. These strategies typically resolve most performance issues you might encounter.

Unclear or Incorrect Results:

• Provide feedback using thumbs down

• Rephrase questions more specifically

• Check query details in the query dialog

• Use entity mentions for precision

Error Messages:

• Note error details for administrator

• Try alternative query approaches

• Check system status with IT suppor