feat(rag): PDF/docx ingest + chat citations

- Upload endpoint (base64 JSON, no multipart dep): extracts text from
  pdf/docx/txt/md via pypdf + python-docx, then runs the existing
  chunk/embed pipeline. Documents page uploads files straight through.
- Citations: the chat stream emits an SSE `sources` event listing the
  documents that fed the answer; the UI shows them as chips under the reply.
- Deps: pypdf, python-docx (both pure-Python, Windows-safe).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

synapse/main.py

@@ -287,10 +287,12 @@ async def chat_stream_endpoint(payload: Dict[str, Any]):
 
         # Retrieve relevant uploaded documents (RAG) and inject the top chunks.
         doc_hits = await store.search_documents(message, get_ollama_manager().embed, limit=3)
+        doc_titles: list = []
         if doc_hits:
             doc_block = "\n\n".join(f"[{d['title']}]\n{d['text']}" for d in doc_hits)
             separator = "\n\n---\nRelevant documents (cite as source material):\n\n"
             system_prompt = (system_prompt + separator + doc_block) if system_prompt else doc_block
+            doc_titles = list(dict.fromkeys(d["title"] for d in doc_hits))  # unique, order-preserving
 
         # Fetch web search results for time-sensitive queries
         search_results = ""
@@ -373,6 +375,10 @@ async def chat_stream_endpoint(payload: Dict[str, Any]):
             meta: dict = {}
             final_model = model
 
+            # Tell the client which documents fed this answer (RAG citations).
+            if doc_titles:
+                yield f"event: sources\ndata: {_json.dumps({'sources': doc_titles})}\n\n"
+
             # ── Phase 1: stream primary model response ────────────────────
             try:
                 async for chunk in stream_chat_response(
@@ -953,6 +959,46 @@ async def add_document(payload: Dict[str, Any] = Body(...)):
     return result
 
 
+def _extract_text(filename: str, data: bytes) -> str:
+    """Pull plain text from an uploaded file by extension. PDF/DOCX use
+    pure-Python parsers; anything else is decoded as UTF-8."""
+    import io
+    name = (filename or "").lower()
+    if name.endswith(".pdf"):
+        from pypdf import PdfReader
+        reader = PdfReader(io.BytesIO(data))
+        return "\n\n".join((page.extract_text() or "") for page in reader.pages)
+    if name.endswith(".docx"):
+        import docx
+        doc = docx.Document(io.BytesIO(data))
+        return "\n\n".join(p.text for p in doc.paragraphs if p.text.strip())
+    return data.decode("utf-8", errors="replace")
+
+
+@app.post("/documents/upload")
+async def upload_document(payload: Dict[str, Any] = Body(...)):
+    """Ingest a file (pdf/docx/txt/md) sent as base64. Extracts text, then runs
+    the same chunk/embed pipeline as a pasted document."""
+    import base64
+    import os as _os
+    filename = (payload.get("filename") or "").strip()
+    b64 = payload.get("data") or ""
+    if not filename or not b64:
+        raise HTTPException(status_code=400, detail="filename and data are required")
+    try:
+        raw = base64.b64decode(b64)
+    except Exception:
+        raise HTTPException(status_code=400, detail="data must be base64")
+    try:
+        text = _extract_text(filename, raw).strip()
+    except Exception as e:
+        raise HTTPException(status_code=400, detail=f"could not read {filename}: {e}")
+    if not text:
+        raise HTTPException(status_code=400, detail="no extractable text in file")
+    title = _os.path.splitext(_os.path.basename(filename))[0] or filename
+    return await store.add_document(title, text, get_ollama_manager().embed)
+
+
 @app.get("/documents/{doc_id}")
 async def get_document(doc_id: str):
     chunks = store.get_document(doc_id)