Release 1.4.6 (#3154)

# Changelog

## New Features:

- **Cerebras Model Provider**: Added Cerebras as a model provider.
- **Claude Web Search**: Added support for [Claude’s new web search
tool](https://www.anthropic.com/news/web-search)
- **Knowledge Base Metadata Filtering (Beta)**: Added support for
filtering documents by metadata
    - **Two Ways to Apply Filters**:
- **Explicit Filtering**: Pass filters directly to Agent or during
run/query
            
            ```python
            # Option 1: Filters on Agent initialization
            agent = Agent(
            					knowledge=knowledge_base, 
            					knowledge_filters={"filter_1": "abc"}
            				)
                 
            # Option 2: Filters on run execution
agent.run("Tell me about...", knowledge_filters={"filter_1": "abc"})
            ```
            
See docs
[here](https://github.com/agno-agi/agno/blob/main/cookbook/agent_concepts/knowledge/filters/pdf/filtering.py)
            
- **Agentic Filtering**: Agent automatically detects and applies filters
from user queries
            
            ```python
            # Enable automatic filter detection
            agent = Agent(
            					knowledge=knowledge_base, 
            					enable_agentic_knowledge_filters=True
            				)
                 
            # Agent extracts filters from query
            agent.run("Tell me about John Doe's experience...")
            ```
            
See docs
[here](https://github.com/agno-agi/agno/blob/main/cookbook/agent_concepts/knowledge/filters/pdf/agentic_filtering.py)
            
    - Two approaches for adding metadata to documents:
        1. **During Knowledge Base Initialization**:
            
            ```python
            knowledge_base = PDFKnowledgeBase(path=[
                 {
            		     "path": "file_1.pdf", 
            		     "metadata": {
            				     "user_id": "abc"
            				  }
            		 },
            		 {
            		     "path": "file_2.pdf", 
            		     "metadata": {
            				     "user_id": "xyz"
            				  }
            		 }
            ])
            ```
            
        2. **During Individual Document Loading:**
            
            ```python
            knowledge_base.load_document(
                 path="file.pdf",
                 metadata={"user_id": "abc"}
            )
            ```
            
    - **Compatibility**
- **Knowledge Base Types**: `PDF`, `Text`, `DOCX`, `JSON`, and `PDF_URL`
        - **Vector Databases**: `Qdrant`, `LanceDB`, and `MongoDB`

## Improvements:

- **User and Session ID in Tools**: Added `current_user_id` and
`current_session_id` as default variables in

## Bug Fixes:

- **Knowledge Base ID Clashes**: Knowledge files with overlapping
names (e.g., `abc.-.xyz.pdf` and `abc.-.def.pdf`)
were being incorrectly identified due to the readers using formatted
names as unique id which were getting uniqueness conflict. Introduced a
unique ID for each document in all the readers using `uuidv4()` to
ensure strict identification and prevent conflicts.

Author: dirkbrnd

libs/agno/agno/agent/agent.py

@@ -1922,8 +1922,10 @@ def create_run_response(
 
     def _initialize_session_state(self, user_id: Optional[str] = None, session_id: Optional[str] = None) -> None:
         self.session_state = self.session_state or {}
-        self.session_state["current_user_id"] = user_id
-        self.session_state["current_session_id"] = session_id
+        if user_id is not None:
+            self.session_state["current_user_id"] = user_id
+        if session_id is not None:
+            self.session_state["current_session_id"] = session_id
 
     def _make_memories_and_summaries(
         self,

libs/agno/agno/document/reader/base.py

@@ -18,7 +18,7 @@ class Reader:
 
     def __init__(self, chunk_size: int = 5000, chunking_strategy: Optional[ChunkingStrategy] = None) -> None:
         self.chunk_size = chunk_size
-        self.chunking_strategy = chunking_strategy or FixedSizeChunking(chunk_size=self.chunk_size)
+        self.chunking_strategy = chunking_strategy
 
     def read(self, obj: Any) -> List[Document]:
         raise NotImplementedError
@@ -27,6 +27,8 @@ async def async_read(self, obj: Any) -> List[Document]:
         raise NotImplementedError
 
     def chunk_document(self, document: Document) -> List[Document]:
+        if self.chunking_strategy is None:
+            self.chunking_strategy = FixedSizeChunking(chunk_size=self.chunk_size)
         return self.chunking_strategy.chunk(document)  # type: ignore
 
     async def chunk_documents_async(self, documents: List[Document]) -> List[Document]:

libs/agno/agno/document/reader/csv_reader.py

@@ -30,7 +30,7 @@ def read(self, file: Union[Path, IO[Any]], delimiter: str = ",", quotechar: str
                 logger.info(f"Reading: {file}")
                 file_content = file.open(newline="", mode="r", encoding="utf-8")
             else:
-                logger.info(f"Reading uploaded file: {file.name}")
+                logger.info(f"Reading retrieved file: {file.name}")
                 file.seek(0)
                 file_content = io.StringIO(file.read().decode("utf-8"))  # type: ignore
 
@@ -82,7 +82,7 @@ async def async_read(
                     content = await file_content.read()
                     file_content_io = io.StringIO(content)
             else:
-                logger.info(f"Reading uploaded file async: {file.name}")
+                logger.info(f"Reading retrieved file async: {file.name}")
                 file.seek(0)
                 file_content_io = io.StringIO(file.read().decode("utf-8"))  # type: ignore
 
@@ -152,7 +152,6 @@ def read(self, url: str) -> List[Document]:
 
         file_obj = io.BytesIO(response.content)
         file_obj.name = filename
-
         documents = CSVReader().read(file=file_obj)
 
         file_obj.close()

libs/agno/agno/models/anthropic/claude.py

@@ -144,10 +144,12 @@ def _format_tools_for_model(self) -> Optional[List[Dict[str, Any]]]:
             return None
 
         tools: List[Dict[str, Any]] = []
-        for func_def in self._tools:
-            if func_def.get("type") != "function":
-                tools.append(func_def)
+        for tool_def in self._tools:
+            if tool_def.get("type", "") != "function":
+                tools.append(tool_def)
                 continue
+
+            func_def = tool_def.get("function", {})
             parameters: Dict[str, Any] = func_def.get("parameters", {})
             properties: Dict[str, Any] = parameters.get("properties", {})
             required_params: List[str] = []
@@ -502,6 +504,7 @@ def parse_provider_response_delta(
 
         # Capture citations from the final response
         elif isinstance(response, MessageStopEvent):
+            model_response.content = ""
             model_response.citations = Citations(raw=[], urls=[], documents=[])
             for block in response.message.content:
                 citations = getattr(block, "citations", None)
@@ -517,9 +520,6 @@ def parse_provider_response_delta(
                         model_response.citations.documents.append(  # type: ignore
                             DocumentCitation(document_title=citation.document_title, cited_text=citation.cited_text)
                         )
-
-        # Handle message completion and usage metrics
-        elif isinstance(response, MessageStopEvent):
             if response.message.usage is not None:
                 model_response.response_usage = response.message.usage
 

libs/agno/agno/team/team.py

@@ -1996,8 +1996,10 @@ async def _arun_stream(
 
     def _initialize_session_state(self, user_id: Optional[str] = None, session_id: Optional[str] = None) -> None:
         self.session_state = self.session_state or {}
-        self.session_state["current_user_id"] = user_id
-        self.session_state["current_session_id"] = session_id
+        if user_id is not None:
+            self.session_state["current_user_id"] = user_id
+        if session_id is not None:
+            self.session_state["current_session_id"] = session_id
 
     def _make_memories_and_summaries(
         self, run_messages: RunMessages, session_id: str, user_id: Optional[str] = None

libs/agno/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agno"
-version = "1.4.5"
+version = "1.4.6"
 description = "Agno: a lightweight library for building Reasoning Agents"
 requires-python = ">=3.7,<4"
 readme = "README.md"

libs/agno/tests/integration/agent/test_metrics.py

@@ -31,7 +31,7 @@ def test_session_metrics():
     response = agent.run("What is current news in France?")
 
     input_tokens_list = response.metrics.get("input_tokens", [])
-    assert len(input_tokens_list) == 2  # Should be 2 assistant messages
+    assert len(input_tokens_list) >= 2  # Should be atleast 2 assistant messages
 
     input_tokens += sum(response.metrics.get("input_tokens", []))
     output_tokens += sum(response.metrics.get("output_tokens", []))

libs/agno/tests/integration/agent/test_reasoning_content_knowledge_tools.py

@@ -256,7 +256,7 @@ def test_docx_knowledge_base_with_metadata_path_invalid_filter(setup_vector_db):
     assert len(response_content) > 50
 
     clarification_phrases = [
-        "specify which",
+        "specify",
         "which candidate",
         "please clarify",
         "need more information",
@@ -321,7 +321,7 @@ async def test_async_docx_knowledge_base_with_metadata_path_invalid_filter(setup
     assert len(response_content) > 50
 
     clarification_phrases = [
-        "specify which",
+        "specify",
         "which candidate",
         "please clarify",
         "need more information",
@@ -550,7 +550,7 @@ async def test_async_knowledge_base_with_invalid_filter(setup_vector_db):
     assert len(response_content) > 50
 
     clarification_phrases = [
-        "specify which",
+        "specify",
         "which candidate",
         "please clarify",
         "need more information",

libs/agno/tests/integration/knowledge/test_docx_knowledge_base.py

@@ -276,7 +276,7 @@ def test_docx_knowledge_base_with_metadata_path_invalid_filter(setup_vector_db):
     assert len(response_content) > 50
 
     clarification_phrases = [
-        "specify which",
+        "specify",
         "which candidate",
         "please clarify",
         "need more information",
@@ -340,7 +340,7 @@ async def test_async_docx_knowledge_base_with_metadata_path_invalid_filter(setup
     assert len(response_content) > 50
 
     clarification_phrases = [
-        "specify which",
+        "specify",
         "which candidate",
         "please clarify",
         "need more information",
@@ -437,7 +437,7 @@ def test_knowledge_base_with_invalid_filter(setup_vector_db):
     assert len(response_content) > 50
 
     clarification_phrases = [
-        "specify which",
+        "specify",
         "which candidate",
         "please clarify",
         "need more information",
@@ -569,7 +569,7 @@ async def test_async_knowledge_base_with_invalid_filter(setup_vector_db):
     assert len(response_content) > 50
 
     clarification_phrases = [
-        "specify which",
+        "specify",
         "which candidate",
         "please clarify",
         "need more information",