qalita 2.9.0__py3-none-any.whl → 2.9.2__py3-none-any.whl

qalita/internal/action_executor.py

@@ -0,0 +1,896 @@
+"""
+# QALITA (c) COPYRIGHT 2025 - ALL RIGHTS RESERVED -
+Action Executor module for Studio Agent integration.
+
+This module provides an abstraction layer that translates LLM agent commands
+into concrete operations on data sources (SQL queries, file manipulations, etc.).
+"""
+
+import json
+import time
+from dataclasses import dataclass, field
+from typing import Any, Dict, List, Optional
+
+import pandas as pd
+
+from qalita.internal.utils import logger
+from qalita.internal.data_preview import (
+    DataPreviewResult,
+    _dataframe_to_preview,
+    _error_result,
+    DEFAULT_ROW_LIMIT,
+)
+
+
+@dataclass
+class ActionResult:
+    """Result of an action execution."""
+    
+    ok: bool
+    action_type: str
+    error: Optional[str] = None
+    result_json: Optional[str] = None  # Structured result as JSON string
+    data: Optional[DataPreviewResult] = None  # Tabular data if applicable
+    execution_time_ms: int = 0
+
+
+# Supported action types
+ACTION_TYPES = {
+    "query": "Execute a SQL query on a database source",
+    "read_data": "Read data from a file or database source",
+    "filter": "Filter data based on conditions",
+    "aggregate": "Perform aggregation on data",
+    "describe": "Get metadata about a source (schema, columns, row count)",
+    "sample": "Get a random sample of data",
+    "count": "Count rows in a source or query result",
+    "distinct": "Get distinct values from a column",
+    "head": "Get first N rows from a source",
+    "tail": "Get last N rows from a source",
+}
+
+
+class ActionExecutor:
+    """
+    Executes actions requested by the LLM agent.
+    
+    This class provides a unified interface for executing various data operations
+    on different source types (databases, files, etc.).
+    """
+    
+    def __init__(self):
+        """Initialize the action executor."""
+        self._engines: Dict[int, Any] = {}  # Cache for database engines
+    
+    def execute(
+        self,
+        action_type: str,
+        source_config: dict,
+        params: dict,
+        timeout_seconds: Optional[int] = None,
+    ) -> ActionResult:
+        """
+        Execute an action on a data source.
+        
+        Args:
+            action_type: Type of action to execute (query, read_data, etc.)
+            source_config: Source configuration dict with 'type' and 'config' keys
+            params: Action parameters (specific to each action type)
+            timeout_seconds: Optional timeout for the action
+        
+        Returns:
+            ActionResult with the execution result
+        """
+        start_time = time.time()
+        
+        if action_type not in ACTION_TYPES:
+            return ActionResult(
+                ok=False,
+                action_type=action_type,
+                error=f"Unknown action type: {action_type}. Supported: {list(ACTION_TYPES.keys())}",
+            )
+        
+        handlers = {
+            "query": self._execute_query,
+            "read_data": self._read_data,
+            "filter": self._filter_data,
+            "aggregate": self._aggregate_data,
+            "describe": self._describe_source,
+            "sample": self._sample_data,
+            "count": self._count_rows,
+            "distinct": self._get_distinct,
+            "head": self._get_head,
+            "tail": self._get_tail,
+        }
+        
+        handler = handlers.get(action_type)
+        if not handler:
+            return ActionResult(
+                ok=False,
+                action_type=action_type,
+                error=f"Handler not implemented for action: {action_type}",
+            )
+        
+        try:
+            result = handler(source_config, params)
+            result.execution_time_ms = int((time.time() - start_time) * 1000)
+            return result
+        except Exception as e:
+            logger.error(f"Error executing action {action_type}: {e}")
+            return ActionResult(
+                ok=False,
+                action_type=action_type,
+                error=str(e),
+                execution_time_ms=int((time.time() - start_time) * 1000),
+            )
+    
+    def _get_database_engine(self, source_config: dict) -> Any:
+        """Get or create a SQLAlchemy engine for a database source."""
+        from sqlalchemy import create_engine
+        
+        config = source_config.get("config", {})
+        source_type = source_config.get("type", "").lower()
+        
+        connection_string = config.get("connection_string")
+        if not connection_string:
+            db_type_map = {
+                "postgresql": "postgresql",
+                "mysql": "mysql",
+                "mssql": "mssql+pymssql",
+                "oracle": "oracle+oracledb",
+                "sqlite": "sqlite",
+            }
+            
+            dialect = db_type_map.get(source_type)
+            if not dialect:
+                raise ValueError(f"Unsupported database type: {source_type}")
+            
+            if source_type == "sqlite":
+                database_path = config.get("database", ":memory:")
+                connection_string = f"sqlite:///{database_path}"
+            elif source_type == "oracle":
+                connection_string = (
+                    f"{dialect}://{config['username']}:{config['password']}"
+                    f"@{config['host']}:{config['port']}/?service_name={config['database']}"
+                )
+            else:
+                connection_string = (
+                    f"{dialect}://{config['username']}:{config['password']}"
+                    f"@{config['host']}:{config['port']}/{config['database']}"
+                )
+        
+        return create_engine(connection_string)
+    
+    def _is_database_source(self, source_config: dict) -> bool:
+        """Check if the source is a database."""
+        source_type = source_config.get("type", "").lower()
+        return source_type in ("postgresql", "mysql", "mssql", "oracle", "sqlite")
+    
+    def _is_file_source(self, source_config: dict) -> bool:
+        """Check if the source is a file."""
+        source_type = source_config.get("type", "").lower()
+        return source_type in ("file", "csv", "excel", "parquet", "json", "folder")
+    
+    def _execute_query(self, source_config: dict, params: dict) -> ActionResult:
+        """Execute a SQL query on a database source."""
+        if not self._is_database_source(source_config):
+            return ActionResult(
+                ok=False,
+                action_type="query",
+                error=f"Query action only supported for database sources, not {source_config.get('type')}",
+            )
+        
+        sql = params.get("sql")
+        if not sql:
+            return ActionResult(
+                ok=False,
+                action_type="query",
+                error="SQL query is required for 'query' action",
+            )
+        
+        limit = params.get("limit", DEFAULT_ROW_LIMIT)
+        
+        # Add LIMIT if not present (for safety)
+        sql_lower = sql.strip().lower()
+        if "limit" not in sql_lower and not sql_lower.startswith(("insert", "update", "delete", "create", "drop", "alter")):
+            sql = f"{sql.rstrip(';')} LIMIT {limit}"
+        
+        try:
+            engine = self._get_database_engine(source_config)
+            with engine.connect() as conn:
+                df = pd.read_sql(sql, conn)
+            
+            preview = _dataframe_to_preview(df, limit)
+            return ActionResult(
+                ok=True,
+                action_type="query",
+                data=preview,
+                result_json=json.dumps({"rows_returned": len(df), "columns": list(df.columns)}),
+            )
+        except Exception as e:
+            return ActionResult(
+                ok=False,
+                action_type="query",
+                error=f"Query execution failed: {str(e)}",
+            )
+    
+    def _read_data(self, source_config: dict, params: dict) -> ActionResult:
+        """Read data from a source."""
+        limit = params.get("limit", DEFAULT_ROW_LIMIT)
+        columns = params.get("columns")  # Optional list of columns to select
+        
+        if self._is_database_source(source_config):
+            config = source_config.get("config", {})
+            table = params.get("table") or config.get("table") or config.get("default_table")
+            if not table:
+                return ActionResult(
+                    ok=False,
+                    action_type="read_data",
+                    error="Table name is required for database sources",
+                )
+            
+            schema = config.get("schema")
+            qualified_table = f"{schema}.{table}" if schema else table
+            
+            cols = ", ".join(columns) if columns else "*"
+            sql = f"SELECT {cols} FROM {qualified_table} LIMIT {limit}"
+            
+            try:
+                engine = self._get_database_engine(source_config)
+                with engine.connect() as conn:
+                    df = pd.read_sql(sql, conn)
+                
+                preview = _dataframe_to_preview(df, limit)
+                return ActionResult(
+                    ok=True,
+                    action_type="read_data",
+                    data=preview,
+                    result_json=json.dumps({"rows_returned": len(df), "columns": list(df.columns)}),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="read_data",
+                    error=f"Failed to read data: {str(e)}",
+                )
+        
+        elif self._is_file_source(source_config):
+            return self._read_file_data(source_config, params)
+        
+        else:
+            return ActionResult(
+                ok=False,
+                action_type="read_data",
+                error=f"Unsupported source type: {source_config.get('type')}",
+            )
+    
+    def _read_file_data(self, source_config: dict, params: dict) -> ActionResult:
+        """Read data from a file source."""
+        import os
+        
+        config = source_config.get("config", {})
+        source_type = source_config.get("type", "").lower()
+        path = config.get("path")
+        
+        if not path:
+            return ActionResult(
+                ok=False,
+                action_type="read_data",
+                error="File path not configured",
+            )
+        
+        if not os.path.exists(path):
+            return ActionResult(
+                ok=False,
+                action_type="read_data",
+                error=f"File not found: {path}",
+            )
+        
+        limit = params.get("limit", DEFAULT_ROW_LIMIT)
+        columns = params.get("columns")
+        
+        try:
+            if source_type in ("csv", "file") and path.endswith(".csv"):
+                usecols = columns if columns else None
+                df = pd.read_csv(path, nrows=limit, usecols=usecols, low_memory=False)
+            elif source_type == "excel" or path.endswith((".xlsx", ".xls")):
+                usecols = columns if columns else None
+                df = pd.read_excel(path, nrows=limit, usecols=usecols, engine="openpyxl")
+            elif source_type == "parquet" or path.endswith(".parquet"):
+                df = pd.read_parquet(path, columns=columns)
+                df = df.head(limit)
+            elif source_type == "json" or path.endswith(".json"):
+                df = pd.read_json(path)
+                if columns:
+                    df = df[columns]
+                df = df.head(limit)
+            else:
+                return ActionResult(
+                    ok=False,
+                    action_type="read_data",
+                    error=f"Unsupported file type: {source_type}",
+                )
+            
+            preview = _dataframe_to_preview(df, limit)
+            return ActionResult(
+                ok=True,
+                action_type="read_data",
+                data=preview,
+                result_json=json.dumps({"rows_returned": len(df), "columns": list(df.columns)}),
+            )
+        except Exception as e:
+            return ActionResult(
+                ok=False,
+                action_type="read_data",
+                error=f"Failed to read file: {str(e)}",
+            )
+    
+    def _filter_data(self, source_config: dict, params: dict) -> ActionResult:
+        """Filter data based on a condition."""
+        condition = params.get("condition")
+        if not condition:
+            return ActionResult(
+                ok=False,
+                action_type="filter",
+                error="Filter condition is required",
+            )
+        
+        limit = params.get("limit", DEFAULT_ROW_LIMIT)
+        
+        if self._is_database_source(source_config):
+            config = source_config.get("config", {})
+            table = params.get("table") or config.get("table") or config.get("default_table")
+            if not table:
+                return ActionResult(
+                    ok=False,
+                    action_type="filter",
+                    error="Table name is required for database sources",
+                )
+            
+            schema = config.get("schema")
+            qualified_table = f"{schema}.{table}" if schema else table
+            columns = params.get("columns")
+            cols = ", ".join(columns) if columns else "*"
+            
+            sql = f"SELECT {cols} FROM {qualified_table} WHERE {condition} LIMIT {limit}"
+            
+            try:
+                engine = self._get_database_engine(source_config)
+                with engine.connect() as conn:
+                    df = pd.read_sql(sql, conn)
+                
+                preview = _dataframe_to_preview(df, limit)
+                return ActionResult(
+                    ok=True,
+                    action_type="filter",
+                    data=preview,
+                    result_json=json.dumps({"rows_returned": len(df), "condition": condition}),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="filter",
+                    error=f"Filter failed: {str(e)}",
+                )
+        
+        elif self._is_file_source(source_config):
+            # First read the data
+            read_result = self._read_file_data(source_config, {"limit": 10000})  # Read more for filtering
+            if not read_result.ok or not read_result.data:
+                return ActionResult(
+                    ok=False,
+                    action_type="filter",
+                    error=read_result.error or "Failed to read data for filtering",
+                )
+            
+            # Reconstruct dataframe and filter
+            try:
+                df = pd.DataFrame(
+                    [row for row in read_result.data.rows],
+                    columns=read_result.data.headers
+                )
+                # Use query for filtering
+                df_filtered = df.query(condition)
+                df_filtered = df_filtered.head(limit)
+                
+                preview = _dataframe_to_preview(df_filtered, limit)
+                return ActionResult(
+                    ok=True,
+                    action_type="filter",
+                    data=preview,
+                    result_json=json.dumps({"rows_returned": len(df_filtered), "condition": condition}),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="filter",
+                    error=f"Filter failed: {str(e)}",
+                )
+        
+        else:
+            return ActionResult(
+                ok=False,
+                action_type="filter",
+                error=f"Unsupported source type: {source_config.get('type')}",
+            )
+    
+    def _aggregate_data(self, source_config: dict, params: dict) -> ActionResult:
+        """Perform aggregation on data."""
+        group_by = params.get("group_by")  # Column(s) to group by
+        agg_func = params.get("agg_func", "count")  # Aggregation function
+        agg_column = params.get("agg_column")  # Column to aggregate
+        
+        if not group_by:
+            return ActionResult(
+                ok=False,
+                action_type="aggregate",
+                error="group_by column is required for aggregation",
+            )
+        
+        limit = params.get("limit", DEFAULT_ROW_LIMIT)
+        
+        if self._is_database_source(source_config):
+            config = source_config.get("config", {})
+            table = params.get("table") or config.get("table") or config.get("default_table")
+            if not table:
+                return ActionResult(
+                    ok=False,
+                    action_type="aggregate",
+                    error="Table name is required for database sources",
+                )
+            
+            schema = config.get("schema")
+            qualified_table = f"{schema}.{table}" if schema else table
+            
+            # Build SQL aggregation
+            group_cols = group_by if isinstance(group_by, list) else [group_by]
+            group_str = ", ".join(group_cols)
+            
+            if agg_func.upper() == "COUNT":
+                agg_expr = "COUNT(*) as count"
+            elif agg_column:
+                agg_expr = f"{agg_func.upper()}({agg_column}) as {agg_func.lower()}_{agg_column}"
+            else:
+                agg_expr = "COUNT(*) as count"
+            
+            sql = f"SELECT {group_str}, {agg_expr} FROM {qualified_table} GROUP BY {group_str} LIMIT {limit}"
+            
+            try:
+                engine = self._get_database_engine(source_config)
+                with engine.connect() as conn:
+                    df = pd.read_sql(sql, conn)
+                
+                preview = _dataframe_to_preview(df, limit)
+                return ActionResult(
+                    ok=True,
+                    action_type="aggregate",
+                    data=preview,
+                    result_json=json.dumps({
+                        "groups": len(df),
+                        "group_by": group_by,
+                        "agg_func": agg_func,
+                    }),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="aggregate",
+                    error=f"Aggregation failed: {str(e)}",
+                )
+        
+        else:
+            return ActionResult(
+                ok=False,
+                action_type="aggregate",
+                error="Aggregation currently only supported for database sources",
+            )
+    
+    def _describe_source(self, source_config: dict, params: dict) -> ActionResult:
+        """Get metadata about a source."""
+        source_type = source_config.get("type", "").lower()
+        config = source_config.get("config", {})
+        
+        metadata = {
+            "source_type": source_type,
+            "name": source_config.get("name", "unknown"),
+        }
+        
+        if self._is_database_source(source_config):
+            try:
+                from sqlalchemy import inspect
+                
+                engine = self._get_database_engine(source_config)
+                inspector = inspect(engine)
+                
+                # Get schema info
+                schema = config.get("schema")
+                tables = inspector.get_table_names(schema=schema)
+                
+                metadata["tables"] = tables
+                metadata["schema"] = schema
+                
+                # Get column info for specified table or first table
+                table = params.get("table") or config.get("table") or (tables[0] if tables else None)
+                if table:
+                    columns = inspector.get_columns(table, schema=schema)
+                    metadata["table"] = table
+                    metadata["columns"] = [
+                        {"name": col["name"], "type": str(col["type"])}
+                        for col in columns
+                    ]
+                    
+                    # Get row count
+                    with engine.connect() as conn:
+                        from sqlalchemy import text
+                        qualified_table = f"{schema}.{table}" if schema else table
+                        result = conn.execute(text(f"SELECT COUNT(*) FROM {qualified_table}"))
+                        metadata["row_count"] = result.scalar()
+                
+                return ActionResult(
+                    ok=True,
+                    action_type="describe",
+                    result_json=json.dumps(metadata),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="describe",
+                    error=f"Failed to describe database source: {str(e)}",
+                )
+        
+        elif self._is_file_source(source_config):
+            import os
+            
+            path = config.get("path")
+            if not path:
+                return ActionResult(
+                    ok=False,
+                    action_type="describe",
+                    error="File path not configured",
+                )
+            
+            if not os.path.exists(path):
+                return ActionResult(
+                    ok=False,
+                    action_type="describe",
+                    error=f"File not found: {path}",
+                )
+            
+            metadata["path"] = path
+            metadata["file_size_bytes"] = os.path.getsize(path)
+            
+            try:
+                # Read a small sample to get column info
+                if source_type in ("csv", "file") and path.endswith(".csv"):
+                    df = pd.read_csv(path, nrows=5, low_memory=False)
+                elif source_type == "excel" or path.endswith((".xlsx", ".xls")):
+                    df = pd.read_excel(path, nrows=5, engine="openpyxl")
+                elif source_type == "parquet" or path.endswith(".parquet"):
+                    df = pd.read_parquet(path)
+                    df = df.head(5)
+                elif source_type == "json" or path.endswith(".json"):
+                    df = pd.read_json(path)
+                    df = df.head(5)
+                else:
+                    df = None
+                
+                if df is not None:
+                    metadata["columns"] = [
+                        {"name": col, "type": str(df[col].dtype)}
+                        for col in df.columns
+                    ]
+                    # Try to get total row count
+                    if source_type in ("csv", "file") and path.endswith(".csv"):
+                        with open(path, 'r', encoding='utf-8', errors='ignore') as f:
+                            metadata["row_count"] = sum(1 for _ in f) - 1  # Exclude header
+                    elif source_type == "parquet" or path.endswith(".parquet"):
+                        metadata["row_count"] = len(pd.read_parquet(path))
+                
+                return ActionResult(
+                    ok=True,
+                    action_type="describe",
+                    result_json=json.dumps(metadata),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="describe",
+                    error=f"Failed to describe file source: {str(e)}",
+                )
+        
+        else:
+            return ActionResult(
+                ok=False,
+                action_type="describe",
+                error=f"Describe not supported for source type: {source_type}",
+            )
+    
+    def _sample_data(self, source_config: dict, params: dict) -> ActionResult:
+        """Get a random sample of data."""
+        n = params.get("n", 10)  # Number of samples
+        
+        if self._is_database_source(source_config):
+            config = source_config.get("config", {})
+            table = params.get("table") or config.get("table") or config.get("default_table")
+            if not table:
+                return ActionResult(
+                    ok=False,
+                    action_type="sample",
+                    error="Table name is required",
+                )
+            
+            schema = config.get("schema")
+            qualified_table = f"{schema}.{table}" if schema else table
+            source_type = source_config.get("type", "").lower()
+            
+            # Different databases have different random sampling syntax
+            if source_type == "postgresql":
+                sql = f"SELECT * FROM {qualified_table} ORDER BY RANDOM() LIMIT {n}"
+            elif source_type == "mysql":
+                sql = f"SELECT * FROM {qualified_table} ORDER BY RAND() LIMIT {n}"
+            else:
+                sql = f"SELECT * FROM {qualified_table} LIMIT {n}"  # Fallback
+            
+            try:
+                engine = self._get_database_engine(source_config)
+                with engine.connect() as conn:
+                    df = pd.read_sql(sql, conn)
+                
+                preview = _dataframe_to_preview(df, n)
+                return ActionResult(
+                    ok=True,
+                    action_type="sample",
+                    data=preview,
+                    result_json=json.dumps({"samples": len(df)}),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="sample",
+                    error=f"Sampling failed: {str(e)}",
+                )
+        
+        elif self._is_file_source(source_config):
+            # Read all data and sample
+            read_result = self._read_file_data(source_config, {"limit": 10000})
+            if not read_result.ok or not read_result.data:
+                return ActionResult(
+                    ok=False,
+                    action_type="sample",
+                    error=read_result.error or "Failed to read data for sampling",
+                )
+            
+            try:
+                df = pd.DataFrame(
+                    [row for row in read_result.data.rows],
+                    columns=read_result.data.headers
+                )
+                df_sample = df.sample(n=min(n, len(df)))
+                
+                preview = _dataframe_to_preview(df_sample, n)
+                return ActionResult(
+                    ok=True,
+                    action_type="sample",
+                    data=preview,
+                    result_json=json.dumps({"samples": len(df_sample)}),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="sample",
+                    error=f"Sampling failed: {str(e)}",
+                )
+        
+        else:
+            return ActionResult(
+                ok=False,
+                action_type="sample",
+                error=f"Sampling not supported for source type: {source_config.get('type')}",
+            )
+    
+    def _count_rows(self, source_config: dict, params: dict) -> ActionResult:
+        """Count rows in a source."""
+        if self._is_database_source(source_config):
+            config = source_config.get("config", {})
+            table = params.get("table") or config.get("table") or config.get("default_table")
+            condition = params.get("condition")
+            
+            if not table:
+                return ActionResult(
+                    ok=False,
+                    action_type="count",
+                    error="Table name is required",
+                )
+            
+            schema = config.get("schema")
+            qualified_table = f"{schema}.{table}" if schema else table
+            
+            sql = f"SELECT COUNT(*) as count FROM {qualified_table}"
+            if condition:
+                sql += f" WHERE {condition}"
+            
+            try:
+                engine = self._get_database_engine(source_config)
+                with engine.connect() as conn:
+                    from sqlalchemy import text
+                    result = conn.execute(text(sql))
+                    count = result.scalar()
+                
+                return ActionResult(
+                    ok=True,
+                    action_type="count",
+                    result_json=json.dumps({"count": count, "table": table}),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="count",
+                    error=f"Count failed: {str(e)}",
+                )
+        
+        else:
+            # Use describe for file sources
+            describe_result = self._describe_source(source_config, params)
+            if describe_result.ok and describe_result.result_json:
+                metadata = json.loads(describe_result.result_json)
+                if "row_count" in metadata:
+                    return ActionResult(
+                        ok=True,
+                        action_type="count",
+                        result_json=json.dumps({"count": metadata["row_count"]}),
+                    )
+            
+            return ActionResult(
+                ok=False,
+                action_type="count",
+                error="Could not determine row count",
+            )
+    
+    def _get_distinct(self, source_config: dict, params: dict) -> ActionResult:
+        """Get distinct values from a column."""
+        column = params.get("column")
+        if not column:
+            return ActionResult(
+                ok=False,
+                action_type="distinct",
+                error="Column name is required for distinct action",
+            )
+        
+        limit = params.get("limit", 100)
+        
+        if self._is_database_source(source_config):
+            config = source_config.get("config", {})
+            table = params.get("table") or config.get("table") or config.get("default_table")
+            
+            if not table:
+                return ActionResult(
+                    ok=False,
+                    action_type="distinct",
+                    error="Table name is required",
+                )
+            
+            schema = config.get("schema")
+            qualified_table = f"{schema}.{table}" if schema else table
+            
+            sql = f"SELECT DISTINCT {column} FROM {qualified_table} LIMIT {limit}"
+            
+            try:
+                engine = self._get_database_engine(source_config)
+                with engine.connect() as conn:
+                    df = pd.read_sql(sql, conn)
+                
+                values = df[column].tolist()
+                return ActionResult(
+                    ok=True,
+                    action_type="distinct",
+                    result_json=json.dumps({
+                        "column": column,
+                        "distinct_count": len(values),
+                        "values": values[:limit],
+                    }),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="distinct",
+                    error=f"Distinct failed: {str(e)}",
+                )
+        
+        else:
+            return ActionResult(
+                ok=False,
+                action_type="distinct",
+                error="Distinct currently only supported for database sources",
+            )
+    
+    def _get_head(self, source_config: dict, params: dict) -> ActionResult:
+        """Get first N rows from a source."""
+        n = params.get("n", 10)
+        params["limit"] = n
+        return self._read_data(source_config, params)
+    
+    def _get_tail(self, source_config: dict, params: dict) -> ActionResult:
+        """Get last N rows from a source."""
+        n = params.get("n", 10)
+        
+        if self._is_database_source(source_config):
+            config = source_config.get("config", {})
+            table = params.get("table") or config.get("table") or config.get("default_table")
+            
+            if not table:
+                return ActionResult(
+                    ok=False,
+                    action_type="tail",
+                    error="Table name is required",
+                )
+            
+            schema = config.get("schema")
+            qualified_table = f"{schema}.{table}" if schema else table
+            
+            # This requires knowing the order - use a subquery with DESC ordering
+            sql = f"""
+                SELECT * FROM (
+                    SELECT * FROM {qualified_table} ORDER BY 1 DESC LIMIT {n}
+                ) sub ORDER BY 1 ASC
+            """
+            
+            try:
+                engine = self._get_database_engine(source_config)
+                with engine.connect() as conn:
+                    df = pd.read_sql(sql, conn)
+                
+                preview = _dataframe_to_preview(df, n)
+                return ActionResult(
+                    ok=True,
+                    action_type="tail",
+                    data=preview,
+                    result_json=json.dumps({"rows_returned": len(df)}),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="tail",
+                    error=f"Tail failed: {str(e)}",
+                )
+        
+        else:
+            # For files, read all and take tail
+            read_result = self._read_file_data(source_config, {"limit": 10000})
+            if not read_result.ok or not read_result.data:
+                return ActionResult(
+                    ok=False,
+                    action_type="tail",
+                    error=read_result.error or "Failed to read data",
+                )
+            
+            try:
+                df = pd.DataFrame(
+                    [row for row in read_result.data.rows],
+                    columns=read_result.data.headers
+                )
+                df_tail = df.tail(n)
+                
+                preview = _dataframe_to_preview(df_tail, n)
+                return ActionResult(
+                    ok=True,
+                    action_type="tail",
+                    data=preview,
+                    result_json=json.dumps({"rows_returned": len(df_tail)}),
+                )
+            except Exception as e:
+                return ActionResult(
+                    ok=False,
+                    action_type="tail",
+                    error=f"Tail failed: {str(e)}",
+                )
+
+
+# Singleton instance
+_executor: Optional[ActionExecutor] = None
+
+
+def get_action_executor() -> ActionExecutor:
+    """Get the singleton ActionExecutor instance."""
+    global _executor
+    if _executor is None:
+        _executor = ActionExecutor()
+    return _executor