agno 2.3.9__py3-none-any.whl → 2.3.11__py3-none-any.whl

agno/knowledge/utils.py

@@ -1,5 +1,6 @@
-from typing import Dict, List
+from typing import Any, Dict, List, Optional
 
+from agno.knowledge.reader.base import Reader
 from agno.knowledge.reader.reader_factory import ReaderFactory
 from agno.knowledge.types import ContentType
 from agno.utils.log import log_debug
@@ -75,8 +76,33 @@ def get_reader_info(reader_key: str) -> Dict:
         raise ValueError(f"Unknown reader: {reader_key}. Error: {str(e)}")
 
 
-def get_all_readers_info() -> List[Dict]:
-    """Get information about all available readers."""
+def get_reader_info_from_instance(reader: Reader, reader_id: str) -> Dict:
+    """Get information about a reader instance."""
+    try:
+        reader_class = reader.__class__
+        supported_strategies = reader_class.get_supported_chunking_strategies()
+        supported_content_types = reader_class.get_supported_content_types()
+
+        return {
+            "id": reader_id,
+            "name": getattr(reader, "name", reader_class.__name__),
+            "description": getattr(reader, "description", f"Custom {reader_class.__name__}"),
+            "chunking_strategies": [strategy.value for strategy in supported_strategies],
+            "content_types": [ct.value for ct in supported_content_types],
+        }
+    except Exception as e:
+        raise ValueError(f"Failed to get info for reader '{reader_id}': {str(e)}")
+
+
+def get_all_readers_info(knowledge_instance: Optional[Any] = None) -> List[Dict]:
+    """Get information about all available readers, including custom readers from a Knowledge instance.
+
+    Args:
+        knowledge_instance: Optional Knowledge instance to include custom readers from.
+
+    Returns:
+        List of reader info dictionaries.
+    """
     readers_info = []
     keys = ReaderFactory.get_all_reader_keys()
     for key in keys:
@@ -88,18 +114,35 @@ def get_all_readers_info() -> List[Dict]:
             # Log the error but don't fail the entire request
             log_debug(f"Skipping reader '{key}': {e}")
             continue
+
+    # Add custom readers from knowledge instance if provided
+    if knowledge_instance is not None:
+        custom_readers = knowledge_instance.get_readers()
+        if isinstance(custom_readers, dict):
+            for reader_id, reader in custom_readers.items():
+                try:
+                    reader_info = get_reader_info_from_instance(reader, reader_id)
+                    # Only add if not already present (custom readers take precedence)
+                    if not any(r["id"] == reader_id for r in readers_info):
+                        readers_info.append(reader_info)
+                except ValueError as e:
+                    log_debug(f"Skipping custom reader '{reader_id}': {e}")
+                    continue
+
     return readers_info
 
 
-def get_content_types_to_readers_mapping() -> Dict[str, List[str]]:
+def get_content_types_to_readers_mapping(knowledge_instance: Optional[Any] = None) -> Dict[str, List[str]]:
     """Get mapping of content types to list of reader IDs that support them.
 
+    Args:
+        knowledge_instance: Optional Knowledge instance to include custom readers from.
+
     Returns:
         Dictionary mapping content type strings (ContentType enum values) to list of reader IDs.
     """
     content_type_mapping: Dict[str, List[str]] = {}
-    readers_info = get_all_readers_info()
-
+    readers_info = get_all_readers_info(knowledge_instance)
     for reader_info in readers_info:
         reader_id = reader_info["id"]
         content_types = reader_info.get("content_types", [])
@@ -107,7 +150,9 @@ def get_content_types_to_readers_mapping() -> Dict[str, List[str]]:
         for content_type in content_types:
             if content_type not in content_type_mapping:
                 content_type_mapping[content_type] = []
-            content_type_mapping[content_type].append(reader_id)
+            # Avoid duplicates
+            if reader_id not in content_type_mapping[content_type]:
+                content_type_mapping[content_type].append(reader_id)
 
     return content_type_mapping
 

agno/models/base.py

@@ -1801,6 +1801,17 @@ class Model(ABC):
                 log_error(f"Error while iterating function result generator for {function_call.function.name}: {e}")
                 function_call.error = str(e)
                 function_call_success = False
+
+            # For generators, re-capture updated_session_state after consumption
+            # since session_state modifications were made during iteration
+            if function_execution_result.updated_session_state is None:
+                if (
+                    function_call.function._run_context is not None
+                    and function_call.function._run_context.session_state is not None
+                ):
+                    function_execution_result.updated_session_state = function_call.function._run_context.session_state
+                elif function_call.function._session_state is not None:
+                    function_execution_result.updated_session_state = function_call.function._session_state
         else:
             from agno.tools.function import ToolResult
 
@@ -2327,7 +2338,29 @@ class Model(ABC):
                     log_error(f"Error while iterating function result generator for {function_call.function.name}: {e}")
                     function_call.error = str(e)
                     function_call_success = False
-            else:
+
+            # For generators (sync or async), re-capture updated_session_state after consumption
+            # since session_state modifications were made during iteration
+            if async_function_call_output is not None or isinstance(
+                function_call.result,
+                (GeneratorType, collections.abc.Iterator, AsyncGeneratorType, collections.abc.AsyncIterator),
+            ):
+                if updated_session_state is None:
+                    if (
+                        function_call.function._run_context is not None
+                        and function_call.function._run_context.session_state is not None
+                    ):
+                        updated_session_state = function_call.function._run_context.session_state
+                    elif function_call.function._session_state is not None:
+                        updated_session_state = function_call.function._session_state
+
+            if not (
+                async_function_call_output is not None
+                or isinstance(
+                    function_call.result,
+                    (GeneratorType, collections.abc.Iterator, AsyncGeneratorType, collections.abc.AsyncIterator),
+                )
+            ):
                 from agno.tools.function import ToolResult
 
                 if isinstance(function_execution_result.result, ToolResult):

agno/models/google/gemini.py

@@ -36,6 +36,7 @@ try:
         GenerateContentResponseUsageMetadata,
         GoogleSearch,
         GoogleSearchRetrieval,
+        GroundingMetadata,
         Operation,
         Part,
         Retrieval,
@@ -244,8 +245,8 @@ class Gemini(Model):
         builtin_tools = []
 
         if self.grounding:
-            log_info(
-                "Grounding enabled. This is a legacy tool. For Gemini 2.0+ Please use enable `search` flag instead."
+            log_debug(
+                "Gemini Grounding enabled. This is a legacy tool. For Gemini 2.0+ Please use enable `search` flag instead."
             )
             builtin_tools.append(
                 Tool(
@@ -258,15 +259,15 @@ class Gemini(Model):
             )
 
         if self.search:
-            log_info("Google Search enabled.")
+            log_debug("Gemini Google Search enabled.")
             builtin_tools.append(Tool(google_search=GoogleSearch()))
 
         if self.url_context:
-            log_info("URL context enabled.")
+            log_debug("Gemini URL context enabled.")
             builtin_tools.append(Tool(url_context=UrlContext()))
 
         if self.vertexai_search:
-            log_info("Vertex AI Search enabled.")
+            log_debug("Gemini Vertex AI Search enabled.")
             if not self.vertexai_search_datastore:
                 log_error("vertexai_search_datastore must be provided when vertexai_search is enabled.")
                 raise ValueError("vertexai_search_datastore must be provided when vertexai_search is enabled.")
@@ -1008,27 +1009,24 @@ class Gemini(Model):
             citations = Citations()
             citations_raw = {}
             citations_urls = []
+            web_search_queries: List[str] = []
 
             if response.candidates and response.candidates[0].grounding_metadata is not None:
-                grounding_metadata = response.candidates[0].grounding_metadata.model_dump()
-                citations_raw["grounding_metadata"] = grounding_metadata
+                grounding_metadata: GroundingMetadata = response.candidates[0].grounding_metadata
+                citations_raw["grounding_metadata"] = grounding_metadata.model_dump()
 
-                chunks = grounding_metadata.get("grounding_chunks", []) or []
-                citation_pairs = []
+                chunks = grounding_metadata.grounding_chunks or []
+                web_search_queries = grounding_metadata.web_search_queries or []
                 for chunk in chunks:
-                    if not isinstance(chunk, dict):
+                    if not chunk:
                         continue
-                    web = chunk.get("web")
-                    if not isinstance(web, dict):
+                    web = chunk.web
+                    if not web:
                         continue
-                    uri = web.get("uri")
-                    title = web.get("title")
+                    uri = web.uri
+                    title = web.title
                     if uri:
-                        citation_pairs.append((uri, title))
-
-                # Create citation objects from filtered pairs
-                grounding_urls = [UrlCitation(url=url, title=title) for url, title in citation_pairs]
-                citations_urls.extend(grounding_urls)
+                        citations_urls.append(UrlCitation(url=uri, title=title))
 
             # Handle URLs from URL context tool
             if (
@@ -1036,22 +1034,29 @@ class Gemini(Model):
                 and hasattr(response.candidates[0], "url_context_metadata")
                 and response.candidates[0].url_context_metadata is not None
             ):
-                url_context_metadata = response.candidates[0].url_context_metadata.model_dump()
-                citations_raw["url_context_metadata"] = url_context_metadata
+                url_context_metadata = response.candidates[0].url_context_metadata
+                citations_raw["url_context_metadata"] = url_context_metadata.model_dump()
 
-                url_metadata_list = url_context_metadata.get("url_metadata", [])
+                url_metadata_list = url_context_metadata.url_metadata or []
                 for url_meta in url_metadata_list:
-                    retrieved_url = url_meta.get("retrieved_url")
-                    status = url_meta.get("url_retrieval_status", "UNKNOWN")
+                    retrieved_url = url_meta.retrieved_url
+                    status = "UNKNOWN"
+                    if url_meta.url_retrieval_status:
+                        status = url_meta.url_retrieval_status.value
                     if retrieved_url and status == "URL_RETRIEVAL_STATUS_SUCCESS":
                         # Avoid duplicate URLs
                         existing_urls = [citation.url for citation in citations_urls]
                         if retrieved_url not in existing_urls:
                             citations_urls.append(UrlCitation(url=retrieved_url, title=retrieved_url))
 
+            if citations_raw:
+                citations.raw = citations_raw
+            if citations_urls:
+                citations.urls = citations_urls
+            if web_search_queries:
+                citations.search_queries = web_search_queries
+
             if citations_raw or citations_urls:
-                citations.raw = citations_raw if citations_raw else None
-                citations.urls = citations_urls if citations_urls else None
                 model_response.citations = citations
 
         # Extract usage metadata if present
@@ -1150,28 +1155,52 @@ class Gemini(Model):
 
                         model_response.tool_calls.append(tool_call)
 
-            if response_delta.candidates[0].grounding_metadata is not None:
-                citations = Citations()
-                grounding_metadata = response_delta.candidates[0].grounding_metadata.model_dump()
-                citations.raw = grounding_metadata
+            citations = Citations()
+            citations.raw = {}
+            citations.urls = []
 
+            if (
+                hasattr(response_delta.candidates[0], "grounding_metadata")
+                and response_delta.candidates[0].grounding_metadata is not None
+            ):
+                grounding_metadata = response_delta.candidates[0].grounding_metadata
+                citations.raw["grounding_metadata"] = grounding_metadata.model_dump()
+                citations.search_queries = grounding_metadata.web_search_queries or []
                 # Extract url and title
-                chunks = grounding_metadata.pop("grounding_chunks", None) or []
-                citation_pairs = []
+                chunks = grounding_metadata.grounding_chunks or []
                 for chunk in chunks:
-                    if not isinstance(chunk, dict):
+                    if not chunk:
                         continue
-                    web = chunk.get("web")
-                    if not isinstance(web, dict):
+                    web = chunk.web
+                    if not web:
                         continue
-                    uri = web.get("uri")
-                    title = web.get("title")
+                    uri = web.uri
+                    title = web.title
                     if uri:
-                        citation_pairs.append((uri, title))
+                        citations.urls.append(UrlCitation(url=uri, title=title))
+
+            # Handle URLs from URL context tool
+            if (
+                hasattr(response_delta.candidates[0], "url_context_metadata")
+                and response_delta.candidates[0].url_context_metadata is not None
+            ):
+                url_context_metadata = response_delta.candidates[0].url_context_metadata
 
-                # Create citation objects from filtered pairs
-                citations.urls = [UrlCitation(url=url, title=title) for url, title in citation_pairs]
+                citations.raw["url_context_metadata"] = url_context_metadata.model_dump()
+
+                url_metadata_list = url_context_metadata.url_metadata or []
+                for url_meta in url_metadata_list:
+                    retrieved_url = url_meta.retrieved_url
+                    status = "UNKNOWN"
+                    if url_meta.url_retrieval_status:
+                        status = url_meta.url_retrieval_status.value
+                    if retrieved_url and status == "URL_RETRIEVAL_STATUS_SUCCESS":
+                        # Avoid duplicate URLs
+                        existing_urls = [citation.url for citation in citations.urls]
+                        if retrieved_url not in existing_urls:
+                            citations.urls.append(UrlCitation(url=retrieved_url, title=retrieved_url))
 
+            if citations.raw or citations.urls:
                 model_response.citations = citations
 
             # Extract usage metadata if present

agno/models/message.py

@@ -42,6 +42,9 @@ class Citations(BaseModel):
     # Raw citations from the model
     raw: Optional[Any] = None
 
+    # Search queries used to retrieve the citations
+    search_queries: Optional[List[str]] = None
+
     # URLs of the citations.
     urls: Optional[List[UrlCitation]] = None
 

agno/models/openai/chat.py

@@ -820,6 +820,16 @@ class OpenAIChat(Model):
         if response.usage is not None:
             model_response.response_usage = self._get_metrics(response.usage)
 
+        if model_response.provider_data is None:
+            model_response.provider_data = {}
+
+        if response.id:
+            model_response.provider_data["id"] = response.id
+        if response.system_fingerprint:
+            model_response.provider_data["system_fingerprint"] = response.system_fingerprint
+        if response.model_extra:
+            model_response.provider_data["model_extra"] = response.model_extra
+
         return model_response
 
     def _parse_provider_response_delta(self, response_delta: ChatCompletionChunk) -> ModelResponse:
@@ -842,6 +852,17 @@ class OpenAIChat(Model):
                 if choice_delta.content is not None:
                     model_response.content = choice_delta.content
 
+                    # We only want to handle these if content is present
+                    if model_response.provider_data is None:
+                        model_response.provider_data = {}
+
+                    if response_delta.id:
+                        model_response.provider_data["id"] = response_delta.id
+                    if response_delta.system_fingerprint:
+                        model_response.provider_data["system_fingerprint"] = response_delta.system_fingerprint
+                    if response_delta.model_extra:
+                        model_response.provider_data["model_extra"] = response_delta.model_extra
+
                 # Add tool calls
                 if choice_delta.tool_calls is not None:
                     model_response.tool_calls = choice_delta.tool_calls  # type: ignore

agno/os/routers/evals/utils.py

@@ -66,14 +66,14 @@ async def run_agent_as_judge_eval(
 
     # Run agent/team to get output
     if agent:
-        agent_response = await agent.arun(eval_run_input.input)
+        agent_response = await agent.arun(eval_run_input.input, stream=False)
         output = str(agent_response.content) if agent_response.content else ""
         model_id = agent.model.id if agent and agent.model else None
         model_provider = agent.model.provider if agent and agent.model else None
         agent_id = agent.id
         team_id = None
     elif team:
-        team_response = await team.arun(eval_run_input.input)
+        team_response = await team.arun(eval_run_input.input, stream=False)
         output = str(team_response.content) if team_response.content else ""
         model_id = team.model.id if team and team.model else None
         model_provider = team.model.provider if team and team.model else None
@@ -125,39 +125,21 @@ async def run_performance_eval(
     default_model: Optional[Model] = None,
 ) -> EvalSchema:
     """Run a performance evaluation for the given agent or team"""
-    # Create sync or async function based on DB type
-    if isinstance(db, AsyncBaseDb):
-        if agent:
-
-            async def run_component():  # type: ignore
-                return await agent.arun(eval_run_input.input)
-
-            model_id = agent.model.id if agent and agent.model else None
-            model_provider = agent.model.provider if agent and agent.model else None
-
-        elif team:
-
-            async def run_component():  # type: ignore
-                return await team.arun(eval_run_input.input)
-
-            model_id = team.model.id if team and team.model else None
-            model_provider = team.model.provider if team and team.model else None
-    else:
-        if agent:
+    if agent:
 
-            def run_component():  # type: ignore
-                return agent.run(eval_run_input.input)
+        async def run_component():  # type: ignore
+            return await agent.arun(eval_run_input.input, stream=False)
 
-            model_id = agent.model.id if agent and agent.model else None
-            model_provider = agent.model.provider if agent and agent.model else None
+        model_id = agent.model.id if agent and agent.model else None
+        model_provider = agent.model.provider if agent and agent.model else None
 
-        elif team:
+    elif team:
 
-            def run_component():
-                return team.run(eval_run_input.input)
+        async def run_component():  # type: ignore
+            return await team.arun(eval_run_input.input, stream=False)
 
-            model_id = team.model.id if team and team.model else None
-            model_provider = team.model.provider if team and team.model else None
+        model_id = team.model.id if team and team.model else None
+        model_provider = team.model.provider if team and team.model else None
 
     performance_eval = PerformanceEval(
         db=db,
@@ -171,11 +153,7 @@ async def run_performance_eval(
         model_provider=model_provider,
     )
 
-    # PerformanceEval needs sync/async check because it wraps a function
-    if isinstance(db, AsyncBaseDb):
-        result = await performance_eval.arun(print_results=False, print_summary=False)
-    else:
-        result = performance_eval.run(print_results=False, print_summary=False)
+    result = await performance_eval.arun(print_results=False, print_summary=False)
     if not result:
         raise HTTPException(status_code=500, detail="Failed to run performance evaluation")
 
@@ -210,7 +188,7 @@ async def run_reliability_eval(
         raise HTTPException(status_code=400, detail="expected_tool_calls is required for reliability evaluations")
 
     if agent:
-        agent_response = await agent.arun(eval_run_input.input)
+        agent_response = await agent.arun(eval_run_input.input, stream=False)
         reliability_eval = ReliabilityEval(
             db=db,
             name=eval_run_input.name,
@@ -221,7 +199,7 @@ async def run_reliability_eval(
         model_provider = agent.model.provider if agent and agent.model else None
 
     elif team:
-        team_response = await team.arun(eval_run_input.input)
+        team_response = await team.arun(eval_run_input.input, stream=False)
         reliability_eval = ReliabilityEval(
             db=db,
             name=eval_run_input.name,

agno/os/routers/knowledge/knowledge.py

@@ -1,7 +1,7 @@
 import json
 import logging
 import math
-from typing import Dict, List, Optional
+from typing import Any, Dict, List, Optional
 
 from fastapi import APIRouter, BackgroundTasks, Depends, File, Form, HTTPException, Path, Query, UploadFile
 
@@ -874,8 +874,8 @@ def attach_routes(router: APIRouter, knowledge_instances: List[Knowledge]) -> AP
     ) -> ConfigResponseSchema:
         knowledge = get_knowledge_instance_by_db_id(knowledge_instances, db_id)
 
-        # Get factory readers info
-        readers_info = get_all_readers_info()
+        # Get factory readers info (including custom readers from this knowledge instance)
+        readers_info = get_all_readers_info(knowledge)
         reader_schemas = {}
         # Add factory readers
         for reader_info in readers_info:
@@ -887,7 +887,13 @@ def attach_routes(router: APIRouter, knowledge_instances: List[Knowledge]) -> AP
             )
 
         # Add custom readers from knowledge.readers
-        readers_dict: Dict[str, Reader] = knowledge.get_readers() or {}
+        readers_result: Any = knowledge.get_readers() or {}
+        print(f"readers_result: {readers_result}")
+        # Ensure readers_dict is a dictionary (defensive check)
+        if not isinstance(readers_result, dict):
+            readers_dict: Dict[str, Reader] = {}
+        else:
+            readers_dict = readers_result
         if readers_dict:
             for reader_id, reader in readers_dict.items():
                 # Get chunking strategies from the reader
@@ -907,8 +913,8 @@ def attach_routes(router: APIRouter, knowledge_instances: List[Knowledge]) -> AP
                         chunkers=chunking_strategies,
                     )
 
-        # Get content types to readers mapping
-        types_of_readers = get_content_types_to_readers_mapping()
+        # Get content types to readers mapping (including custom readers from this knowledge instance)
+        types_of_readers = get_content_types_to_readers_mapping(knowledge)
         chunkers_list = get_all_chunkers_info()
 
         # Convert chunkers list to dictionary format expected by schema
@@ -961,20 +967,26 @@ async def process_content(
     try:
         if reader_id:
             reader = None
-            if knowledge.readers and reader_id in knowledge.readers:
-                reader = knowledge.readers[reader_id]
+            # Use get_readers() to ensure we get a dict (handles list conversion)
+            custom_readers = knowledge.get_readers()
+            if custom_readers and reader_id in custom_readers:
+                reader = custom_readers[reader_id]
+                log_debug(f"Found custom reader: {reader.__class__.__name__}")
             else:
+                # Try to resolve from factory readers
                 key = reader_id.lower().strip().replace("-", "_").replace(" ", "_")
                 candidates = [key] + ([key[:-6]] if key.endswith("reader") else [])
                 for cand in candidates:
                     try:
                         reader = ReaderFactory.create_reader(cand)
-                        log_debug(f"Resolved reader: {reader.__class__.__name__}")
+                        log_debug(f"Resolved reader from factory: {reader.__class__.__name__}")
                         break
                     except Exception:
                         continue
             if reader:
                 content.reader = reader
+            else:
+                log_debug(f"Could not resolve reader with id: {reader_id}")
         if chunker and content.reader:
             # Set the chunker name on the reader - let the reader handle it internally
             content.reader.set_chunking_strategy_from_string(chunker, chunk_size=chunk_size, overlap=chunk_overlap)

agno/team/team.py

@@ -1541,6 +1541,8 @@ class Team:
             add_history_to_context=add_history_to_context,
             add_session_state_to_context=add_session_state_to_context,
             add_dependencies_to_context=add_dependencies_to_context,
+            stream=False,
+            stream_events=False,
         )
 
         # 3. Prepare run messages
@@ -1753,6 +1755,8 @@ class Team:
             add_history_to_context=add_history_to_context,
             add_session_state_to_context=add_session_state_to_context,
             add_dependencies_to_context=add_dependencies_to_context,
+            stream=True,
+            stream_events=stream_events,
         )
 
         # 3. Prepare run messages
@@ -2182,9 +2186,6 @@ class Team:
         if stream_events is None:
             stream_events = False if self.stream_events is None else self.stream_events
 
-        self.stream = self.stream or stream
-        self.stream_events = self.stream_events or stream_events
-
         self.model = cast(Model, self.model)
 
         if self.metadata is not None:
@@ -2394,6 +2395,8 @@ class Team:
             add_history_to_context=add_history_to_context,
             add_dependencies_to_context=add_dependencies_to_context,
             add_session_state_to_context=add_session_state_to_context,
+            stream=False,
+            stream_events=False,
         )
 
         # 5. Prepare run messages
@@ -2638,6 +2641,10 @@ class Team:
             files=run_input.files,
             debug_mode=debug_mode,
             add_history_to_context=add_history_to_context,
+            add_dependencies_to_context=add_dependencies_to_context,
+            add_session_state_to_context=add_session_state_to_context,
+            stream=True,
+            stream_events=stream_events,
         )
 
         # 6. Prepare run messages
@@ -3046,9 +3053,6 @@ class Team:
         if stream_events is None:
             stream_events = False if self.stream_events is None else self.stream_events
 
-        self.stream = self.stream or stream
-        self.stream_events = self.stream_events or stream_events
-
         self.model = cast(Model, self.model)
 
         if self.metadata is not None:
@@ -5397,6 +5401,8 @@ class Team:
         add_history_to_context: Optional[bool] = None,
         add_dependencies_to_context: Optional[bool] = None,
         add_session_state_to_context: Optional[bool] = None,
+        stream: Optional[bool] = None,
+        stream_events: Optional[bool] = None,
         check_mcp_tools: bool = True,
     ) -> List[Union[Function, dict]]:
         # Connect tools that require connection management
@@ -5491,8 +5497,8 @@ class Team:
                 team_run_context=team_run_context,
                 input=user_message_content,
                 user_id=user_id,
-                stream=self.stream or False,
-                stream_events=self.stream_events or False,
+                stream=stream or False,
+                stream_events=stream_events or False,
                 async_mode=async_mode,
                 images=images,  # type: ignore
                 videos=videos,  # type: ignore