Platform Components

Solid Build

For: Semantic engineers, data engineers, and modelers.
Role: The primary authoring and management interface for creating, editing, testing, and publishing semantic models.

Key capabilities:

Connect to your data warehouse; ingest schema and query logs
Review and edit auto-generated semantic models
Run benchmark evaluations and review results
Apply AI-generated fix recommendations
Manage model lifecycle (In Progress → Pending Review → Certified)
Access the company glossary, metrics library, and asset documentation
Monitor MCP usage, including human-initiated vs. agent-initiated queries

Solid Analyze

For: Business users.
Role: Natural-language interface for querying certified semantic models. Business users ask questions in plain English and receive answers backed by the governed semantic layer.

Key capabilities:

Natural language query → SQL generation → results
Grounded in certified semantic models (no hallucination outside model scope)
Users see answers, not necessarily raw SQL
Can be used for ad hoc model validation during model development ("Ask model")

Solid MCP Server

For: AI agents and agent platforms.
Role: Exposes Solid's semantic layer to any MCP-compatible agent. Agents call Solid MCP tools to get grounded SQL, business term definitions, asset metadata, and model context.

Key capabilities:

SQL Generation tool (text-to-SQL)
Glossary tool (business term definitions)
Asset Information tool (metadata about specific assets)
Semantic Model QA tool (questions about model purpose and business questions)

See Getting Started with the Solid MCP Server for the full tool reference.

Solid DB Agent

For: Internal use — not user-facing. This is the security boundary component.
Role: The bridge between Solid's platform and your data warehouse. Executes validation queries during SQL generation, enforces read-only and scoping constraints, and returns pass/fail plus error metadata only.

Key properties:

Detachable — can run wherever you want, next to the database or in your network
Runs as a service principal with read-only permissions scoped to only the tables/schemas exposed to semantic modeling
Applies a LIMIT and a strict, configurable timeout to all validation queries
Uses a communication queue with limited throughput to prevent warehouse overload
Returns only pass/fail status and error information — never raw query results
All query execution is transient; no data is stored or cached

Supporting Subsystems

Component	Role
Model Engine	The AI subsystem that generates initial semantic models and documentation from extracted knowledge, during the offline ingestion/analysis phase.
Automated Validation Suite	A test harness that validates semantic models against real usage data and constraints before human review.
Intake	The data collection layer. Ingests schema and metadata, query logs, non-PII data samples, wikis and documentation, and BI reports and metrics.
Metrics Library	A central, reusable store of metric entities extracted from query history. Metrics are referenced, not duplicated, across models.
Company Glossary	An auto-generated business glossary mapping organizational terminology to the technical schema.
Asset Documentation	Auto-generated per-table and per-column descriptions and quality scores.
Benchmark Engine	The testing subsystem that runs benchmark evaluations, scores accuracy, and surfaces failures with root cause analysis.

See Glossary, Asset Documentation, and Metrics Library for more on the last three.