agno 2.3.13__py3-none-any.whl → 2.3.15__py3-none-any.whl

agno/db/migrations/manager.py

@@ -66,13 +66,13 @@ class MigrationManager:
                 current_version = packaging_version.parse(self.db.get_latest_schema_version(table_name))
 
             if current_version is None:
-                log_warning(f"Skipping up migration: No version found for table {table_name}.")
+                log_info(f"Skipping migration: No version found for table {table_name}.")
                 continue
 
             # If the target version is less or equal to the current version, no migrations needed
             if _target_version <= current_version and not force:
-                log_warning(
-                    f"Skipping up migration: the version of table '{table_name}' ({current_version}) is less or equal to the target version ({_target_version})."
+                log_info(
+                    f"Skipping migration: the version of table '{table_name}' ({current_version}) is less or equal to the target version ({_target_version})."
                 )
                 continue
 

agno/eval/__init__.py

@@ -1,12 +1,4 @@
-from agno.eval.accuracy import AccuracyAgentResponse, AccuracyEval, AccuracyEvaluation, AccuracyResult
-from agno.eval.agent_as_judge import (
-    AgentAsJudgeEval,
-    AgentAsJudgeEvaluation,
-    AgentAsJudgeResult,
-)
 from agno.eval.base import BaseEval
-from agno.eval.performance import PerformanceEval, PerformanceResult
-from agno.eval.reliability import ReliabilityEval, ReliabilityResult
 
 __all__ = [
     "AccuracyAgentResponse",
@@ -22,3 +14,24 @@ __all__ = [
     "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/knowledge/embedder/azure_openai.py

@@ -18,7 +18,6 @@ except ImportError:
 @dataclass
 class AzureOpenAIEmbedder(Embedder):
     id: str = "text-embedding-3-small"  # This has to match the model that you deployed at the provided URL
-
     dimensions: int = 1536
     encoding_format: Literal["float", "base64"] = "float"
     user: Optional[str] = None

agno/knowledge/embedder/google.py

@@ -15,7 +15,7 @@ except ImportError:
 
 @dataclass
 class GeminiEmbedder(Embedder):
-    id: str = "gemini-embedding-exp-03-07"
+    id: str = "gemini-embedding-001"
     task_type: str = "RETRIEVAL_QUERY"
     title: Optional[str] = None
     dimensions: Optional[int] = 1536

agno/models/anthropic/claude.py

@@ -102,8 +102,8 @@ class Claude(Model):
         # Claude Sonnet 4.x family (versions before 4.5)
         "claude-sonnet-4-20250514",
         "claude-sonnet-4",
-        # Claude Opus 4.x family (versions before 4.1)
-        # (Add any Opus 4.x models released before 4.1 if they exist)
+        # Claude Opus 4.x family (versions before 4.1 and 4.5)
+        # (Add any Opus 4.x models released before 4.1/4.5 if they exist)
     }
 
     id: str = "claude-sonnet-4-5-20250929"
@@ -191,7 +191,9 @@ class Claude(Model):
             return False
         if self.id.startswith("claude-sonnet-4-") and not self.id.startswith("claude-sonnet-4-5"):
             return False
-        if self.id.startswith("claude-opus-4-") and not self.id.startswith("claude-opus-4-1"):
+        if self.id.startswith("claude-opus-4-") and not (
+            self.id.startswith("claude-opus-4-1") or self.id.startswith("claude-opus-4-5")
+        ):
             return False
 
         return True
@@ -320,8 +322,11 @@ class Claude(Model):
 
             return {"type": "json_schema", "schema": schema}
 
-        # Handle dict format (already in correct structure)
+        # Handle dict format
         elif isinstance(response_format, dict):
+            # Claude only supports json_schema, not json_object
+            if response_format.get("type") == "json_object":
+                return None
             return response_format
 
         return None

agno/models/base.py

@@ -196,7 +196,8 @@ class Model(ABC):
                     )
                     sleep(delay)
                 else:
-                    log_error(f"Model provider error after {self.retries + 1} attempts: {e}")
+                    if self.retries > 0:
+                        log_error(f"Model provider error after {self.retries + 1} attempts: {e}")
             except RetryableModelProviderError as e:
                 current_count = retries_with_guidance_count
                 if current_count >= self.retry_with_guidance_limit:
@@ -238,7 +239,8 @@ class Model(ABC):
                     )
                     await asyncio.sleep(delay)
                 else:
-                    log_error(f"Model provider error after {self.retries + 1} attempts: {e}")
+                    if self.retries > 0:
+                        log_error(f"Model provider error after {self.retries + 1} attempts: {e}")
             except RetryableModelProviderError as e:
                 current_count = retries_with_guidance_count
                 if current_count >= self.retry_with_guidance_limit:
@@ -283,7 +285,8 @@ class Model(ABC):
                     )
                     sleep(delay)
                 else:
-                    log_error(f"Model provider error after {self.retries + 1} attempts: {e}")
+                    if self.retries > 0:
+                        log_error(f"Model provider error after {self.retries + 1} attempts: {e}")
             except RetryableModelProviderError as e:
                 current_count = retries_with_guidance_count
                 if current_count >= self.retry_with_guidance_limit:
@@ -330,7 +333,8 @@ class Model(ABC):
                     )
                     await asyncio.sleep(delay)
                 else:
-                    log_error(f"Model provider error after {self.retries + 1} attempts: {e}")
+                    if self.retries > 0:
+                        log_error(f"Model provider error after {self.retries + 1} attempts: {e}")
             except RetryableModelProviderError as e:
                 current_count = retries_with_guidance_count
                 if current_count >= self.retry_with_guidance_limit:

agno/models/metrics.py

@@ -13,6 +13,10 @@ class Metrics:
     output_tokens: int = 0
     total_tokens: int = 0
 
+    # Cost of the run
+    # Currently only supported by some providers
+    cost: Optional[float] = None
+
     # Audio token usage
     audio_input_tokens: int = 0
     audio_output_tokens: int = 0
@@ -43,6 +47,7 @@ class Metrics:
         metrics_dict = asdict(self)
         # Remove the timer util if present
         metrics_dict.pop("timer", None)
+        # Remove any None, 0, or empty dict values
         metrics_dict = {
             k: v
             for k, v in metrics_dict.items()
@@ -65,6 +70,13 @@ class Metrics:
             reasoning_tokens=self.reasoning_tokens + other.reasoning_tokens,
         )
 
+        if self.cost is not None and other.cost is not None:
+            result.cost = self.cost + other.cost
+        elif self.cost is not None:
+            result.cost = self.cost
+        elif other.cost is not None:
+            result.cost = other.cost
+
         # Handle provider_metrics
         if self.provider_metrics or other.provider_metrics:
             result.provider_metrics = {}

agno/models/openai/chat.py

@@ -945,4 +945,6 @@ class OpenAIChat(Model):
             metrics.audio_output_tokens = completion_tokens_details.audio_tokens or 0
             metrics.reasoning_tokens = completion_tokens_details.reasoning_tokens or 0
 
+        metrics.cost = getattr(response_usage, "cost", None)
+
         return metrics

agno/models/openai/responses.py

@@ -234,8 +234,8 @@ class OpenAIResponses(Model):
                     "strict": self.strict_output,
                 }
             else:
-                # JSON mode
-                text_params["format"] = {"type": "json_object"}
+                # Pass through directly, user handles everything
+                text_params["format"] = response_format
 
         # Add text parameter if there are any text-level params
         if text_params:

agno/os/app.py

@@ -34,6 +34,7 @@ from agno.os.config import (
 from agno.os.interfaces.base import BaseInterface
 from agno.os.router import get_base_router
 from agno.os.routers.agents import get_agent_router
+from agno.os.routers.database import get_database_router
 from agno.os.routers.evals import get_eval_router
 from agno.os.routers.health import get_health_router
 from agno.os.routers.home import get_home_router
@@ -225,6 +226,9 @@ class AgentOS:
         self._initialize_teams()
         self._initialize_workflows()
 
+        # Check for duplicate IDs
+        self._raise_if_duplicate_ids()
+
         if self.tracing:
             self._setup_tracing()
 
@@ -266,6 +270,9 @@ class AgentOS:
         self._initialize_agents()
         self._initialize_teams()
         self._initialize_workflows()
+
+        # Check for duplicate IDs
+        self._raise_if_duplicate_ids()
         self._auto_discover_databases()
         self._auto_discover_knowledge_instances()
 
@@ -334,6 +341,31 @@ class AgentOS:
             self.interfaces.append(a2a_interface)
             self._add_router(app, a2a_interface.get_router())
 
+    def _raise_if_duplicate_ids(self) -> None:
+        """Check for duplicate IDs within each entity type.
+
+        Raises:
+            ValueError: If duplicate IDs are found within the same entity type
+        """
+        duplicate_ids: List[str] = []
+
+        for entities in [self.agents, self.teams, self.workflows]:
+            if not entities:
+                continue
+            seen_ids: set[str] = set()
+            for entity in entities:
+                entity_id = entity.id
+                if entity_id is None:
+                    continue
+                if entity_id in seen_ids:
+                    if entity_id not in duplicate_ids:
+                        duplicate_ids.append(entity_id)
+                else:
+                    seen_ids.add(entity_id)
+
+        if duplicate_ids:
+            raise ValueError(f"Duplicate IDs found in AgentOS: {', '.join(repr(id_) for id_ in duplicate_ids)}")
+
     def _make_app(self, lifespan: Optional[Any] = None) -> FastAPI:
         # Adjust the FastAPI app lifespan to handle MCP connections if relevant
         app_lifespan = lifespan
@@ -543,6 +575,7 @@ class AgentOS:
             get_metrics_router(dbs=self.dbs),
             get_knowledge_router(knowledge_instances=self.knowledge_instances),
             get_traces_router(dbs=self.dbs),
+            get_database_router(self, settings=self.settings),
         ]
 
         for router in routers:
@@ -583,6 +616,18 @@ class AgentOS:
 
         # Add JWT middleware if authorization is enabled
         if self.authorization:
+            # Set authorization_enabled flag on settings so security key validation is skipped
+            self.settings.authorization_enabled = True
+
+            jwt_configured = bool(getenv("JWT_VERIFICATION_KEY") or getenv("JWT_JWKS_FILE"))
+            security_key_set = bool(self.settings.os_security_key)
+            if jwt_configured and security_key_set:
+                log_warning(
+                    "Both JWT configuration (JWT_VERIFICATION_KEY or JWT_JWKS_FILE) and OS_SECURITY_KEY are set. "
+                    "With authorization=True, only JWT authorization will be used. "
+                    "Consider removing OS_SECURITY_KEY from your environment."
+                )
+
             self._add_jwt_middleware(fastapi_app)
 
         return fastapi_app
@@ -981,6 +1026,8 @@ class AgentOS:
         host: str = "localhost",
         port: int = 7777,
         reload: bool = False,
+        reload_includes: Optional[List[str]] = None,
+        reload_excludes: Optional[List[str]] = None,
         workers: Optional[int] = None,
         access_log: bool = False,
         **kwargs,
@@ -1000,8 +1047,12 @@ class AgentOS:
             Align.center(f"[bold cyan]{public_endpoint}[/bold cyan]"),
             Align.center(f"\n\n[bold dark_orange]OS running on:[/bold dark_orange] http://{host}:{port}"),
         ]
-        if bool(self.settings.os_security_key):
-            panel_group.append(Align.center("\n\n[bold chartreuse3]:lock: Security Enabled[/bold chartreuse3]"))
+        if self.authorization:
+            panel_group.append(
+                Align.center("\n\n[bold chartreuse3]:lock: JWT Authorization Enabled[/bold chartreuse3]")
+            )
+        elif bool(self.settings.os_security_key):
+            panel_group.append(Align.center("\n\n[bold chartreuse3]:lock: Security Key Enabled[/bold chartreuse3]"))
 
         console = Console()
         console.print(
@@ -1015,11 +1066,17 @@ class AgentOS:
             )
         )
 
+        # Adding *.yaml to reload_includes to reload the app when the yaml config file changes.
+        if reload and reload_includes is not None:
+            reload_includes = ["*.yaml", "*.yml"]
+
         uvicorn.run(
             app=app,
             host=host,
             port=port,
             reload=reload,
+            reload_includes=reload_includes,
+            reload_excludes=reload_excludes,
             workers=workers,
             access_log=access_log,
             lifespan="on",

agno/os/auth.py

@@ -1,3 +1,4 @@
+from os import getenv
 from typing import List, Set
 
 from fastapi import Depends, HTTPException, Request
@@ -10,18 +11,43 @@ from agno.os.settings import AgnoAPISettings
 security = HTTPBearer(auto_error=False)
 
 
+def _is_jwt_configured() -> bool:
+    """Check if JWT authentication is configured via environment variables.
+
+    This covers cases where JWT middleware is set up manually (not via authorization=True).
+    """
+    return bool(getenv("JWT_VERIFICATION_KEY") or getenv("JWT_JWKS_FILE"))
+
+
 def get_authentication_dependency(settings: AgnoAPISettings):
     """
     Create an authentication dependency function for FastAPI routes.
 
+    This handles security key authentication (OS_SECURITY_KEY).
+    When JWT authorization is enabled (via authorization=True, JWT environment variables,
+    or manually added JWT middleware), this dependency is skipped as JWT middleware
+    handles authentication.
+
     Args:
-        settings: The API settings containing the security key
+        settings: The API settings containing the security key and authorization flag
 
     Returns:
         A dependency function that can be used with FastAPI's Depends()
     """
 
-    async def auth_dependency(credentials: HTTPAuthorizationCredentials = Depends(security)) -> bool:
+    async def auth_dependency(request: Request, credentials: HTTPAuthorizationCredentials = Depends(security)) -> bool:
+        # If JWT authorization is enabled via settings (authorization=True on AgentOS)
+        if settings and settings.authorization_enabled:
+            return True
+
+        # Check if JWT middleware has already handled authentication
+        if getattr(request.state, "authenticated", False):
+            return True
+
+        # Also skip if JWT is configured via environment variables
+        if _is_jwt_configured():
+            return True
+
         # If no security key is set, skip authentication entirely
         if not settings or not settings.os_security_key:
             return True
@@ -45,13 +71,24 @@ def validate_websocket_token(token: str, settings: AgnoAPISettings) -> bool:
     """
     Validate a bearer token for WebSocket authentication (legacy os_security_key method).
 
+    When JWT authorization is enabled (via authorization=True or JWT environment variables),
+    this validation is skipped as JWT middleware handles authentication.
+
     Args:
         token: The bearer token to validate
-        settings: The API settings containing the security key
+        settings: The API settings containing the security key and authorization flag
 
     Returns:
         True if the token is valid or authentication is disabled, False otherwise
     """
+    # If JWT authorization is enabled, skip security key validation
+    if settings and settings.authorization_enabled:
+        return True
+
+    # Also skip if JWT is configured via environment variables (manual JWT middleware setup)
+    if _is_jwt_configured():
+        return True
+
     # If no security key is set, skip authentication entirely
     if not settings or not settings.os_security_key:
         return True