langroid 0.47.2__tar.gz → 0.48.1__tar.gz

{langroid-0.47.2 → langroid-0.48.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langroid
-Version: 0.47.2
+Version: 0.48.1
 Summary: Harness LLMs with Multi-Agent Programming
 Author-email: Prasad Chalasani <pchalasani@gmail.com>
 License: MIT
@@ -108,7 +108,7 @@ Requires-Dist: pytesseract<0.4.0,>=0.3.10; extra == 'doc-chat'
 Requires-Dist: python-docx<2.0.0,>=1.1.0; extra == 'doc-chat'
 Requires-Dist: unstructured[docx,pdf,pptx]<1.0.0,>=0.16.15; extra == 'doc-chat'
 Provides-Extra: doc-parsers
-Requires-Dist: markitdown>=0.0.1a3; extra == 'doc-parsers'
+Requires-Dist: markitdown[docx,pptx,xlsx]>=0.0.1a3; extra == 'doc-parsers'
 Requires-Dist: openpyxl>=3.1.5; extra == 'doc-parsers'
 Requires-Dist: python-docx>=1.1.2; extra == 'doc-parsers'
 Requires-Dist: python-pptx>=1.0.2; extra == 'doc-parsers'
@@ -144,6 +144,8 @@ Requires-Dist: litellm<2.0.0,>=1.30.1; extra == 'litellm'
 Provides-Extra: marker-pdf
 Requires-Dist: marker-pdf[full]>=1.6.0; (sys_platform != 'darwin' or platform_machine != 'x86_64') and extra == 'marker-pdf'
 Requires-Dist: opencv-python>=4.11.0.86; extra == 'marker-pdf'
+Provides-Extra: markitdown
+Requires-Dist: markitdown[docx,pptx,xlsx]>=0.0.1a3; extra == 'markitdown'
 Provides-Extra: meilisearch
 Requires-Dist: meilisearch-python-sdk<3.0.0,>=2.2.3; extra == 'meilisearch'
 Provides-Extra: metaphor
@@ -157,7 +159,7 @@ Requires-Dist: neo4j<6.0.0,>=5.14.1; extra == 'neo4j'
 Provides-Extra: pdf-parsers
 Requires-Dist: docling<3.0.0,>=2.16.0; extra == 'pdf-parsers'
 Requires-Dist: marker-pdf; extra == 'pdf-parsers'
-Requires-Dist: markitdown>=0.0.1a3; extra == 'pdf-parsers'
+Requires-Dist: markitdown[docx,pptx,xlsx]>=0.0.1a3; extra == 'pdf-parsers'
 Requires-Dist: pdf2image<2.0.0,>=1.17.0; extra == 'pdf-parsers'
 Requires-Dist: pymupdf4llm<0.1.0,>=0.0.17; extra == 'pdf-parsers'
 Requires-Dist: pymupdf<2.0.0,>=1.23.3; extra == 'pdf-parsers'

{langroid-0.47.2 → langroid-0.48.1}/langroid/mytypes.py

@@ -45,6 +45,8 @@ class DocMetaData(BaseModel):
 
     source: str = "context"  # just reference
     source_content: str = "context"  # reference and content
+    title: str = "unknown"
+    published_date: str = "unknown"
     is_chunk: bool = False  # if it is a chunk, don't split
     id: str = Field(default_factory=lambda: str(uuid4()))
     window_ids: List[str] = []  # for RAG: ids of chunks around this one

{langroid-0.47.2 → langroid-0.48.1}/langroid/parsing/document_parser.py

@@ -161,6 +161,8 @@ class DocumentParser(Parser):
                 return UnstructuredDocxParser(source, config)
             elif config.docx.library == "python-docx":
                 return PythonDocxParser(source, config)
+            elif config.docx.library == "markitdown-docx":
+                return MarkitdownDocxParser(source, config)
             else:
                 raise ValueError(
                     f"Unsupported DOCX library specified: {config.docx.library}"
@@ -887,6 +889,44 @@ class PythonDocxParser(DocumentParser):
         )
 
 
+class MarkitdownDocxParser(DocumentParser):
+    def iterate_pages(self) -> Generator[Tuple[int, Any], None, None]:
+        try:
+            from markitdown import MarkItDown
+        except ImportError:
+            LangroidImportError("markitdown", ["markitdown", "doc-parsers"])
+        md = MarkItDown()
+        self.doc_bytes.seek(0)  # Reset to start
+
+        # Direct conversion from stream works for DOCX (unlike XLSX)
+        result = md.convert_stream(self.doc_bytes, file_extension=".docx")
+
+        # Split content into logical sections (paragraphs, sections, etc.)
+        # This approach differs from the strict page-based approach used for PDFs
+        sections = re.split(r"(?=# |\n## |\n### )", result.text_content)
+
+        # Filter out empty sections
+        sections = [section for section in sections if section.strip()]
+
+        for i, section in enumerate(sections):
+            yield i, section
+
+    def get_document_from_page(self, md_content: str) -> Document:
+        """
+        Get Document object from a given markdown section.
+
+        Args:
+            md_content (str): The markdown content for the section.
+
+        Returns:
+            Document: Document object, with content and possible metadata.
+        """
+        return Document(
+            content=self.fix_text(md_content),
+            metadata=DocMetaData(source=self.source),
+        )
+
+
 class MarkitdownXLSXParser(DocumentParser):
     def iterate_pages(self) -> Generator[Tuple[int, Any], None, None]:
         try:

{langroid-0.47.2 → langroid-0.48.1}/langroid/parsing/parser.py

@@ -78,7 +78,7 @@ class PdfParsingConfig(BaseParsingConfig):
 
 
 class DocxParsingConfig(BaseSettings):
-    library: Literal["python-docx", "unstructured"] = "unstructured"
+    library: Literal["python-docx", "unstructured", "markitdown-docx"] = "unstructured"
 
 
 class DocParsingConfig(BaseSettings):

{langroid-0.47.2 → langroid-0.48.1}/langroid/parsing/url_loader.py

@@ -48,6 +48,15 @@ class FirecrawlConfig(BaseCrawlerConfig):
         env_prefix = "FIRECRAWL_"
 
 
+class ExaCrawlerConfig(BaseCrawlerConfig):
+    api_key: str = ""
+
+    class Config:
+        # Allow setting of fields via env vars with prefix EXA_
+        # e.g., EXA_API_KEY=your_api_key
+        env_prefix = "EXA_"
+
+
 class BaseCrawler(ABC):
     """Abstract base class for web crawlers."""
 
@@ -150,6 +159,8 @@ class CrawlerFactory:
             return TrafilaturaCrawler(config)
         elif isinstance(config, FirecrawlConfig):
             return FirecrawlCrawler(config)
+        elif isinstance(config, ExaCrawlerConfig):
+            return ExaCrawler(config)
         else:
             raise ValueError(f"Unsupported crawler configuration type: {type(config)}")
 
@@ -247,7 +258,13 @@ class FirecrawlCrawler(BaseCrawler):
                     with open(filename, "w") as f:
                         f.write(content)
                     docs.append(
-                        Document(content=content, metadata=DocMetaData(source=url))
+                        Document(
+                            content=content,
+                            metadata=DocMetaData(
+                                source=url,
+                                title=page["metadata"].get("title", ""),
+                            ),
+                        )
                     )
                     processed_urls.add(url)
                     new_pages += 1
@@ -289,7 +306,10 @@ class FirecrawlCrawler(BaseCrawler):
                         docs.append(
                             Document(
                                 content=result["markdown"],
-                                metadata=DocMetaData(source=url),
+                                metadata=DocMetaData(
+                                    source=url,
+                                    title=metadata.get("title", ""),
+                                ),
                             )
                         )
                 except Exception as e:
@@ -311,6 +331,77 @@ class FirecrawlCrawler(BaseCrawler):
         return docs
 
 
+class ExaCrawler(BaseCrawler):
+    """Crawler implementation using Exa API."""
+
+    def __init__(self, config: ExaCrawlerConfig) -> None:
+        """Initialize the Exa crawler.
+
+        Args:
+            config: Configuration for the crawler
+        """
+        super().__init__(config)
+        self.config: ExaCrawlerConfig = config
+
+    @property
+    def needs_parser(self) -> bool:
+        return True
+
+    def crawl(self, urls: List[str]) -> List[Document]:
+        """Crawl the given URLs using Exa SDK.
+
+        Args:
+            urls: List of URLs to crawl
+
+        Returns:
+            List of Documents with content extracted from the URLs
+
+        Raises:
+            LangroidImportError: If the exa package is not installed
+            ValueError: If the Exa API key is not set
+        """
+        try:
+            from exa_py import Exa
+        except ImportError:
+            raise LangroidImportError("exa", "exa")
+
+        if not self.config.api_key:
+            raise ValueError("EXA_API_KEY key is required in your env or .env")
+
+        exa = Exa(self.config.api_key)
+        docs = []
+
+        try:
+            for url in urls:
+                parsed_doc_chunks = self._process_document(url)
+                if parsed_doc_chunks:
+                    docs.extend(parsed_doc_chunks)
+                    continue
+                else:
+                    results = exa.get_contents([url], livecrawl="always", text=True)
+                    result = results.results[0]
+                    if result.text:
+                        # append a NON-chunked document
+                        # (metadata.is_chunk = False, so will be chunked downstream)
+                        docs.append(
+                            Document(
+                                content=result.text,
+                                metadata=DocMetaData(
+                                    source=url,
+                                    title=getattr(result, "title", ""),
+                                    published_date=getattr(
+                                        result, "published_date", ""
+                                    ),
+                                ),
+                            )
+                        )
+
+        except Exception as e:
+            logging.error(f"Error retrieving content from Exa API: {e}")
+
+        return docs
+
+
 class URLLoader:
     """Loads URLs and extracts text using a specified crawler."""
 
@@ -334,6 +425,8 @@ class URLLoader:
             crawler_config = TrafilaturaConfig(parser=Parser(parsing_config))
 
         self.crawler = CrawlerFactory.create_crawler(crawler_config)
+        if self.crawler.needs_parser:
+            self.crawler.parser = Parser(parsing_config)
 
     def load(self) -> List[Document]:
         """Load the URLs using the specified crawler."""

{langroid-0.47.2 → langroid-0.48.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "langroid"
-version = "0.47.2"
+version = "0.48.1"
 authors = [
     {name = "Prasad Chalasani", email = "pchalasani@gmail.com"},
 ]
@@ -147,7 +147,7 @@ pdf-parsers = [
     "pymupdf4llm<0.1.0,>=0.0.17",
     "pdf2image<2.0.0,>=1.17.0",
     "pytesseract<0.4.0,>=0.3.10",
-    "markitdown>=0.0.1a3",
+    "markitdown[docx,xlsx,pptx]>=0.0.1a3",
     "marker-pdf",
 ]
 
@@ -155,6 +155,10 @@ docx = [
     "python-docx<2.0.0,>=1.1.0",
 ]
 
+markitdown = [
+    "markitdown[docx,xlsx,pptx]>=0.0.1a3",
+]
+
 marker-pdf = [
     "marker-pdf[full]>=1.6.0; sys_platform != 'darwin' or platform_machine != 'x86_64'",
     "opencv-python>=4.11.0.86",
@@ -252,7 +256,7 @@ google-generativeai = [
     "google-genai>=1.0.0",
 ]
 doc-parsers = [
-    "markitdown>=0.0.1a3",
+    "markitdown[docx,xlsx,pptx]>=0.0.1a3",
     "openpyxl>=3.1.5",
     "python-docx>=1.1.2",
     "python-pptx>=1.0.2",