agno 2.2.13__py3-none-any.whl → 2.4.3__py3-none-any.whl

agno/db/utils.py

@@ -2,10 +2,36 @@
 
 import json
 from datetime import date, datetime
+from typing import TYPE_CHECKING, Any, Dict, Optional, Union
 from uuid import UUID
 
 from agno.models.message import Message
 from agno.models.metrics import Metrics
+from agno.utils.log import log_error, log_warning
+
+if TYPE_CHECKING:
+    from agno.db.base import BaseDb
+
+
+def get_sort_value(record: Dict[str, Any], sort_by: str) -> Any:
+    """Get the sort value for a record, with fallback to created_at for updated_at.
+
+    When sorting by 'updated_at', this function falls back to 'created_at' if
+    'updated_at' is None. This ensures pre-2.0 records (which may have NULL
+    updated_at values) are sorted correctly by their creation time.
+
+    Args:
+        record: The record dictionary to get the sort value from
+        sort_by: The field to sort by
+
+    Returns:
+        The value to use for sorting
+    """
+    value = record.get(sort_by)
+    # For updated_at, fall back to created_at if updated_at is None
+    if value is None and sort_by == "updated_at":
+        value = record.get("created_at")
+    return value
 
 
 class CustomJSONEncoder(json.JSONEncoder):
@@ -20,6 +46,8 @@ class CustomJSONEncoder(json.JSONEncoder):
             return obj.to_dict()
         elif isinstance(obj, Metrics):
             return obj.to_dict()
+        elif isinstance(obj, type):
+            return str(obj)
 
         return super().default(obj)
 
@@ -114,3 +142,35 @@ def deserialize_session_json_fields(session: dict) -> dict:
             log_warning(f"Warning: Could not parse runs as JSON, keeping as string: {e}")
 
     return session
+
+
+def db_from_dict(db_data: Dict[str, Any]) -> Optional[Union["BaseDb"]]:
+    """
+    Create a database instance from a dictionary.
+
+    Args:
+        db_data: Dictionary containing database configuration
+
+    Returns:
+        Database instance or None if creation fails
+    """
+    db_type = db_data.get("type")
+    if db_type == "postgres":
+        try:
+            from agno.db.postgres import PostgresDb
+
+            return PostgresDb.from_dict(db_data)
+        except Exception as e:
+            log_error(f"Error reconstructing PostgresDb from dictionary: {e}")
+            return None
+    elif db_type == "sqlite":
+        try:
+            from agno.db.sqlite import SqliteDb
+
+            return SqliteDb.from_dict(db_data)
+        except Exception as e:
+            log_error(f"Error reconstructing SqliteDb from dictionary: {e}")
+            return None
+    else:
+        log_warning(f"Unknown database type: {db_type}")
+        return None

agno/eval/__init__.py

@@ -1,14 +1,37 @@
-from agno.eval.accuracy import AccuracyAgentResponse, AccuracyEval, AccuracyEvaluation, AccuracyResult
-from agno.eval.performance import PerformanceEval, PerformanceResult
-from agno.eval.reliability import ReliabilityEval, ReliabilityResult
+from agno.eval.base import BaseEval
 
 __all__ = [
     "AccuracyAgentResponse",
     "AccuracyEvaluation",
     "AccuracyResult",
     "AccuracyEval",
+    "AgentAsJudgeEval",
+    "AgentAsJudgeEvaluation",
+    "AgentAsJudgeResult",
+    "BaseEval",
     "PerformanceEval",
     "PerformanceResult",
     "ReliabilityEval",
     "ReliabilityResult",
 ]
+
+
+def __getattr__(name: str):
+    """Lazy import for eval implementations to avoid circular imports with Agent."""
+    if name in ("AccuracyAgentResponse", "AccuracyEval", "AccuracyEvaluation", "AccuracyResult"):
+        from agno.eval import accuracy
+
+        return getattr(accuracy, name)
+    elif name in ("AgentAsJudgeEval", "AgentAsJudgeEvaluation", "AgentAsJudgeResult"):
+        from agno.eval import agent_as_judge
+
+        return getattr(agent_as_judge, name)
+    elif name in ("PerformanceEval", "PerformanceResult"):
+        from agno.eval import performance
+
+        return getattr(performance, name)
+    elif name in ("ReliabilityEval", "ReliabilityResult"):
+        from agno.eval import reliability
+
+        return getattr(reliability, name)
+    raise AttributeError(f"module '{__name__}' has no attribute '{name}'")

agno/eval/accuracy.py

@@ -282,7 +282,8 @@ Remember: You must only compare the agent_output to the expected_output. The exp
     ) -> Optional[AccuracyEvaluation]:
         """Orchestrate the evaluation process."""
         try:
-            accuracy_agent_response = evaluator_agent.run(evaluation_input).content
+            response = evaluator_agent.run(evaluation_input, stream=False)
+            accuracy_agent_response = response.content
             if accuracy_agent_response is None or not isinstance(accuracy_agent_response, AccuracyAgentResponse):
                 raise EvalError(f"Evaluator Agent returned an invalid response: {accuracy_agent_response}")
             return AccuracyEvaluation(
@@ -306,7 +307,7 @@ Remember: You must only compare the agent_output to the expected_output. The exp
     ) -> Optional[AccuracyEvaluation]:
         """Orchestrate the evaluation process asynchronously."""
         try:
-            response = await evaluator_agent.arun(evaluation_input)
+            response = await evaluator_agent.arun(evaluation_input, stream=False)
             accuracy_agent_response = response.content
             if accuracy_agent_response is None or not isinstance(accuracy_agent_response, AccuracyAgentResponse):
                 raise EvalError(f"Evaluator Agent returned an invalid response: {accuracy_agent_response}")
@@ -359,10 +360,14 @@ Remember: You must only compare the agent_output to the expected_output. The exp
                 status = Status(f"Running evaluation {i + 1}...", spinner="dots", speed=1.0, refresh_per_second=10)
                 live_log.update(status)
 
+                agent_session_id = f"eval_{self.eval_id}_{i + 1}"
+
                 if self.agent is not None:
-                    output = self.agent.run(input=eval_input).content
+                    agent_response = self.agent.run(input=eval_input, session_id=agent_session_id, stream=False)
+                    output = agent_response.content
                 elif self.team is not None:
-                    output = self.team.run(input=eval_input).content
+                    team_response = self.team.run(input=eval_input, session_id=agent_session_id, stream=False)
+                    output = team_response.content
 
                 if not output:
                     logger.error(f"Failed to generate a valid answer on iteration {i + 1}: {output}")
@@ -500,12 +505,14 @@ Remember: You must only compare the agent_output to the expected_output. The exp
                 status = Status(f"Running evaluation {i + 1}...", spinner="dots", speed=1.0, refresh_per_second=10)
                 live_log.update(status)
 
+                agent_session_id = f"eval_{self.eval_id}_{i + 1}"
+
                 if self.agent is not None:
-                    response = await self.agent.arun(input=eval_input)
-                    output = response.content
+                    agent_response = await self.agent.arun(input=eval_input, session_id=agent_session_id, stream=False)
+                    output = agent_response.content
                 elif self.team is not None:
-                    response = await self.team.arun(input=eval_input)  # type: ignore
-                    output = response.content
+                    team_response = await self.team.arun(input=eval_input, session_id=agent_session_id, stream=False)
+                    output = team_response.content
 
                 if not output:
                     logger.error(f"Failed to generate a valid answer on iteration {i + 1}: {output}")
@@ -612,11 +619,14 @@ Remember: You must only compare the agent_output to the expected_output. The exp
         print_results: bool = True,
     ) -> Optional[AccuracyResult]:
         """Run the evaluation logic against the given answer, instead of generating an answer with the Agent"""
+        # Generate unique run_id for this execution (don't modify self.eval_id due to concurrency)
+        run_id = str(uuid4())
+
         set_log_level_to_debug() if self.debug_mode else set_log_level_to_info()
 
         self.result = AccuracyResult()
 
-        logger.debug(f"************ Evaluation Start: {self.eval_id} ************")
+        logger.debug(f"************ Evaluation Start: {run_id} ************")
 
         evaluator_agent = self.get_evaluator_agent()
         eval_input = self.get_eval_input()
@@ -721,7 +731,7 @@ Remember: You must only compare the agent_output to the expected_output. The exp
                 ),
             )
 
-        logger.debug(f"*********** Evaluation End: {self.eval_id} ***********")
+        logger.debug(f"*********** Evaluation End: {run_id} ***********")
         return self.result
 
     async def arun_with_output(
@@ -732,11 +742,14 @@ Remember: You must only compare the agent_output to the expected_output. The exp
         print_results: bool = True,
     ) -> Optional[AccuracyResult]:
         """Run the evaluation logic against the given answer, instead of generating an answer with the Agent"""
+        # Generate unique run_id for this execution (don't modify self.eval_id due to concurrency)
+        run_id = str(uuid4())
+
         set_log_level_to_debug() if self.debug_mode else set_log_level_to_info()
 
         self.result = AccuracyResult()
 
-        logger.debug(f"************ Evaluation Start: {self.eval_id} ************")
+        logger.debug(f"************ Evaluation Start: {run_id} ************")
 
         evaluator_agent = self.get_evaluator_agent()
         eval_input = self.get_eval_input()
@@ -820,7 +833,7 @@ Remember: You must only compare the agent_output to the expected_output. The exp
                 eval_input=log_eval_input,
             )
 
-        logger.debug(f"*********** Evaluation End: {self.eval_id} ***********")
+        logger.debug(f"*********** Evaluation End: {run_id} ***********")
         return self.result
 
     def _get_telemetry_data(self) -> Dict[str, Any]: