Connecty AI System Overview

Overview

Connecty AI as an end-to-end agentic layer that sits between users, enterprise data warehouses, and downstream data consumption tools. It securely connects to client data platforms (e.g., Snowflake, BigQuery, Databricks), ingesting metadata and query results while enforcing role-based access control. At its core, Connecty builds an autonomous semantic graph ('context graph') that models datasets, schemas, metrics, dimensions, relations, and business logic, continuously kept in sync via a context engine that handles schema evolution, mapping, versioning, and reconciliation. On top of this semantic foundation, agentic AI components enable natural language interaction, multi-agent reasoning, and intent understanding to support analytics use cases such as analysis, discovery, metric exploration, and SQL authoring, as well as decision-oriented workflows through decision trees and goal-driven AI. Outputs are delivered through Connecty’s UI or integrated into APIs and common data consumption tools, with human-in-the-loop controls for inference validation and explainability, all governed by centralized security and governance.

Autonomy vs. Control by Design

Connecty AI is architected to operate across a full spectrum—from fully autonomous execution to fine-grained, human-controlled workflows—without changing the underlying system. In autonomous mode, the platform automatically discovers metadata, constructs the semantic graph, infers metrics, generates queries, and delivers insights or decisions end-to-end with minimal configuration. In parallel, every layer of the system exposes granular controls, allowing teams to explicitly define semantics, constrain agent behavior, approve actions, scope automation, and insert human validation at specific points. This design enables teams to start with fast, low-friction onboarding and progressively increase control and governance as requirements mature.

1. User Interaction (Not Just Questions)

Users can interact with Connecty AI in multiple ways: by asking natural language questions, by allowing the system to learn their data and business context autonomously, or by simply reviewing and validating inferred goals, KPIs, and decisions. Interaction ranges from active exploration to passive oversight, where users primarily see summaries, validations, and outcomes rather than raw analysis.

2. Learning Client Data & Context

Connecty AI connects to the client’s data warehouse and autonomously learns schemas, metrics, relationships, and historical patterns. Over time, it builds contextual understanding of business logic, metric usage, and organizational intent. Users can let this learning run automatically, or intervene to refine definitions, approve inferred metrics, or lock critical KPIs.

3. Agentic AI Operation

Rather than acting as a single chatbot, Connecty AI operates as a coordinated set of agents that continuously reason over data and context. Agents interpret intent, analyze changes, track metrics, and evaluate decisions. These agents can operate fully autonomously or be constrained to advisory mode with explicit human verification.

4. Context & Semantic Foundation

All reasoning is grounded in an autonomous semantic graph that represents datasets, metrics, dimensions, and business concepts. This context is continuously updated as schemas evolve or new data appears, ensuring the system stays aligned with the underlying warehouse while preserving semantic consistency.

5. Goals, KPIs & Decisions

Users can define or approve business goals and KPIs, after which Connecty AI continuously monitors performance, reasons about deviations, and evaluates outcomes. Decision logic is represented as dynamic decision trees, allowing the system to recommend actions, trigger workflows, or escalate issues based on real-time data.

6. Human Verification, Not Micromanagement

Human-in-the-loop is applied where it matters most: validating goals, approving critical metrics, and reviewing high-impact decisions. Users can verify, override, or refine system behavior without managing individual queries or analyses. Explainability is built-in, showing why a conclusion or action was produced.

7. Automated Delivery & Consumption

Insights and decisions are delivered proactively, not just on request. Users receive automated emails, inbox summaries, alerts, or embedded outputs in internal tools. Connecty AI can operate continuously in the background, surfacing only what requires attention or confirmation.