Upsolve Data Models

Product Introduction

1. Definition: Upsolve Data Models is a semantic modeling and context management platform for AI data agents. It functions as a centralized, version-controlled repository for an organization's data schema, metric definitions, and business vocabulary, designed to ground AI-generated analytics in a single source of truth.
2. Core Value Proposition: It exists to eliminate AI hallucination and context drift in data analytics by programmatically enforcing canonical definitions. The platform ensures every AI agent answer is consistent, accurate, and based on up-to-date data, solving the critical problem of unreliable AI outputs in business intelligence.

Main Features

1. Descriptive Data Model Registry: This feature allows users to formally register database tables and columns, attaching human-readable descriptions, data types, and defining primary and foreign key relationships. It works by creating a machine-readable ontology that the AI agent references before querying, ensuring it understands the data structure and relationships.
2. Versioned System Prompt & Definitions: Business logic, metric formulas (e.g., "ARR = Sum of annualized contract value for active subscriptions"), and output formats are encoded into a version-controlled system prompt. This enables safe iteration, rollback capabilities, and audit trails for changes to business rules, treating context management like code (Git-like versioning).
3. Pre-cached Selectable Values with Nightly Refresh: Users can mark specific columns (e.g., contract_status, payment_terms) as "selectable." Upsolve's engine then pre-computes and caches the distinct values from these columns, refreshing them on a configurable schedule (nightly by default). This ensures the agent has immediate access to current, valid data enumerations without live database lookups for every query.

Problems Solved

1. Pain Point: "Definition Drift" and "Context Hallucination." In organizations, multiple definitions for core metrics (like "revenue" or "active user") exist across documents, Slack, and individual minds. AI agents, relying on inconsistent prompt context or guessing from raw schemas, produce numerically plausible but fundamentally incorrect answers, leading to faulty decision-making.
2. Target Audience: Data teams (Analysts, Engineers, Analytics Engineers) responsible for BI and AI agent accuracy; Product teams embedding analytics features; Business operations and finance teams requiring consistent reporting; Companies implementing internal or customer-facing AI data chatbots.
3. Use Cases: Ensuring a customer-facing AI chat feature in a SaaS product always calculates "Seat Usage" correctly; guaranteeing that an internal Slack bot for sales metrics uses the latest, board-approved "Pipeline Coverage" formula; onboarding new data team members with a living, executable data dictionary; providing deterministic context for multi-agent AI analytics systems.

Unique Advantages

1. Differentiation: Unlike standalone semantic layers (e.g., dbt, LookML) which require significant upfront modeling, Upsolve Data Models allows incremental definition starting from raw warehouse tables. Unlike static, long system prompts in LLM chats, it dynamically combines versioned business logic with a fresh, pre-cached understanding of the underlying data state.
2. Key Innovation: The integration of a versioned semantic layer with scheduled data value caching specifically for AI agents. This creates a "closed-loop" grounding system where the agent's context is both semantically precise (definitions) and temporally accurate (fresh data values), which is a distinct architectural approach compared to prompt-only or RAG-only solutions.

Frequently Asked Questions (FAQ)

1. How does Upsolve Data Models differ from just using a long, detailed system prompt in ChatGPT or Claude? A system prompt contains business practice and instructions but lacks a live connection to the data structure and its current values. Upsolve Data Models provides the AI agent with the actual schema, keys, descriptions, and pre-fetched valid column values, creating a deterministic understanding of the data truth that a static prompt cannot.
2. Do I need an existing tool like dbt or a semantic layer to use Upsolve Data Models? No, it is not a prerequisite. Upsolve is designed to be the starting semantic layer. You can connect it directly to your data warehouse tables and incrementally add table descriptions, keys, and metric definitions, effectively building your governed context layer within the platform.
3. What happens when my underlying data changes—how does the agent stay accurate? Upsolve employs a nightly refresh (or a user-defined schedule) for its cached "selectable values." When a new contract_status is added to the database, it is automatically picked up in the next refresh cycle, ensuring the agent's knowledge of possible values does not drift from reality.
4. Can Upsolve Data Models handle row-level security (RLS) for AI agents? While this specific feature page focuses on the data model, Upsolve's broader platform includes Row-Level Security as a core component (noted in the Launch Month calendar). This ensures that when the agent queries data, it respects the same permission models as your BI tools, providing secure, personalized answers.