alita-sdk 0.3.395__py3-none-any.whl → 0.3.397__py3-none-any.whl

alita_sdk/runtime/langchain/document_loaders/AlitaDocxMammothLoader.py

@@ -1,4 +1,5 @@
 import re
+import uuid
 from io import BytesIO
 
 import mammoth.images
@@ -8,6 +9,9 @@ from langchain_core.document_loaders import BaseLoader
 from langchain_core.documents import Document
 from mammoth import convert_to_html
 from markdownify import markdownify
+from docx import Document as DocxDocument
+from docx.oxml.ns import qn
+from bs4 import BeautifulSoup
 
 from alita_sdk.tools.chunkers.sematic.markdown_chunker import markdown_by_headers_chunker
 from .utils import perform_llm_prediction_for_image_bytes
@@ -17,6 +21,7 @@ class AlitaDocxMammothLoader(BaseLoader):
     """
     Loader for Docx files using Mammoth to convert to HTML, with image handling,
     and then Markdownify to convert HTML to markdown.
+    Detects bordered paragraphs and text boxes and treats them as code blocks.
     """
     def __init__(self, **kwargs):
         """
@@ -97,6 +102,295 @@ class AlitaDocxMammothLoader(BaseLoader):
         new_md = pattern.sub(replace_placeholder, original_md)
         return new_md
 
+    def __has_border(self, paragraph):
+        """
+        Check if a paragraph has border formatting.
+        
+        Args:
+            paragraph: A python-docx Paragraph object.
+            
+        Returns:
+            bool: True if paragraph has any border, False otherwise.
+        """
+        pPr = paragraph._element.pPr
+        if pPr is not None:
+            pBdr = pPr.find(qn('w:pBdr'))
+            if pBdr is not None:
+                # Check if any border side exists (top, bottom, left, right)
+                for side in ['top', 'bottom', 'left', 'right']:
+                    border = pBdr.find(qn(f'w:{side}'))
+                    if border is not None:
+                        # Check if border is not "none" or has a width
+                        val = border.get(qn('w:val'))
+                        if val and val != 'none':
+                            return True
+        return False
+
+    def __find_text_boxes(self, doc):
+        """
+        Find all text boxes in document by searching OOXML structure.
+        Text boxes are typically in w:txbxContent elements.
+        
+        Args:
+            doc: A python-docx Document object.
+            
+        Returns:
+            list: List of tuples (element, paragraphs_inside_textbox).
+        """
+        text_boxes = []
+        
+        # Iterate through document body XML to find text box content elements
+        for element in doc.element.body.iter():
+            # Look for text box content elements
+            if element.tag.endswith('txbxContent'):
+                # Collect all paragraphs inside this text box
+                txbx_paragraphs = []
+                for txbx_para_element in element.iter():
+                    if txbx_para_element.tag.endswith('p'):
+                        txbx_paragraphs.append(txbx_para_element)
+                
+                if txbx_paragraphs:
+                    text_boxes.append((element, txbx_paragraphs))
+        
+        return text_boxes
+
+    def __create_marker_paragraph(self, marker_text):
+        """
+        Create a paragraph element with marker text.
+        
+        Args:
+            marker_text (str): The marker text to insert.
+            
+        Returns:
+            Element: An OOXML paragraph element.
+        """
+        from docx.oxml import OxmlElement
+        
+        p = OxmlElement('w:p')
+        r = OxmlElement('w:r')
+        t = OxmlElement('w:t')
+        t.text = marker_text
+        r.append(t)
+        p.append(r)
+        return p
+
+    def __inject_markers_for_paragraph(self, paragraph, start_marker, end_marker):
+        """
+        Inject marker paragraphs before and after a bordered paragraph.
+        
+        Args:
+            paragraph: A python-docx Paragraph object.
+            start_marker (str): The start marker text.
+            end_marker (str): The end marker text.
+        """
+        # Insert start marker paragraph before
+        marker_p_start = self.__create_marker_paragraph(start_marker)
+        paragraph._element.addprevious(marker_p_start)
+        
+        # Insert end marker paragraph after
+        marker_p_end = self.__create_marker_paragraph(end_marker)
+        paragraph._element.addnext(marker_p_end)
+
+    def __inject_markers_for_textbox(self, textbox_element, paragraph_elements, start_marker, end_marker):
+        """
+        Inject markers around text box content.
+        
+        Args:
+            textbox_element: The w:txbxContent element.
+            paragraph_elements: List of paragraph elements inside the text box.
+            start_marker (str): The start marker text.
+            end_marker (str): The end marker text.
+        """
+        if not paragraph_elements:
+            return
+        
+        # Insert start marker before first paragraph in text box
+        first_para = paragraph_elements[0]
+        marker_p_start = self.__create_marker_paragraph(start_marker)
+        first_para.addprevious(marker_p_start)
+        
+        # Insert end marker after last paragraph in text box
+        last_para = paragraph_elements[-1]
+        marker_p_end = self.__create_marker_paragraph(end_marker)
+        last_para.addnext(marker_p_end)
+
+    def __detect_and_mark_bordered_content(self, docx_stream):
+        """
+        Detects bordered paragraphs and text boxes, injects unique markers around them.
+        Groups consecutive bordered paragraphs into single code blocks.
+        
+        Args:
+            docx_stream: A file-like object containing the DOCX document.
+            
+        Returns:
+            tuple: (modified_docx_stream, start_marker, end_marker)
+        """
+        # Load document with python-docx
+        doc = DocxDocument(docx_stream)
+        
+        # Generate unique markers to avoid conflicts with document content
+        unique_id = uuid.uuid4().hex[:8]
+        start_marker = f"<<<BORDERED_BLOCK_START_{unique_id}>>>"
+        end_marker = f"<<<BORDERED_BLOCK_END_{unique_id}>>>"
+        
+        # Group consecutive bordered paragraphs together
+        bordered_groups = []
+        current_group = []
+        
+        for para in doc.paragraphs:
+            if self.__has_border(para):
+                current_group.append(para)
+            else:
+                if current_group:
+                    # End of a bordered group
+                    bordered_groups.append(current_group)
+                    current_group = []
+        
+        # Don't forget the last group if document ends with bordered paragraphs
+        if current_group:
+            bordered_groups.append(current_group)
+        
+        # Collect all text boxes
+        # text_boxes = self.__find_text_boxes(doc)
+        
+        # Inject markers around each group of consecutive bordered paragraphs
+        for group in bordered_groups:
+            if group:
+                # Add start marker before first paragraph in group
+                first_para = group[0]
+                marker_p_start = self.__create_marker_paragraph(start_marker)
+                first_para._element.addprevious(marker_p_start)
+                
+                # Add end marker after last paragraph in group
+                last_para = group[-1]
+                marker_p_end = self.__create_marker_paragraph(end_marker)
+                last_para._element.addnext(marker_p_end)
+        
+        # Inject markers around text box content
+        # for textbox_element, para_elements in text_boxes:
+        #     self.__inject_markers_for_textbox(textbox_element, para_elements, start_marker, end_marker)
+        
+        # Save modified document to BytesIO
+        output = BytesIO()
+        doc.save(output)
+        output.seek(0)
+        
+        return output, start_marker, end_marker
+
+    def __contains_complex_structure(self, content_html):
+        """
+        Check if HTML content contains tables, lists, or other complex structures.
+        
+        Args:
+            content_html (str): HTML content to analyze.
+            
+        Returns:
+            bool: True if content contains tables/lists, False otherwise.
+        """
+        content_soup = BeautifulSoup(content_html, 'html.parser')
+        
+        # Check for tables
+        if content_soup.find('table'):
+            return True
+        
+        # Check for lists (ul, ol)
+        if content_soup.find('ul') or content_soup.find('ol'):
+            return True
+        
+        return False
+
+    def __escape_hash_symbols(self, html_content):
+        """
+        Escape hash (#) symbols at the beginning of lines in HTML to prevent
+        them from being treated as markdown headers.
+        
+        Args:
+            html_content (str): HTML content.
+            
+        Returns:
+            str: HTML with escaped hash symbols.
+        """
+        soup = BeautifulSoup(html_content, 'html.parser')
+        
+        # Process all text-containing elements
+        for element in soup.find_all(['p', 'li', 'td', 'th', 'h1', 'h2', 'h3', 'h4', 'h5', 'h6']):
+            if element.string:
+                text = element.string
+                # If line starts with #, escape it
+                if text.strip().startswith('#'):
+                    element.string = text.replace('#', '\\#', 1)
+        
+        return str(soup)
+
+    def __wrap_marked_sections_in_code_blocks(self, html, start_marker, end_marker):
+        """
+        Find content between markers and wrap appropriately:
+        - Simple text/code → <pre><code> block
+        - Tables/lists → Custom wrapper with preserved structure
+        
+        Args:
+            html (str): The HTML content from Mammoth.
+            start_marker (str): The start marker text.
+            end_marker (str): The end marker text.
+            
+        Returns:
+            str: HTML with marked sections wrapped appropriately.
+        """
+        import html as html_module
+        
+        # Mammoth escapes < and > to &lt; and &gt;, so we need to escape our markers too
+        escaped_start = html_module.escape(start_marker)
+        escaped_end = html_module.escape(end_marker)
+        
+        # Pattern to find content between HTML-escaped markers (including HTML tags)
+        # The markers will be in separate <p> tags, and content in between
+        pattern = re.compile(
+            f'<p>{re.escape(escaped_start)}</p>(.*?)<p>{re.escape(escaped_end)}</p>',
+            re.DOTALL
+        )
+        
+        def replace_with_appropriate_wrapper(match):
+            content = match.group(1)
+            
+            # Detect if content has complex structure (tables, lists)
+            has_complex_structure = self.__contains_complex_structure(content)
+            
+            if has_complex_structure:
+                # Preserve structure: keep HTML as-is, escape # symbols
+                escaped_content = self.__escape_hash_symbols(content)
+                # Wrap in a div with special class for potential custom handling
+                return f'<div class="alita-bordered-content">{escaped_content}</div>'
+            else:
+                # Simple text/code: extract as plain text and wrap in code block
+                content_soup = BeautifulSoup(content, 'html.parser')
+                
+                # Extract text from each paragraph separately to preserve line breaks
+                lines = []
+                for element in content_soup.find_all(['p', 'h1', 'h2', 'h3', 'h4', 'h5', 'h6']):
+                    # Replace <br /> within paragraphs with newlines
+                    for br in element.find_all('br'):
+                        br.replace_with('\n')
+                    text = element.get_text()
+                    # Preserve leading whitespace (indentation), only strip trailing
+                    lines.append(text.rstrip())
+                
+                # If no paragraphs found, just get all text
+                if not lines:
+                    content = content.replace('<br />', '\n').replace('<br/>', '\n').replace('<br>', '\n')
+                    content_text = content_soup.get_text()
+                    lines = [line.rstrip() for line in content_text.split('\n')]
+                
+                # Join lines, strip only leading/trailing empty lines
+                content_text = '\n'.join(lines).strip()
+                # Return as code block (need to HTML-escape the content)
+                content_escaped = html_module.escape(content_text)
+                return f'<pre><code>{content_escaped}</code></pre>'
+        
+        # Replace all marked sections with appropriate wrappers
+        result_html = pattern.sub(replace_with_appropriate_wrapper, html)
+        
+        return result_html
+
     def load(self):
         """
         Loads and converts the Docx file to markdown format.
@@ -131,6 +425,7 @@ class AlitaDocxMammothLoader(BaseLoader):
     def _convert_docx_to_markdown(self, docx_file):
         """
         Converts the content of a Docx file to markdown format.
+        Detects bordered content and treats it as code blocks.
 
         Args:
             docx_file (BinaryIO): The Docx file object.
@@ -138,11 +433,28 @@ class AlitaDocxMammothLoader(BaseLoader):
         Returns:
             str: The markdown content extracted from the Docx file.
         """
+        # Step 1: Detect and mark bordered content
+        # Reset stream position if needed
+        if hasattr(docx_file, 'seek'):
+            docx_file.seek(0)
+        
+        marked_docx, start_marker, end_marker = self.__detect_and_mark_bordered_content(docx_file)
+        
+        # Step 2: Convert marked DOCX to HTML using Mammoth
         if self.extract_images:
             # Extract images using the provided image handler
-            result = convert_to_html(docx_file, convert_image=mammoth.images.img_element(self.__handle_image))
+            result = convert_to_html(marked_docx, convert_image=mammoth.images.img_element(self.__handle_image))
         else:
             # Ignore images
-            result = convert_to_html(docx_file, convert_image=lambda image: "")
-        content = markdownify(result.value, heading_style="ATX")
+            result = convert_to_html(marked_docx, convert_image=lambda image: "")
+        
+        # Step 3: Wrap marked sections in <pre><code> tags
+        html_with_code_blocks = self.__wrap_marked_sections_in_code_blocks(
+            result.value, start_marker, end_marker
+        )
+        
+        # Step 4: Convert HTML to markdown
+        content = markdownify(html_with_code_blocks, heading_style="ATX")
+        
+        # Step 5: Post-process markdown (for image transcripts, etc.)
         return self.__postprocess_original_md(content)

alita_sdk/runtime/tools/llm.py

@@ -149,6 +149,8 @@ class LLMNode(BaseTool):
 
                     output_msgs = {"messages": new_messages}
                     if self.output_variables:
+                        if self.output_variables[0] == 'messages':
+                            return output_msgs
                         output_msgs[self.output_variables[0]] = current_completion.content if current_completion else None
 
                     return output_msgs

alita_sdk/tools/__init__.py

@@ -113,6 +113,10 @@ def get_tools(tools_list, alita, llm, store: Optional[BaseStore] = None, *args,
         elif tool_type in AVAILABLE_TOOLS and 'get_tools' in AVAILABLE_TOOLS[tool_type]:
             try:
                 get_tools_func = AVAILABLE_TOOLS[tool_type]['get_tools']
+                if tool['settings'].get('pgvector_configuration'):
+                    # Set collection schema to toolkit_id and put it to pgvector configuration
+                    # to propagate it to the all toolkits level from single place
+                    tool['settings']['pgvector_configuration']['collection_schema'] = str(tool['id'])
                 tools.extend(get_tools_func(tool))
 
             except Exception as e:

alita_sdk/tools/base_indexer_toolkit.py

@@ -110,7 +110,7 @@ class BaseIndexerToolkit(VectorStoreWrapperBase):
     def __init__(self, **kwargs):
         conn = kwargs.get('connection_string', None)
         connection_string = conn.get_secret_value() if isinstance(conn, SecretStr) else conn
-        collection_name = kwargs.get('collection_name')
+        collection_name = kwargs.get('collection_schema')
         
         if 'vectorstore_type' not in kwargs:
             kwargs['vectorstore_type'] = 'PGVector'

{alita_sdk-0.3.395.dist-info → alita_sdk-0.3.397.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: alita_sdk
-Version: 0.3.395
+Version: 0.3.397
 Summary: SDK for building langchain agents using resources from Alita
 Author-email: Artem Rozumenko <artyom.rozumenko@gmail.com>, Mikalai Biazruchka <mikalai_biazruchka@epam.com>, Roman Mitusov <roman_mitusov@epam.com>, Ivan Krakhmaliuk <lifedj27@gmail.com>, Artem Dubrovskiy <ad13box@gmail.com>
 License-Expression: Apache-2.0

{alita_sdk-0.3.395.dist-info → alita_sdk-0.3.397.dist-info}/RECORD

@@ -56,7 +56,7 @@ alita_sdk/runtime/langchain/document_loaders/AlitaBDDScenariosLoader.py,sha256=4
 alita_sdk/runtime/langchain/document_loaders/AlitaCSVLoader.py,sha256=3ne-a5qIkBuGL2pzIePxDr79n3RJhASbOdS5izYWDMg,2321
 alita_sdk/runtime/langchain/document_loaders/AlitaConfluenceLoader.py,sha256=NzpoL4C7UzyzLouTSL_xTQw70MitNt-WZz3Eyl7QkTA,8294
 alita_sdk/runtime/langchain/document_loaders/AlitaDirectoryLoader.py,sha256=fKezkgvIcLG7S2PVJp1a8sZd6C4XQKNZKAFC87DbQts,7003
-alita_sdk/runtime/langchain/document_loaders/AlitaDocxMammothLoader.py,sha256=9hi5eHgDIfa9wBWqTuwMM6D6W64czrDTfZl_htooe8Y,5943
+alita_sdk/runtime/langchain/document_loaders/AlitaDocxMammothLoader.py,sha256=HU4NZHYxLAgPkIrR-NKOfNz2SsfmkWp7F414SGSm3W0,18453
 alita_sdk/runtime/langchain/document_loaders/AlitaExcelLoader.py,sha256=YI8QaHRjCl8WtxuQKMXi_iTJBZ6da3OTNgoDFqNjz1g,9294
 alita_sdk/runtime/langchain/document_loaders/AlitaGitRepoLoader.py,sha256=5WXGcyHraSVj3ANHj_U6X4EDikoekrIYtS0Q_QqNIng,2608
 alita_sdk/runtime/langchain/document_loaders/AlitaImageLoader.py,sha256=QwgBJE-BvOasjgT1hYHZc0MP0F_elirUjSzKixoM6fY,6610
@@ -114,7 +114,7 @@ alita_sdk/runtime/tools/function.py,sha256=jk_JrtuYByR9Df5EFOGFheB9HktNPJcOwf4js
 alita_sdk/runtime/tools/graph.py,sha256=7jImBBSEdP5Mjnn2keOiyUwdGDFhEXLUrgUiugO3mgA,3503
 alita_sdk/runtime/tools/image_generation.py,sha256=Kls9D_ke_SK7xmVr7I9SlQcAEBJc86gf66haN0qIj9k,7469
 alita_sdk/runtime/tools/indexer_tool.py,sha256=whSLPevB4WD6dhh2JDXEivDmTvbjiMV1MrPl9cz5eLA,4375
-alita_sdk/runtime/tools/llm.py,sha256=gOk2bSOYL6CnUeb6kIPTZFn-NZC1LggDy3qj6eYL97U,15338
+alita_sdk/runtime/tools/llm.py,sha256=rn7gYn__yLyKyou61SN3DadtWzY3pcYR1G2IBbrtL9M,15452
 alita_sdk/runtime/tools/loop.py,sha256=uds0WhZvwMxDVFI6MZHrcmMle637cQfBNg682iLxoJA,8335
 alita_sdk/runtime/tools/loop_output.py,sha256=U4hO9PCQgWlXwOq6jdmCGbegtAxGAPXObSxZQ3z38uk,8069
 alita_sdk/runtime/tools/mcp_server_tool.py,sha256=MhLxZJ44LYrB_0GrojmkyqKoDRaqIHkEQAsg718ipog,4277
@@ -135,8 +135,8 @@ alita_sdk/runtime/utils/streamlit.py,sha256=GQ69CsjfRMcGXcCrslL0Uoj24Cl07Jeji0rZ
 alita_sdk/runtime/utils/toolkit_runtime.py,sha256=MU63Fpxj0b5_r1IUUc0Q3-PN9VwL7rUxp2MRR4tmYR8,5136
 alita_sdk/runtime/utils/toolkit_utils.py,sha256=I9QFqnaqfVgN26LUr6s3XlBlG6y0CoHURnCzG7XcwVs,5311
 alita_sdk/runtime/utils/utils.py,sha256=PJK8A-JVIzY1IowOjGG8DIqsIiEFe65qDKvFcjJCKWA,1041
-alita_sdk/tools/__init__.py,sha256=NrZyTEdEhmO1NnAR9RFMQ05Mb-kgu68mAQz3n5r0HYs,10692
-alita_sdk/tools/base_indexer_toolkit.py,sha256=JZR69D-ABZZEIE5I5LzthqM9-2xwLvuQPtLC3UYIPeE,26437
+alita_sdk/tools/__init__.py,sha256=6g3Y2zI88IEgVu3BhmsBJfqrV0DFuFehkpID_ip0Eus,11038
+alita_sdk/tools/base_indexer_toolkit.py,sha256=7UTcrmvGvmIBF3WGKrsEp7zJL-XB1JIgaRkbE1ZSS9A,26439
 alita_sdk/tools/code_indexer_toolkit.py,sha256=p3zVnCnQTUf7JUGra9Rl6GEK2W1-hvvz0Xsgz0v0muM,7292
 alita_sdk/tools/elitea_base.py,sha256=34fmVdYgd2YXifU5LFNjMQysr4OOIZ6AOZjq4GxLgSw,34417
 alita_sdk/tools/non_code_indexer_toolkit.py,sha256=6Lrqor1VeSLbPLDHAfg_7UAUqKFy1r_n6bdsc4-ak98,1315
@@ -353,8 +353,8 @@ alita_sdk/tools/zephyr_scale/api_wrapper.py,sha256=kT0TbmMvuKhDUZc0i7KO18O38JM9S
 alita_sdk/tools/zephyr_squad/__init__.py,sha256=0ne8XLJEQSLOWfzd2HdnqOYmQlUliKHbBED5kW_Vias,2895
 alita_sdk/tools/zephyr_squad/api_wrapper.py,sha256=kmw_xol8YIYFplBLWTqP_VKPRhL_1ItDD0_vXTe_UuI,14906
 alita_sdk/tools/zephyr_squad/zephyr_squad_cloud_client.py,sha256=R371waHsms4sllHCbijKYs90C-9Yu0sSR3N4SUfQOgU,5066
-alita_sdk-0.3.395.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-alita_sdk-0.3.395.dist-info/METADATA,sha256=4QMYM5jEWJT3NUl1TVc0r1iicEnE5fzYEDMZu_5Vghk,19071
-alita_sdk-0.3.395.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-alita_sdk-0.3.395.dist-info/top_level.txt,sha256=0vJYy5p_jK6AwVb1aqXr7Kgqgk3WDtQ6t5C-XI9zkmg,10
-alita_sdk-0.3.395.dist-info/RECORD,,
+alita_sdk-0.3.397.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+alita_sdk-0.3.397.dist-info/METADATA,sha256=RNLoAgkEYlqzitDQjSuZ1MLFzbki4fr5PH9UMiur3Kw,19071
+alita_sdk-0.3.397.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+alita_sdk-0.3.397.dist-info/top_level.txt,sha256=0vJYy5p_jK6AwVb1aqXr7Kgqgk3WDtQ6t5C-XI9zkmg,10
+alita_sdk-0.3.397.dist-info/RECORD,,