langroid 0.36.0__tar.gz → 0.36.1__tar.gz

{langroid-0.36.0 → langroid-0.36.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langroid
-Version: 0.36.0
+Version: 0.36.1
 Summary: Harness LLMs with Multi-Agent Programming
 Author-email: Prasad Chalasani <pchalasani@gmail.com>
 License: MIT
@@ -292,20 +292,28 @@ teacher_task.run()
 <summary> <b>Click to expand</b></summary>
 
 - **Jan 2025:**
-  - [0.33.0](https://github.com/langroid/langroid/releases/tag/0.33.3) Move from Poetry to uv!
+  - [0.36.0](https://github.com/langroid/langroid/releases/tag/0.36.0): Weaviate vector-db support (thanks @abab-dev).
+  - [0.35.0](https://github.com/langroid/langroid/releases/tag/0.35.0): Capture/Stream reasoning content from 
+    Reasoning LLMs (e.g. DeepSeek, OpenAI o1) in addition to final answer.
+  - [0.34.0](https://github.com/langroid/langroid/releases/tag/0.34.0): DocChatAgent 
+    chunk enrichment to improve retrieval. (collaboration with @dfm88). 
+  - [0.33.0](https://github.com/langroid/langroid/releases/tag/0.33.3) Move from Poetry to uv! (thanks @abab-dev).
   - [0.32.0](https://github.com/langroid/langroid/releases/tag/0.32.0) DeepSeek v3 support.
 - **Dec 2024:**
   - [0.31.0](https://github.com/langroid/langroid/releases/tag/0.31.0) Azure OpenAI Embeddings
-  - [0.30.0](https://github.com/langroid/langroid/releases/tag/0.30.0) Llama-cpp embeddings.
-  - [0.29.0](https://github.com/langroid/langroid/releases/tag/0.29.0) Custom Azure OpenAI Client 
+  - [0.30.0](https://github.com/langroid/langroid/releases/tag/0.30.0) Llama-cpp embeddings (thanks @Kwigg).
+  - [0.29.0](https://github.com/langroid/langroid/releases/tag/0.29.0) Custom Azure OpenAI Client (thanks 
+    @johannestang).
   - [0.28.0](https://github.com/langroid/langroid/releases/tag/0.28.0) `ToolMessage`: `_handler` field to override 
-default handler method name in `request` field. 
+default handler method name in `request` field (thanks @alexagr).
   - [0.27.0](https://github.com/langroid/langroid/releases/tag/0.27.0) OpenRouter Support.
   - [0.26.0](https://github.com/langroid/langroid/releases/tag/0.26.0) Update to latest Chainlit.
-  - [0.25.0](https://github.com/langroid/langroid/releases/tag/0.25.0) True Async Methods for agent and user-response.
+  - [0.25.0](https://github.com/langroid/langroid/releases/tag/0.25.0) True Async Methods for agent and 
+    user-response (thanks @alexagr).
 - **Nov 2024:**
   - **[0.24.0](https://langroid.github.io/langroid/notes/structured-output/)**: 
      Enables support for `Agent`s with strict JSON schema output format on compatible LLMs and strict mode for the OpenAI tools API.
+    (thanks @nilspalumbo).
   - **[0.23.0](https://langroid.github.io/langroid/tutorials/local-llm-setup/#local-llms-hosted-on-glhfchat)**: 
       support for LLMs (e.g. `Qwen2.5-Coder-32b-Instruct`) hosted on glhf.chat 
   - **[0.22.0](https://langroid.github.io/langroid/notes/large-tool-results/)**: 

{langroid-0.36.0 → langroid-0.36.1}/README.md

@@ -136,20 +136,28 @@ teacher_task.run()
 <summary> <b>Click to expand</b></summary>
 
 - **Jan 2025:**
-  - [0.33.0](https://github.com/langroid/langroid/releases/tag/0.33.3) Move from Poetry to uv!
+  - [0.36.0](https://github.com/langroid/langroid/releases/tag/0.36.0): Weaviate vector-db support (thanks @abab-dev).
+  - [0.35.0](https://github.com/langroid/langroid/releases/tag/0.35.0): Capture/Stream reasoning content from 
+    Reasoning LLMs (e.g. DeepSeek, OpenAI o1) in addition to final answer.
+  - [0.34.0](https://github.com/langroid/langroid/releases/tag/0.34.0): DocChatAgent 
+    chunk enrichment to improve retrieval. (collaboration with @dfm88). 
+  - [0.33.0](https://github.com/langroid/langroid/releases/tag/0.33.3) Move from Poetry to uv! (thanks @abab-dev).
   - [0.32.0](https://github.com/langroid/langroid/releases/tag/0.32.0) DeepSeek v3 support.
 - **Dec 2024:**
   - [0.31.0](https://github.com/langroid/langroid/releases/tag/0.31.0) Azure OpenAI Embeddings
-  - [0.30.0](https://github.com/langroid/langroid/releases/tag/0.30.0) Llama-cpp embeddings.
-  - [0.29.0](https://github.com/langroid/langroid/releases/tag/0.29.0) Custom Azure OpenAI Client 
+  - [0.30.0](https://github.com/langroid/langroid/releases/tag/0.30.0) Llama-cpp embeddings (thanks @Kwigg).
+  - [0.29.0](https://github.com/langroid/langroid/releases/tag/0.29.0) Custom Azure OpenAI Client (thanks 
+    @johannestang).
   - [0.28.0](https://github.com/langroid/langroid/releases/tag/0.28.0) `ToolMessage`: `_handler` field to override 
-default handler method name in `request` field. 
+default handler method name in `request` field (thanks @alexagr).
   - [0.27.0](https://github.com/langroid/langroid/releases/tag/0.27.0) OpenRouter Support.
   - [0.26.0](https://github.com/langroid/langroid/releases/tag/0.26.0) Update to latest Chainlit.
-  - [0.25.0](https://github.com/langroid/langroid/releases/tag/0.25.0) True Async Methods for agent and user-response.
+  - [0.25.0](https://github.com/langroid/langroid/releases/tag/0.25.0) True Async Methods for agent and 
+    user-response (thanks @alexagr).
 - **Nov 2024:**
   - **[0.24.0](https://langroid.github.io/langroid/notes/structured-output/)**: 
      Enables support for `Agent`s with strict JSON schema output format on compatible LLMs and strict mode for the OpenAI tools API.
+    (thanks @nilspalumbo).
   - **[0.23.0](https://langroid.github.io/langroid/tutorials/local-llm-setup/#local-llms-hosted-on-glhfchat)**: 
       support for LLMs (e.g. `Qwen2.5-Coder-32b-Instruct`) hosted on glhf.chat 
   - **[0.22.0](https://langroid.github.io/langroid/notes/large-tool-results/)**: 

{langroid-0.36.0 → langroid-0.36.1}/langroid/agent/special/doc_chat_agent.py

@@ -15,6 +15,7 @@ pip install "langroid[hf-embeddings]"
 """
 
 import logging
+import textwrap
 from collections import OrderedDict
 from functools import cache
 from typing import Any, Callable, Dict, List, Optional, Set, Tuple, no_type_check
@@ -81,7 +82,7 @@ You will be given various passages from these documents, and asked to answer que
 about them, or summarize them into coherent answers.
 """
 
-CHUNK_ENRICHMENT_DELIMITER = "<##-##-##>"
+CHUNK_ENRICHMENT_DELIMITER = "\n<##-##-##>"
 
 has_sentence_transformers = False
 try:
@@ -810,9 +811,11 @@ class DocChatAgent(ChatAgent):
         return "\n".join(
             [
                 f"""
-                [{i+1}]
+                -----[EXTRACT #{i+1}]----------
                 {content}
                 {source}
+                -----END OF EXTRACT------------
+                
                 """
                 for i, (content, source) in enumerate(zip(contents, sources))
             ]
@@ -949,12 +952,13 @@ class DocChatAgent(ChatAgent):
                     continue
 
                 # Combine original content with questions in a structured way
-                combined_content = f"""
-                {doc.content}
-                
+                combined_content = textwrap.dedent(
+                    f"""\
+                {doc.content}                
                 {enrichment_config.delimiter}
                 {enrichment}
-                """.strip()
+                """
+                )
 
                 new_doc = doc.copy(
                     update={
@@ -1440,7 +1444,7 @@ class DocChatAgent(ChatAgent):
         delimiter = self.config.chunk_enrichment_config.delimiter
         return [
             (
-                doc.copy(update={"content": doc.content.split(delimiter)[0].strip()})
+                doc.copy(update={"content": doc.content.split(delimiter)[0]})
                 if doc.content and getattr(doc.metadata, "has_enrichment", False)
                 else doc
             )

{langroid-0.36.0 → langroid-0.36.1}/langroid/parsing/parser.py

@@ -267,9 +267,11 @@ class Parser:
                 # Truncate the chunk text at the punctuation mark
                 chunk_text = chunk_text[: last_punctuation + 1]
 
-            # Remove any newline characters and strip any leading or
-            # trailing whitespace
-            chunk_text_to_append = re.sub(r"\n{2,}", "\n", chunk_text).strip()
+            # Replace redundant (3 or more) newlines with 2 newlines to preser
+            # paragraph separation!
+            # But do NOT strip leading/trailing whitespace, to preserve formatting
+            # (e.g. code blocks, or in case we want to stitch chunks back together)
+            chunk_text_to_append = re.sub(r"\n{3,}", "\n\n", chunk_text)
 
             if len(chunk_text_to_append) > self.config.discard_chunk_chars:
                 # Append the chunk text to the list of chunks

{langroid-0.36.0 → langroid-0.36.1}/langroid/parsing/utils.py

@@ -310,9 +310,9 @@ def extract_numbered_segments(s: str, specs: str) -> str:
         ]
 
         # If we extracted any segments from this paragraph,
-        # join them and append to results
+        # join them with ellipsis (...) and append to results.
         if extracted_segments:
-            extracted_paragraphs.append(" ".join(extracted_segments))
+            extracted_paragraphs.append("...".join(extracted_segments))
 
     return "\n\n".join(extracted_paragraphs)
 

langroid-0.36.1/langroid/utils/output/citations.py

@@ -0,0 +1,61 @@
+def extract_markdown_references(md_string: str) -> list[int]:
+    """
+    Extracts markdown references (e.g., [^1], [^2]) from a string and returns
+    them as a sorted list of integers.
+
+    Args:
+        md_string (str): The markdown string containing references.
+
+    Returns:
+        list[int]: A sorted list of unique integers from the markdown references.
+    """
+    import re
+
+    # Regex to find all occurrences of [^<number>]
+    matches = re.findall(r"\[\^(\d+)\]", md_string)
+    # Convert matches to integers, remove duplicates with set, and sort
+    return sorted(set(int(match) for match in matches))
+
+
+def format_footnote_text(content: str, width: int = 0) -> str:
+    """
+    Formats the content so that each original line is individually processed.
+    - If width=0, no wrapping is done (lines remain as is).
+    - If width>0, lines are wrapped to that width.
+    - Blank lines remain blank (with indentation).
+    - Everything is indented by 4 spaces (for markdown footnotes).
+
+    Args:
+        content (str): The text of the footnote to be formatted.
+        width (int): Maximum width of the text lines. If 0, lines are not wrapped.
+
+    Returns:
+        str: Properly formatted markdown footnote text.
+    """
+    import textwrap
+
+    indent = "    "  # 4 spaces for markdown footnotes
+    lines = content.split("\n")  # keep original line structure
+
+    output_lines = []
+    for line in lines:
+        # If the line is empty (or just spaces), keep it blank (but indented)
+        if not line.strip():
+            output_lines.append(indent)
+            continue
+
+        if width > 0:
+            # Wrap each non-empty line to the specified width
+            wrapped = textwrap.wrap(line, width=width)
+            if not wrapped:
+                # If textwrap gives nothing, add a blank (indented) line
+                output_lines.append(indent)
+            else:
+                for subline in wrapped:
+                    output_lines.append(indent + subline)
+        else:
+            # No wrapping: just indent the original line
+            output_lines.append(indent + line)
+
+    # Join them with newline so we preserve the paragraph/blank line structure
+    return "\n".join(output_lines)

{langroid-0.36.0 → langroid-0.36.1}/langroid/vector_store/base.py

@@ -264,7 +264,7 @@ class VectorStore(ABC):
             metadata = copy.deepcopy(id2metadata[w[0]])
             metadata.window_ids = w
             document = Document(
-                content=" ".join([d.content for d in self.get_documents_by_ids(w)]),
+                content="".join([d.content for d in self.get_documents_by_ids(w)]),
                 metadata=metadata,
             )
             # make a fresh id since content is in general different

{langroid-0.36.0 → langroid-0.36.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "langroid"
-version = "0.36.0"
+version = "0.36.1"
 authors = [
     {name = "Prasad Chalasani", email = "pchalasani@gmail.com"},
 ]

langroid-0.36.0/langroid/utils/output/citations.py

@@ -1,41 +0,0 @@
-def extract_markdown_references(md_string: str) -> list[int]:
-    """
-    Extracts markdown references (e.g., [^1], [^2]) from a string and returns
-    them as a sorted list of integers.
-
-    Args:
-        md_string (str): The markdown string containing references.
-
-    Returns:
-        list[int]: A sorted list of unique integers from the markdown references.
-    """
-    import re
-
-    # Regex to find all occurrences of [^<number>]
-    matches = re.findall(r"\[\^(\d+)\]", md_string)
-    # Convert matches to integers, remove duplicates with set, and sort
-    return sorted(set(int(match) for match in matches))
-
-
-def format_footnote_text(content: str, width: int = 80) -> str:
-    """
-    Formats the content part of a footnote (i.e. not the first line that
-    appears right after the reference [^4])
-    It wraps the text so that no line is longer than the specified width and indents
-    lines as necessary for markdown footnotes.
-
-    Args:
-        content (str): The text of the footnote to be formatted.
-        width (int): Maximum width of the text lines.
-
-    Returns:
-        str: Properly formatted markdown footnote text.
-    """
-    import textwrap
-
-    # Wrap the text to the specified width
-    wrapped_lines = textwrap.wrap(content, width)
-    if len(wrapped_lines) == 0:
-        return ""
-    indent = "    "  # Indentation for markdown footnotes
-    return indent + ("\n" + indent).join(wrapped_lines)