dbt Semantic Layer vs LLMSchema

The dbt Semantic Layer centralizes metric definitions on dbt models and handles joins so downstream tools and applications can use metrics consistently. MetricFlow generates SQL at query time from semantic model and metric configurations. Teams define those models and metrics in YAML as version-controlled, lineage-aware code. MetricFlow can optimize joins, rewrite queries to reduce compute and latency, and cache repeated queries. The product offers an API, the dbt MCP Server and integrations for BI tools, notebooks, spreadsheets, embedded analytics and AI systems. GraphQL and JDBC APIs and a Python SDK provide ways to query metrics and dimensions; Claude and ChatGPT can connect to governed metrics through the MCP server. Documented warehouse platforms include Snowflake, BigQuery, Databricks and Redshift. Defining and querying metrics requires a dbt Starter or Enterprise-tier account, and single-tenant accounts need account-representative setup and enablement. Starter includes 5,000 queried metrics per month; Enterprise and Enterprise+ each include 20,000. Starter costs $100 a month per user. The free Developer plan does not list the Semantic Layer.

LLMSchema is free, MIT-licensed software that generates database schema documentation for LLMs and AI agents. It extracts relational schemas from PostgreSQL, MySQL, and SQLite, then documents tables, columns, types, indexes, constraints, and relationships in Markdown. The default output is one portable Markdown document. An alternate layout creates an overview plus a separate Markdown file for each table. Users can include selected tables or exclude tables from the generated documentation. LLMSchema is available as a command-line tool and a Go library, with installation instructions for Go and a quick install script for macOS and Linux. The README shows adding schema generation to a Makefile or migration script to keep documentation current. It also suggests pointing a coding agent to the schema files from AGENTS.md, CLAUDE.md, or an equivalent instruction file. The project is self-hosted and listed for Linux, macOS, and Windows. Its maker positions it for development databases and says not to rely on it for production-critical documentation.