v0.3.0

[AivanF]

SLayer 0.3 Release Notes

Redesigned MCP capabilities

The MCP server has been substantially reworked. datasource_summary is replaced by models_summary — a per-datasource brief overview — and inspect_model now returns a complete single-model view with sampled dimension values, measure metadata, reachable join fields, and auto-pruned markdown tables. A new help tool provides contextual explanations of SLayer concepts (queries, transforms, joins, etc.) directly inside the agent conversation. All MCP tools are now fully async, and query output supports format parameter ("markdown", "json", "csv").

Async-first architecture

The engine, storage backends, and SQL client are now async-first. SlayerQueryEngine.execute() is async def; use execute_sync() for CLI, notebooks, and scripts. Postgres and MySQL queries use native async drivers (asyncpg, aiomysql) with pooled connections. SQLite and DuckDB fall back to asyncio.to_thread. Storage backends (YAMLStorage, SQLiteStorage) are async, ready for true async implementations like Postgres-backed multi-tenant storage. All tests use pytest-asyncio with asyncio_mode = "auto".

Variables in filters

Filters now support {variable} placeholders, substituted from the query's variables dict. This keeps filter templates reusable across different parameter values without string concatenation: "status = '{status}' AND amount > {min}" with variables={"status": "completed", "min": 100}. Variable names must be alphanumeric, values must be strings or numbers (inserted as-is). {{/}} produce literal braces. Undefined variables raise clear errors.

Meta fields on models, measures, and dimensions

Models, dimensions, and measures now have an optional meta: Dict[str, Any] field for arbitrary user-defined JSON metadata (e.g., {"owner": "analytics", "pii": true}). Persisted in YAML and SQLite storage, editable via MCP edit_model (pass null to clear), HTTP API, and CLI. The MCP sentinel pattern distinguishes "not provided" from "explicitly clear."

Default storage path

slayer serve and slayer mcp now work with zero configuration. When no --storage flag or $SLAYER_STORAGE is set, SLayer uses a platform-appropriate default: ~/.local/share/slayer on Linux (XDG-compliant), ~/Library/Application Support/slayer on macOS, %LOCALAPPDATA%\slayer on Windows.

dbt Semantic Layer import

SLayer can now import models from a dbt project via slayer import-dbt. Semantic models and metrics from dbt_project.yml are converted into SLayer models with dimensions, measures, and joins. The --include-hidden-models flag additionally imports every regular dbt model not wrapped by a semantic_model as a hidden SLayer model (queryable by name but excluded from discovery endpoints). Requires the dbt extra: pip install motley-slayer[dbt].

---

This discussion was created from the release v0.3.0.