Relationalai

RelationalAI is a Berkeley, California-based software company that develops a decision intelligence platform delivered as a Snowflake-native application. Its central product, an agent called Rel, analyzes tables and documents already held in a customer's Snowflake account, constructs a semantic model of the business, and answers operational questions grounded in that model. The platform exposes four categories of reasoning: graph reasoning for discovering groups, clusters and connections (with pathfinding in preview), rule-based reasoning for business qualification rules and decision trees, predictive reasoning for demand forecasts, churn likelihood and risk or anomaly alerts (early access), and prescriptive reasoning for pricing, inventory management and staffing optimization (early access).

The company describes its system as learning business logic once and applying it across use cases, fine-tuning large language models inside the customer's data cloud and enriching them with semantic models and relational knowledge graphs. Because it runs on a single relational foundation inside Snowflake, it supports zero-copy cloning, versioning, time travel and consumption pricing without data syncs. Corrections made by users are said to update the system's understanding permanently, a behavior the company labels "superalignment." Developers interact through PyRel, a Python API that sends queries to the RelationalAI Native App to execute within the Snowflake security perimeter; administrators manage app access and monitor usage.

Earlier company materials from 2022 positioned RelationalAI as the creator of a "relational knowledge graph" system for building intelligent data apps, aimed at eliminating the "knowledge silos" created when application logic remains locked in source applications. At that stage the system was offered as a fully hosted cloud service in early access preview, with early enterprise use cases including data fabric, fraud detection, securities analysis, structural search and contextual recommendation applications.