posit-dev
diff --git a/‎.github/workflows/docs-r-pkgdown.yml‎
Lines changed: 1 addition & 1 deletion b/‎.github/workflows/docs-r-pkgdown.yml‎
Lines changed: 1 addition & 1 deletion
diff --git a/‎pkg-py/CHANGELOG.md‎
Lines changed: 6 additions & 0 deletions b/‎pkg-py/CHANGELOG.md‎
Lines changed: 6 additions & 0 deletions
diff --git a/‎pkg-py/src/querychat/_datasource.py‎
Lines changed: 24 additions & 4 deletions b/‎pkg-py/src/querychat/_datasource.py‎
Lines changed: 24 additions & 4 deletions
diff --git a/‎pkg-py/src/querychat/_snowflake.py‎
Lines changed: 126 additions & 0 deletions b/‎pkg-py/src/querychat/_snowflake.py‎
Lines changed: 126 additions & 0 deletions
diff --git a/‎pkg-py/src/querychat/_system_prompt.py‎
Lines changed: 4 additions & 2 deletions b/‎pkg-py/src/querychat/_system_prompt.py‎
Lines changed: 4 additions & 2 deletions
diff --git a/‎pkg-py/src/querychat/prompts/prompt.md‎
Lines changed: 1 addition & 0 deletions b/‎pkg-py/src/querychat/prompts/prompt.md‎
Lines changed: 1 addition & 0 deletions
diff --git a/‎pkg-py/src/querychat/prompts/semantic-views/prompt.md‎
Lines changed: 5 additions & 0 deletions b/‎pkg-py/src/querychat/prompts/semantic-views/prompt.md‎
Lines changed: 5 additions & 0 deletions
diff --git a/‎pkg-py/src/querychat/prompts/semantic-views/syntax.md‎
Lines changed: 96 additions & 0 deletions b/‎pkg-py/src/querychat/prompts/semantic-views/syntax.md‎
Lines changed: 96 additions & 0 deletions
@@ -48,7 +48,7 @@ jobs:
 
       - uses: r-lib/actions/setup-r-dependencies@v2
         with:
-          extra-packages: any::pkgdown, local::.
+          extra-packages: any::pkgdown, any::brand.yml, local::.
           needs: website
           working-directory: pkg-r
 
 
@@ -5,6 +5,12 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased]
+
+### New features
+
+* Added support for Snowflake Semantic Views. When connected to Snowflake (via SQLAlchemy or Ibis), querychat automatically discovers available Semantic Views and includes their definitions in the system prompt. This helps the LLM generate correct queries using the `SEMANTIC_VIEW()` table function with certified business metrics and dimensions. (#200)
+
 ## [0.5.1] - 2026-01-23
 
 ### New features
 
@@ -11,6 +11,10 @@
 from sqlalchemy.sql import sqltypes
 
 from ._df_compat import read_sql
+from ._snowflake import (
+    discover_semantic_views,
+    format_semantic_views,
+)
 from ._utils import as_narwhals, check_query
 
 if TYPE_CHECKING:
@@ -179,6 +183,10 @@ def cleanup(self) -> None:
 
         """
 
+    def get_semantic_views_description(self) -> str:
+        """Get information about semantic views (if any) for the system prompt."""
+        return ""
+
 
 class DataFrameSource(DataSource[IntoDataFrameT]):
     """A DataSource implementation that wraps a DataFrame using DuckDB."""
@@ -489,6 +497,13 @@ def get_schema(self, *, categorical_threshold: int) -> str:
         self._add_column_stats(columns, categorical_threshold)
         return format_schema(self.table_name, columns)
 
+    def get_semantic_views_description(self) -> str:
+        """Get information about semantic views (if any) for the system prompt."""
+        if self._engine.dialect.name.lower() != "snowflake":
+            return ""
+        views = discover_semantic_views(self._engine)
+        return format_semantic_views(views)
+
     @staticmethod
     def _make_column_meta(name: str, sa_type: sqltypes.TypeEngine) -> ColumnMeta:
         """Create ColumnMeta from SQLAlchemy type."""
@@ -895,8 +910,7 @@ def _add_column_stats(
 
         # Find text columns that qualify as categorical
         categorical_cols = [
-            col
-            for col in columns
+            col for col in columns
             if col.kind == "text"
             and (nunique := stats.get(f"{col.name}__nunique"))
             and nunique <= categorical_threshold
@@ -960,6 +974,13 @@ def get_schema(self, *, categorical_threshold: int) -> str:
         self._add_column_stats(columns, self._table, categorical_threshold)
         return format_schema(self.table_name, columns)
 
+    def get_semantic_views_description(self) -> str:
+        """Get information about semantic views (if any) for the system prompt."""
+        if self._backend.name.lower() != "snowflake":
+            return ""
+        views = discover_semantic_views(self._backend)
+        return format_semantic_views(views)
+
     @staticmethod
     def _make_column_meta(name: str, dtype: IbisDataType) -> ColumnMeta:
         """Create ColumnMeta from an ibis dtype."""
@@ -1018,8 +1039,7 @@ def _add_column_stats(
                 col.max_val = stats.get(f"{col.name}__max")
 
         categorical_cols = [
-            col
-            for col in columns
+            col for col in columns
             if col.kind == "text"
             and (nunique := stats.get(f"{col.name}__nunique"))
             and nunique <= categorical_threshold
 
@@ -0,0 +1,126 @@
+"""
+Snowflake-specific utilities for semantic view discovery.
+
+This module provides functions for discovering Snowflake Semantic Views,
+supporting both SQLAlchemy engines and Ibis backends via isinstance() checks.
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+from dataclasses import dataclass
+from typing import TYPE_CHECKING, Any
+
+import sqlalchemy
+
+if TYPE_CHECKING:
+    from ibis.backends.sql import SQLBackend
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class SemanticViewInfo:
+    """Metadata for a Snowflake Semantic View."""
+
+    name: str
+    """Fully qualified name (database.schema.view_name)."""
+
+    ddl: str
+    """The DDL definition from GET_DDL()."""
+
+
+def execute_raw_sql(
+    query: str,
+    backend: sqlalchemy.Engine | SQLBackend,
+) -> list[dict[str, Any]]:
+    """Execute raw SQL and return results as list of row dicts."""
+    if isinstance(backend, sqlalchemy.Engine):
+        with backend.connect() as conn:
+            result = conn.execute(sqlalchemy.text(query))
+            keys = list(result.keys())
+            return [dict(zip(keys, row, strict=False)) for row in result.fetchall()]
+    else:
+        with backend.raw_sql(query) as cursor:  # type: ignore[union-attr]
+            columns = [desc[0] for desc in cursor.description]
+            return [dict(zip(columns, row, strict=False)) for row in cursor.fetchall()]
+
+
+def discover_semantic_views(
+    backend: sqlalchemy.Engine | SQLBackend,
+) -> list[SemanticViewInfo]:
+    """Discover semantic views in the current schema."""
+    if os.environ.get("QUERYCHAT_DISABLE_SEMANTIC_VIEWS"):
+        return []
+
+    rows = execute_raw_sql("SHOW SEMANTIC VIEWS", backend)
+
+    if not rows:
+        logger.debug("No semantic views found in current schema")
+        return []
+
+    views: list[SemanticViewInfo] = []
+    for row in rows:
+        db = row.get("database_name")
+        schema = row.get("schema_name")
+        name = row.get("name")
+
+        if not name:
+            continue
+
+        fq_name = f"{db}.{schema}.{name}"
+        ddl = get_semantic_view_ddl(backend, fq_name)
+        if ddl:
+            views.append(SemanticViewInfo(name=fq_name, ddl=ddl))
+
+    return views
+
+
+def get_semantic_view_ddl(
+    backend: sqlalchemy.Engine | SQLBackend,
+    fq_name: str,
+) -> str | None:
+    """Get DDL for a semantic view by fully qualified name."""
+    safe_name = fq_name.replace("'", "''")
+    rows = execute_raw_sql(f"SELECT GET_DDL('SEMANTIC_VIEW', '{safe_name}')", backend)
+    if rows:
+        return str(next(iter(rows[0].values())))
+    return None
+
+
+def format_semantic_view_ddls(semantic_views: list[SemanticViewInfo]) -> str:
+    """Format just the DDL definitions for semantic views."""
+    lines: list[str] = []
+
+    for sv in semantic_views:
+        lines.append(f"### Semantic View: `{sv.name}`")
+        lines.append("")
+        lines.append("```sql")
+        lines.append(sv.ddl)
+        lines.append("```")
+        lines.append("")
+
+    return "\n".join(lines)
+
+
+def format_semantic_views(semantic_views: list[SemanticViewInfo]) -> str:
+    """Build the complete semantic views section for the prompt."""
+    if not semantic_views:
+        return ""
+
+    from importlib.resources import files
+
+    prompts = files("querychat.prompts.semantic-views")
+    prompt_text = (prompts / "prompt.md").read_text()
+    syntax_text = (prompts / "syntax.md").read_text()
+    ddls_text = format_semantic_view_ddls(semantic_views)
+
+    return f"""{prompt_text}
+
+{syntax_text}
+
+<semantic_views>
+{ddls_text}
+</semantic_views>
+"""
@@ -65,11 +65,13 @@ def render(self, tools: tuple[TOOL_GROUPS, ...] | None) -> str:
             Fully rendered system prompt string
 
         """
-        is_duck_db = self.data_source.get_db_type().lower() == "duckdb"
+        db_type = self.data_source.get_db_type()
+        is_duck_db = db_type.lower() == "duckdb"
 
         context = {
-            "db_type": self.data_source.get_db_type(),
+            "db_type": db_type,
             "is_duck_db": is_duck_db,
+            "semantic_views": self.data_source.get_semantic_views_description(),
             "schema": self.schema,
             "data_description": self.data_description,
             "extra_instructions": self.extra_instructions,
 
@@ -71,6 +71,7 @@ quantile_cont(salary, 0.5)
 ```
 
 {{/is_duck_db}}
+{{{semantic_views}}}
 ## Your Capabilities
 
 You can handle these types of requests:
 
@@ -0,0 +1,5 @@
+## Semantic Views
+
+**IMPORTANT**: This database has Semantic Views available. Semantic Views provide a curated layer over raw data with pre-defined metrics, dimensions, and relationships. They encode business logic and calculation rules that ensure consistent, accurate results. When a Semantic View covers the data you need, prefer it over raw tables to benefit from these certified definitions (that is, use the `SEMANTIC_VIEW()` table function where appropriate when generating SQL).
+
+**Real-world example**: A legacy ERP database had a revenue column (`X_AMT`) with hidden business rules—only status code 90 transactions count as realized revenue, and a discount factor (`ADJ_FCTR`) must be applied. Querying raw tables for "external customer revenue" returned **$184B**. The same query using the semantic model's certified `NET_REVENUE` metric returned **$84.5B**—the correct answer. The raw query was **2x+ too high** because it ignored discounts and included invalid transaction codes.
@@ -0,0 +1,96 @@
+### SEMANTIC_VIEW() Query Syntax
+
+#### Basic Syntax
+
+```sql
+SELECT * FROM SEMANTIC_VIEW(
+    {view_name}
+    METRICS {logical_table}.{metric_name}
+    DIMENSIONS {logical_table}.{dimension_name}
+    [WHERE {dimension} = 'value']  -- Optional: pre-aggregation filter
+)
+[WHERE {column} = 'value']  -- Optional: post-aggregation filter
+```
+
+#### Key Rules
+
+1. **Use `SEMANTIC_VIEW()` function** - Not direct SELECT FROM the view
+2. **No GROUP BY needed** - Semantic layer handles aggregation via DIMENSIONS
+3. **No JOINs needed within model** - Relationships are pre-defined
+4. **No aggregate functions needed** - Metrics are pre-aggregated
+5. **Use DDL-defined names** - Metrics and dimensions must match the DDL exactly
+
+#### WHERE Clause: Inside vs Outside
+
+- **Inside** (pre-aggregation): Filters base data BEFORE metrics are computed
+- **Outside** (post-aggregation): Filters results AFTER metrics are computed
+
+```sql
+-- Pre-aggregation: only include 'EXT' accounts in the calculation
+SELECT * FROM SEMANTIC_VIEW(
+    MODEL_NAME
+    METRICS T_DATA.NET_REVENUE
+    DIMENSIONS REF_ENTITIES.ACC_TYPE_CD
+    WHERE REF_ENTITIES.ACC_TYPE_CD = 'EXT'
+)
+
+-- Post-aggregation: compute all, then filter results
+SELECT * FROM SEMANTIC_VIEW(
+    MODEL_NAME
+    METRICS T_DATA.NET_REVENUE
+    DIMENSIONS REF_ENTITIES.ACC_TYPE_CD
+)
+WHERE NET_REVENUE > 1000000
+```
+
+#### Common Patterns
+
+**Single metric (total):**
+```sql
+SELECT * FROM SEMANTIC_VIEW(MODEL_NAME METRICS T_DATA.NET_REVENUE)
+```
+
+**Metric by dimension:**
+```sql
+SELECT * FROM SEMANTIC_VIEW(
+    MODEL_NAME
+    METRICS T_DATA.NET_REVENUE
+    DIMENSIONS REF_ENTITIES.ACC_TYPE_CD
+)
+```
+
+**Multiple metrics and dimensions:**
+```sql
+SELECT * FROM SEMANTIC_VIEW(
+    MODEL_NAME
+    METRICS T_DATA.NET_REVENUE, T_DATA.GROSS_REVENUE
+    DIMENSIONS REF_ENTITIES.ACC_TYPE_CD, T_DATA.LOG_DT
+)
+ORDER BY LOG_DT ASC
+```
+
+**Time series:**
+```sql
+SELECT * FROM SEMANTIC_VIEW(
+    MODEL_NAME
+    METRICS T_DATA.NET_REVENUE
+    DIMENSIONS T_DATA.LOG_DT
+)
+ORDER BY LOG_DT ASC
+```
+
+**Join results with other data:**
+```sql
+SELECT sv.*, lookup.category_name
+FROM SEMANTIC_VIEW(
+    MODEL_NAME
+    METRICS T_DATA.NET_REVENUE
+    DIMENSIONS REF_ENTITIES.ACC_TYPE_CD
+) AS sv
+JOIN category_lookup AS lookup ON sv.ACC_TYPE_CD = lookup.code
+```
+
+#### Troubleshooting
+
+- **"Invalid identifier"**: Verify metric/dimension names match exactly what's in the DDL
+- **Syntax error**: Use SEMANTIC_VIEW() function, GROUP BY isn't needed