feat(api): PDF + DOCX extraction in folder indexer

Add pypdf/python-docx deps, _extract_pdf_text/_extract_docx_text helpers,
and summarize_pdf/summarize_docx wrappers that delegate to summarize_text.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app/core/folder_indexer.py

@@ -1,9 +1,13 @@
 """Per-file summarisation for project folder integration."""
 from __future__ import annotations
 
+import base64
+import io
 from dataclasses import dataclass
 
 from langchain_core.messages import HumanMessage, SystemMessage
+from pypdf import PdfReader
+from docx import Document as DocxDocument
 
 from app.core.langfuse_client import (
     compile_prompt,
@@ -98,3 +102,53 @@ async def summarize_text(*, content: str, ext: str, name: str) -> IndexResult:
     usage = extract_usage(response)
     summary = (response.content or "").strip()[:500]
     return IndexResult(summary=summary, tokens_used=usage.get("total", 0))
+
+
+def _extract_pdf_text(pdf_b64: str) -> str:
+    buf = io.BytesIO(base64.b64decode(pdf_b64))
+    reader = PdfReader(buf)
+    parts: list[str] = []
+    for page in reader.pages:
+        try:
+            parts.append(page.extract_text() or "")
+        except Exception:
+            continue
+    return "\n".join(parts).strip()
+
+
+def _extract_docx_text(docx_b64: str) -> str:
+    buf = io.BytesIO(base64.b64decode(docx_b64))
+    doc = DocxDocument(buf)
+    return "\n".join(p.text for p in doc.paragraphs if p.text).strip()
+
+
+async def summarize_pdf(*, pdf_b64: str, name: str) -> IndexResult:
+    """Return a compact summary of a PDF file.
+
+    Parameters
+    ----------
+    pdf_b64:
+        Base64-encoded PDF bytes.
+    name:
+        File name, e.g. ``"report.pdf"``.
+    """
+    text = _extract_pdf_text(pdf_b64)
+    if not text:
+        return IndexResult(summary="Could not extract text", tokens_used=0)
+    return await summarize_text(content=text, ext=".pdf", name=name)
+
+
+async def summarize_docx(*, docx_b64: str, name: str) -> IndexResult:
+    """Return a compact summary of a DOCX file.
+
+    Parameters
+    ----------
+    docx_b64:
+        Base64-encoded DOCX bytes.
+    name:
+        File name, e.g. ``"spec.docx"``.
+    """
+    text = _extract_docx_text(docx_b64)
+    if not text:
+        return IndexResult(summary="Could not extract text", tokens_used=0)
+    return await summarize_text(content=text, ext=".docx", name=name)

requirements.txt

@@ -39,3 +39,5 @@ lxml>=5.0.0
 PyYAML>=6.0.0
 apscheduler>=3.10.0
 ruff>=0.8.0
+pypdf>=4.0
+python-docx>=1.1

tests/test_folder_indexer.py

@@ -51,3 +51,33 @@ async def test_summarize_image_uses_vision_content_blocks():
         isinstance(p, dict) and p.get("type") == "image_url"
         for p in (last.content if isinstance(last.content, list) else [])
     )
+
+
+async def test_summarize_pdf_extracts_then_summarizes(monkeypatch):
+    # pypdf.PdfReader returns text from pages
+    from app.core import folder_indexer
+    class FakePage:
+        def extract_text(self): return "PDF page content with project info."
+    class FakeReader:
+        pages = [FakePage(), FakePage()]
+    monkeypatch.setattr(folder_indexer, "PdfReader", lambda buf: FakeReader())
+    mock_resp = AsyncMock(); mock_resp.content = "Project info doc."; mock_resp.usage_metadata = {"total_tokens": 50}
+    async def fake_llm(messages): return mock_resp
+    with patch("app.core.folder_indexer._llm_text", new=fake_llm):
+        result = await folder_indexer.summarize_pdf(pdf_b64="SGVsbG8=", name="doc.pdf")
+    assert "Project info" in result.summary
+    assert result.tokens_used == 50
+
+
+async def test_summarize_docx_extracts_then_summarizes(monkeypatch):
+    from app.core import folder_indexer
+    class FakePara:
+        def __init__(self, t): self.text = t
+    class FakeDoc:
+        paragraphs = [FakePara("Heading"), FakePara("Body paragraph one.")]
+    monkeypatch.setattr(folder_indexer, "DocxDocument", lambda buf: FakeDoc())
+    mock_resp = AsyncMock(); mock_resp.content = "Heading and body."; mock_resp.usage_metadata = {"total_tokens": 30}
+    async def fake_llm(messages): return mock_resp
+    with patch("app.core.folder_indexer._llm_text", new=fake_llm):
+        result = await folder_indexer.summarize_docx(docx_b64="UEsDBBQ=", name="doc.docx")
+    assert result.summary == "Heading and body."