SLayer: An Open-Source Semantic Layer for AI Agents That Learns from Queries

SLayer is an open-source semantic layer designed for AI agents to query databases, manage data models, and improve over time through natural-language memories. It sits between your database and agents (or internal tools), providing a structured DSL for measures, dimensions, and filters—avoiding the mess of raw SQL generated by LLMs.

Key Features from the Source

• Auto-creation of models from database schema introspection for a warm start.
• Runtime model editing: agents can edit columns/measures or create new models on the fly from SQL or other models.
• Natural-language memories: save and retrieve memories linked to models, columns, or queries to form a knowledge base.
• Embeddability: runs in-process as a Python module or serverless via CLI; no server required.
• Schema drift detection and handling – agents can adapt to changing table structures.
• Expressive DSL supports multi-stage queries, custom aggregations, time shifts, and combining metrics from multiple models.
• Multiple interfaces: MCP (stdio and SSE), REST API, CLI, and Python client for dataframes.
• No caching or pre-aggregation engine yet – noted as a limitation; on roadmap.

Quickstart Examples

Install via uv:

uv tool install motley-slayer
slayer

Instant demo with bundled Jaffle Shop DuckDB:

uvx --from 'motley-slayer[all]' slayer serve --demo

Connect to Claude Code via stdio MCP (serverless):

claude mcp add slayer -- uvx --from motley-slayer slayer mcp --demo

Query via REST API:

curl -X POST http://localhost:5143/query \
  -H "Content-Type: application/json" \
  -d '{"source_model": "orders", "measures": ["*:count"], "dimensions": ["status"]}'

List models:

curl http://localhost:5143/models

Python client usage:

from slayer.client.slayer_client import Slay

Who It's For

Developers building AI data analyst chatbots, agentic apps, or any tool where agents need to explore databases iteratively and learn from past queries.

Docs: motley-slayer.readthedocs.io