AI Copilot Overview
The AI Copilot is Qarion's built-in intelligent assistant. It helps users find data through natural-language questions, automatically generates documentation, and reduces manual effort for data stewards and consumers alike.
What is the AI Copilot?
The AI Copilot is an embedded assistant that understands your organization's data landscape. Instead of manually searching the catalog with filters or writing SQL queries, you can ask questions in plain language — and the Copilot responds with relevant data products, descriptions, and suggestions.
The Copilot serves two primary roles: data discovery for consumers looking for the right data, and documentation automation for stewards managing large catalogs.
Key Capabilities
Natural-Language Search
Ask the Copilot questions like "Where is the customer sales data?" or "Which products contain PII fields?" The Copilot analyzes your question, searches across the catalog metadata, and returns relevant data products with direct links. Each response cites its sources, so you can verify the answer and navigate directly to the relevant products.
Description Generation
For data products with rich schema metadata but little documentation, the Copilot can automatically generate descriptive text. Click Generate Description on any data product, and the AI analyzes the table name, column names, data types, tags, lineage information, product type, and associated metadata to produce a coherent paragraph describing the asset's purpose and contents.
The same capability extends to individual fields (columns). When viewing a product's schema tab, you can generate a description for any field. The AI considers the field's name, data type, the parent table's context, and surrounding column names to produce an accurate, contextual description.
The generation leverages enriched context from multiple dimensions:
- Schema details — Column names, data types, and nullability
- Product classification — Product type, environment, and location
- Existing metadata — Tags, descriptions, and governance assignments
- Lineage — Upstream sources and downstream consumers
Generated descriptions follow a consistent professional tone and can be edited after generation. This feature dramatically speeds up catalog documentation — especially during initial onboarding when hundreds of products need descriptions.
Contextual Assistance
The Copilot is aware of the platform context you are working in. When accessed from within a specific space, it prioritizes results from that space. When viewing a specific data product, it can provide insights about related assets, upstream sources, or downstream consumers.
Using the Copilot
Chat Interface
The Copilot chat is accessible from the platform's main interface. Type your question in natural language, and the Copilot responds with answers and relevant links. The conversation maintains context, so you can ask follow-up questions to refine your search.
Generating Descriptions
Navigate to any data product's detail page. In the description section, click the Generate Description button. The AI analyzes the product's metadata — column names, table identifiers, data types, tags, and lineage information — and produces a draft description. Review the generated text, make any edits, and save.
Generating Field Descriptions
On any product's Schema tab, individual fields also support AI description generation. Click the Generate button next to a field's description to automatically produce a contextual description. The AI uses:
- The field name and data type
- The parent table name and its existing description
- Neighboring column names for additional context
This is especially useful for large tables with dozens of columns where manual documentation is impractical.
For large-scale catalog onboarding, the AI description generation feature can save hours of manual documentation work. Focus your manual effort on adding business context that the AI cannot infer from technical metadata alone.
How It Works
The AI Copilot is powered by large language models (LLMs) — currently Anthropic Claude — integrated via the Model Context Protocol (MCP). When a user sends a message, the Copilot:
- Receives the query with conversation history and platform context
- Selects relevant MCP tools — search, catalog lookup, lineage traversal, quality checks
- Executes tool calls against the Qarion API using the user's authenticated session
- Synthesizes a response combining tool results with the LLM's reasoning
The architecture uses JWT token forwarding so the Copilot operates within the user's permissions — it can only access data the user is authorized to see.
MCP Tools
The Copilot has access to platform tools via the Model Context Protocol:
| Tool | Description |
|---|---|
search | Search products, issues, meetings across the catalog |
get_product | Retrieve detailed product metadata and schema |
get_lineage | Explore upstream/downstream dependencies |
get_quality | Check quality status and recent failures |
Source Attribution
Every Copilot response includes references to the data products and metadata sources it used to build the answer. This transparency helps users verify the accuracy of responses and navigate to the original assets for more detail.
Multi-Source Awareness
The Copilot synthesizes information from across the platform. When answering a question about data availability, it draws on catalog metadata, lineage connections, quality status, and governance assignments to provide a comprehensive picture.
Integration Points
With the Data Catalog
The Copilot's primary data source is the catalog. All product metadata — names, descriptions, tags, schema information, and governance assignments — is searchable through natural language.
With Data Lineage
When you ask about data dependencies or impact, the Copilot leverages lineage data to identify upstream sources and downstream consumers, giving you context that goes beyond simple keyword matching.
With Quality Management
The Copilot can surface quality information in its responses. If you ask about the reliability of a dataset, it can indicate whether quality checks are passing, highlight recent failures, or flag drift warnings.
With MCP Tools
The Copilot uses the Model Context Protocol to access platform APIs as tools. This allows it to perform multi-step reasoning — for example, searching for a product, fetching its schema, then checking its quality status — all within a single conversation turn.
Best Practices
Ask Specific Questions
The more specific your question, the better the results. Instead of "show me all data," ask "which tables contain customer transaction data in the finance space?" Specific queries return focused, actionable results.
Review Generated Descriptions
AI-generated descriptions are a starting point — not a final product. Always review generated text for accuracy, add business-specific context, and ensure it reflects the actual use of the data product within your organization.
Combine with Filters
Use the Copilot alongside traditional catalog filters for the best results. The Copilot excels at understanding intent, while filters excel at precise, structured queries. Together, they cover the full spectrum of data discovery needs.