Basedash AI Sources

Product Introduction

1. Definition: Basedash AI Sources is a provenance and explainability layer for AI-generated business intelligence (BI) and data analysis. It is a technical feature within the Basedash platform that automatically documents and surfaces the complete data lineage behind every answer provided by its AI assistant.
2. Core Value Proposition: It exists to solve the "black box" problem in AI analytics by providing immediate, granular transparency. Its primary value is enabling trusted AI data analysis through auditable AI workflows, allowing users to verify conclusions by inspecting the exact source data, context, and SQL queries used.

Main Features

1. Unified Source Context Panel: This feature aggregates all inputs the AI used into a single, collapsible panel beneath each answer. It works by tagging and categorizing every piece of context accessed during an AI session. This includes live database tables, semantic layer definitions (like dbt models or BI tool metrics), saved chart references, data from read-only Model Context Protocol (MCP) connections, and web pages crawled for research. Each source is displayed as a contextual chip, providing a high-level map of the analysis's data foundation before deep inspection.
2. Query-Level SQL & Result Inspection: For every data operation performed, AI Sources logs the executed query. Users can expand any query to view the exact, generated SQL syntax and a preview of the returned data rows. This technical deep dive allows for validation of join logic, filter periods, aggregation methods, and raw result sets, turning abstract answers into debuggable, data-driven processes.
3. Intelligent Activity Stream Collapsing: The system differentiates between analytical "thinking" steps and consequential actions. During processing, the chat interface shows real-time steps. Upon completion, all non-mutative analysis (data reads, context checks) collapses into a concise "Analyzed for [time period]" summary. This UI/UX innovation reduces visual noise, keeping the final answer prominent while preserving access to the full audit trail.
4. Explicit Mutation & Action Cards: When the AI performs a state-changing operation—such as running a mutating SQL query, creating a new data definition, or updating system context—these actions are not buried in the provenance log. Instead, they are surfaced as distinct, clear cards below the answer. This design ensures data governance and change management are front-and-center, making any modification to the data environment immediately obvious and reviewable.

Problems Solved

1. Pain Point: The lack of transparency in AI-driven data analysis, often called the "hallucination" or trust gap, where users cannot verify how an AI arrived at a specific number or insight. This leads to wasted time manually replicating analysis or making decisions based on unverified outputs.
2. Target Audience: Data Analysts and BI Professionals who need to trust and explain their reports; Product and Operations Managers who rely on data for decisions but lack SQL expertise; Data Engineers and Architects responsible for governance and ensuring correct data usage; Executives and Finance Leaders who require confidence in the metrics driving strategy.
3. Use Cases: Auditing and Compliance: Providing a verifiable trail for how regulated metrics were calculated. Onboarding and Training: New team members can deconstruct complex analyses to understand company data models. Debugging Discrepancies: Quickly identifying if a surprising KPI result stems from a data anomaly, a flawed query, or an incorrect assumption. Collaborative Analysis: Sharing an AI-generated insight along with its full source pedigree for team review and validation.

Unique Advantages

1. Differentiation: Unlike generic AI tools that may provide a list of source documents or a final SQL query, Basedash AI Sources provides a holistic, step-by-step reconstruction of the entire analytical workflow. It integrates this natively into the conversational UI, unlike traditional BI tools where SQL history and data lineage are separate, complex modules. It contrasts with "chat-to-SQL" tools by focusing on the why and how behind the SQL, not just the query output.
2. Key Innovation: The product's core innovation is its context-aware provenance engine. It doesn't just log queries; it intelligently links each step of the AI's reasoning to the specific business context (definitions, charts, sources) that was active at that moment. The separation of "evidence" (collapsible sources) from "actions" (prominent cards) is a novel UX paradigm that aligns with how professionals mentally categorize read-only analysis versus state-changing operations.

Frequently Asked Questions (FAQ)

1. How does Basedash AI Sources ensure data accuracy and prevent AI hallucinations? Basedash AI Sources mitigates hallucinations by forcing the AI to ground all answers in retrieved data and by making the retrieval process fully transparent. Users can cross-reference every claim in the answer against the previewed source rows and executed SQL, allowing for immediate factual verification and identification of any logical leaps made by the model.
2. Can I use AI Sources for compliance and audit reporting? Yes, AI Sources functions as a detailed audit log for AI-driven data analysis. It provides a timestamped, immutable record of the specific data sources accessed, the exact queries run, and the results returned for any given insight, which can be critical for meeting internal governance or external regulatory compliance requirements.
3. What types of data sources can be tracked by the AI Sources feature? The feature tracks a wide range of structured and semi-structured sources, including connected SQL databases (PostgreSQL, MySQL, etc.), data warehouses (Snowflake, BigQuery, Redshift), defined metrics and models in the semantic layer, saved Basedash charts, web pages via its research function, and any data pulled through read-only MCP (Model Context Protocol) connections to other tools.
4. Does the AI Sources feature impact the performance of my queries or the Basedash AI assistant? The provenance logging is designed as a low-overhead metadata collection layer. It records the queries and context as they are executed for the primary analysis, meaning it does not re-run queries or significantly slow down response times. The performance impact is minimal, focused on logging and UI rendering rather than computational overhead.