update
May 11, 2026
By Teun
SLayer puts a semantic layer under AI agents
SLayer is a semantic layer that sits between databases and AI agents, internal tools, dashboards, and scripts. The project says it can auto-create models from schema, compile queries into SQL, and let agents update models at runtime through MCP, REST, CLI, and Python interfaces.
SLayer is a semantic layer for AI agents and internal tools that queries databases, manages data models, and stores natural-language memories about data and queries. The project was posted on Hacker News as “Show HN: SLayer, a semantic layer maintained by your agent.”
According to the project description, SLayer sits between a database and the systems that need to use it, including AI agents, dashboards, scripts, and other internal tools. It is designed to expose structured concepts such as measures, dimensions, and filters rather than forcing every request into hand-written SQL.
The project says one of its main features is “warm start” model creation. SLayer can inspect a database schema, detect foreign-key relationships, and generate models automatically, including joined dimensions and count-distinct measures. It also supports editing models at runtime, with changes taking effect immediately.
SLayer also lets users specify aggregations at query time instead of baking them into the model. In the examples provided, queries can ask for metrics such as month-over-month percentage change, cumulative sums, lagged values, or comparisons against previous periods. The project says its DSL supports nested formulas and can compile them into the correct SQL for the target database, including joins, aggregations, time-based calculations, and dialect differences.
The project supports multiple ways to access the same functionality. According to the documentation snippets in the post, SLayer exposes MCP, REST API, CLI, and Python interfaces. It can run in-process as a Python module or as a serverless tool via the CLI, and it supports both HTTP and stdio-based MCP transports.
For agent workflows, SLayer can create a datasource, ingest models from a schema, inspect models, and run queries conversationally over MCP. The post also shows CLI examples for starting a server, querying data from the terminal, and connecting to a database with a connection URL. The examples include a demo setup based on the Jaffle Shop DuckDB database and instructions for using Claude Code with an in-process MCP server.
The project says it supports most popular databases and offers two storage backends. YAMLStorage stores models and datasources as YAML files on disk for version control, while SQLiteStorage keeps everything in a single SQLite file. The docs also say custom storage backends can be added for use cases such as tenant isolation.
SLayer’s query language is built around model names, measures, dimensions, time dimensions, and filters. The post gives an example JSON query for counting orders by status, and another for monthly revenue by store with month-over-month percentage change. The project says models are defined as YAML by default, with optional descriptions and named formulas such as average order value.
The author also listed features that are not yet implemented. According to the post, SLayer currently does not include caching or pre-aggregation, and the roadmap also lists access controls, governance, unpivoting, asof joins, and chart generation as incomplete items. The project is released under MIT license.
