vn6295337's picture
Add automatic Docling parsing display in indexing flow
aa663e1
"""API routes for RAG application."""
import os
import shutil
import httpx
from pathlib import Path
from typing import List
from fastapi import APIRouter, HTTPException, UploadFile, File, Form
from src.api.models import (
QueryRequest, QueryResponse,
IngestRequest, IngestResponse,
SyncRequest, SyncResponse,
StatusResponse, Citation
)
from src.orchestrator import orchestrate_query, set_chunks_path
from src.ingestion.api import ingest_from_directory, sync_to_pinecone, get_index_status
from src.retrieval.keyword_search import reload_index
router = APIRouter()
# Upload directory for user documents
UPLOAD_DIR = Path("uploads")
UPLOAD_DIR.mkdir(exist_ok=True)
@router.post("/query", response_model=QueryResponse)
async def query(request: QueryRequest):
"""Execute RAG query and return answer with citations."""
try:
result = orchestrate_query(
query=request.query,
top_k=request.top_k,
use_hybrid=request.use_hybrid,
use_reranking=request.use_reranking
)
# Convert sources/citations to Citation models
sources = [
Citation(
id=s.get("id"),
score=s.get("score", 0.0),
snippet=s.get("snippet", "")
)
for s in result.get("sources", [])
]
citations = [
Citation(
id=c.get("id"),
score=c.get("score", 0.0),
snippet=c.get("snippet", "")
)
for c in result.get("citations", [])
]
return QueryResponse(
answer=result.get("answer", ""),
sources=sources,
citations=citations,
query_rewrite=result.get("query_rewrite"),
retrieval_meta=result.get("retrieval_meta"),
error=result.get("llm_meta", {}).get("error")
)
except Exception as e:
return QueryResponse(answer="", error=str(e))
@router.post("/ingest", response_model=IngestResponse)
async def ingest(request: IngestRequest):
"""Ingest documents from directory and create chunks."""
try:
result = ingest_from_directory(
docs_dir=request.docs_dir,
output_path=request.output_path,
provider=request.provider
)
# Reload BM25 index if successful
if result.status == "success":
reload_index(request.output_path)
set_chunks_path(request.output_path)
return IngestResponse(
status=result.status,
documents=result.documents,
chunks=result.chunks,
output_path=result.output_path,
errors=result.errors
)
except Exception as e:
return IngestResponse(status="error", errors=[str(e)])
@router.post("/sync-pinecone", response_model=SyncResponse)
async def sync_pinecone(request: SyncRequest):
"""Sync embeddings to Pinecone vector database."""
try:
result = sync_to_pinecone(
chunks_path=request.chunks_path,
batch_size=request.batch_size
)
return SyncResponse(
status=result.status,
vectors_upserted=result.vectors_upserted,
errors=result.errors
)
except Exception as e:
return SyncResponse(status="error", errors=[str(e)])
@router.get("/status", response_model=StatusResponse)
async def status(chunks_path: str = "data/chunks.jsonl"):
"""Get current index status."""
try:
result = get_index_status(chunks_path)
return StatusResponse(
exists=result.get("exists", False),
chunks=result.get("chunks", 0),
documents=result.get("documents", 0),
path=result.get("path"),
error=result.get("error")
)
except Exception as e:
return StatusResponse(error=str(e))
@router.get("/health")
async def health():
"""Health check endpoint."""
return {"status": "ok"}
@router.delete("/clear-index")
async def clear_index():
"""
Clear all vectors from Pinecone index.
Use before uploading new documents to avoid stale data.
"""
from pinecone import Pinecone
import src.config as cfg
try:
pc = Pinecone(api_key=cfg.PINECONE_API_KEY)
idx_meta = pc.describe_index(cfg.PINECONE_INDEX_NAME)
host = getattr(idx_meta, "host", None) or idx_meta.get("host")
index = pc.Index(host=host)
# Delete all vectors
index.delete(delete_all=True)
return {"status": "success", "message": "Index cleared"}
except Exception as e:
return {"status": "error", "error": str(e)}
@router.post("/embed-chunks")
async def embed_chunks(request: dict):
"""
Embed pre-chunked text and upsert to Pinecone.
ZERO-STORAGE PRIVACY:
- Text is used ONLY for embedding generation
- Only embeddings + file metadata stored in Pinecone
- NO text content stored anywhere
- Original text must be re-fetched from Dropbox at query time
"""
from src.ingestion.embeddings import batch_embed_chunks
from pinecone import Pinecone
import src.config as cfg
chunks = request.get("chunks", [])
if not chunks:
return {"status": "error", "error": "No chunks provided", "vectors_upserted": 0}
try:
# Prepare chunks for embedding
chunk_data = []
for i, chunk in enumerate(chunks):
text = chunk.get("text", "")
metadata = chunk.get("metadata", {})
chunk_data.append({
"text": text,
"filename": metadata.get("filename", f"doc_{i}"),
"chunk_id": metadata.get("chunkIndex", i),
"chars": len(text),
})
# Generate embeddings (text processed in memory only)
embedded = batch_embed_chunks(chunk_data, provider="sentence-transformers", dim=384)
# Prepare vectors for Pinecone - NO TEXT STORED
vectors = []
for j, emb in enumerate(embedded):
chunk_meta = chunks[j].get("metadata", {})
# Use filename for readable IDs (sanitize for Pinecone compatibility)
filename = chunk_meta.get("filename", "doc")
vectors.append({
"id": f"{filename}::{chunk_meta.get('chunkIndex', j)}",
"values": emb["embedding"],
"metadata": {
# File info for re-fetching
"filename": chunk_meta.get("filename", ""),
"file_path": chunk_meta.get("filePath", ""), # Dropbox path
"file_id": chunk_meta.get("fileId", ""),
# Chunk position for extraction
"chunk_index": chunk_meta.get("chunkIndex", j),
"start_char": chunk_meta.get("startChar", 0),
"end_char": chunk_meta.get("endChar", 0),
# NO TEXT STORED - zero storage compliance
}
})
# Upsert to Pinecone
pc = Pinecone(api_key=cfg.PINECONE_API_KEY)
idx_meta = pc.describe_index(cfg.PINECONE_INDEX_NAME)
host = getattr(idx_meta, "host", None) or idx_meta.get("host")
index = pc.Index(host=host)
# Batch upsert
batch_size = 100
upserted = 0
for i in range(0, len(vectors), batch_size):
batch = vectors[i:i + batch_size]
index.upsert(vectors=batch)
upserted += len(batch)
# PRIVACY: Explicitly delete all text references from memory
del chunks
del chunk_data
del embedded
return {
"status": "success",
"vectors_upserted": upserted,
"error": None
}
except Exception as e:
return {
"status": "error",
"vectors_upserted": 0,
"error": str(e)
}
@router.post("/query-secure")
async def query_secure(request: dict):
"""
ZERO-STORAGE QUERY: Re-fetches text from Dropbox at query time.
Flow:
1. Generate query embedding
2. Search Pinecone for similar chunks (returns file paths + positions)
3. Re-fetch files from Dropbox using provided access token
4. Extract chunk text using stored positions
5. Send to LLM for answer generation
6. Return answer (text never stored)
"""
from src.ingestion.embeddings import get_embedding
from pinecone import Pinecone
import src.config as cfg
query = request.get("query", "")
access_token = request.get("access_token")
top_k = request.get("top_k", 3)
if not query:
return {"error": "No query provided", "answer": ""}
if not access_token:
return {"error": "Dropbox access token required for zero-storage queries", "answer": ""}
try:
# 1. Generate query embedding
query_embedding = get_embedding(query, provider="sentence-transformers", dim=384)
# 2. Search Pinecone
pc = Pinecone(api_key=cfg.PINECONE_API_KEY)
idx_meta = pc.describe_index(cfg.PINECONE_INDEX_NAME)
host = getattr(idx_meta, "host", None) or idx_meta.get("host")
index = pc.Index(host=host)
results = index.query(
vector=query_embedding,
top_k=top_k,
include_metadata=True
)
if not results.matches:
return {"answer": "No relevant documents found.", "citations": []}
# 3. Group chunks by file for efficient fetching
files_to_fetch = {}
for match in results.matches:
meta = match.metadata or {}
file_path = meta.get("file_path", "")
if file_path:
if file_path not in files_to_fetch:
files_to_fetch[file_path] = []
files_to_fetch[file_path].append({
"id": match.id,
"score": match.score,
"start_char": meta.get("start_char", 0),
"end_char": meta.get("end_char", 0),
"filename": meta.get("filename", ""),
})
# 4. Re-fetch files from Dropbox and extract chunks
chunks_with_text = []
async with httpx.AsyncClient(timeout=60.0) as client:
for file_path, chunks in files_to_fetch.items():
# Fetch file content
response = await client.post(
"https://content.dropboxapi.com/2/files/download",
headers={
"Authorization": f"Bearer {access_token}",
"Dropbox-API-Arg": f'{{"path": "{file_path}"}}'
}
)
if response.status_code == 200:
# Handle PDF vs text
if file_path.lower().endswith('.pdf'):
import io
from PyPDF2 import PdfReader
pdf_file = io.BytesIO(response.content)
reader = PdfReader(pdf_file)
file_content = "\n\n".join(
page.extract_text() or "" for page in reader.pages
)
else:
file_content = response.text
# Extract each chunk using stored positions
for chunk in chunks:
start = chunk["start_char"]
end = chunk["end_char"]
chunk_text = file_content[start:end] if end > start else file_content[:500]
chunks_with_text.append({
"id": chunk["id"],
"score": chunk["score"],
"text": chunk_text.strip(),
"filename": chunk["filename"],
})
if not chunks_with_text:
return {"answer": "Could not retrieve document content. Please reconnect to Dropbox.", "citations": []}
# Sort by score
chunks_with_text.sort(key=lambda x: x["score"], reverse=True)
# 5. Build prompt and call LLM
from src.prompts.rag_prompt import build_rag_prompt
from src.llm_providers import call_llm
prompt = build_rag_prompt(query=query, chunks=chunks_with_text, k=top_k)
llm_resp = call_llm(prompt=prompt, temperature=0.0, max_tokens=512)
# 6. Build response
citations = [
{"id": c["id"], "score": c["score"], "snippet": c["text"][:200]}
for c in chunks_with_text[:top_k]
]
return {
"answer": llm_resp.get("text", "").strip(),
"citations": citations,
"error": None
}
except Exception as e:
return {
"answer": "",
"citations": [],
"error": str(e)
}
@router.post("/dropbox/token")
async def dropbox_token_exchange(request: dict):
"""
Exchange Dropbox authorization code for access token.
Client secret is kept server-side for security.
"""
code = request.get("code")
redirect_uri = request.get("redirect_uri")
if not code:
return {"error": "No authorization code provided"}
app_key = os.environ.get("DROPBOX_APP_KEY")
app_secret = os.environ.get("DROPBOX_APP_SECRET")
if not app_key or not app_secret:
return {"error": "Dropbox credentials not configured on server"}
try:
async with httpx.AsyncClient() as client:
response = await client.post(
"https://api.dropboxapi.com/oauth2/token",
data={
"grant_type": "authorization_code",
"code": code,
"client_id": app_key,
"client_secret": app_secret,
"redirect_uri": redirect_uri,
}
)
if response.status_code == 200:
return response.json()
else:
return {"error": f"Dropbox API error: {response.text}"}
except Exception as e:
return {"error": str(e)}
@router.post("/dropbox/folder")
async def dropbox_folder(request: dict):
"""
Proxy Dropbox folder API calls to avoid CORS issues.
"""
path = request.get("path", "")
access_token = request.get("access_token")
if not access_token:
return {"error": "No access token provided"}
try:
async with httpx.AsyncClient() as client:
response = await client.post(
"https://api.dropboxapi.com/2/files/list_folder",
json={"path": path, "limit": 100},
headers={
"Authorization": f"Bearer {access_token}",
"Content-Type": "application/json"
}
)
if response.status_code == 200:
return response.json()
else:
return {"error": f"Dropbox API error: {response.text}", "status": response.status_code}
except Exception as e:
return {"error": str(e)}
@router.post("/eval/parsing")
async def eval_parsing(request: dict):
"""
Evaluate Docling parsing on a file from Dropbox.
Request:
- path: Dropbox file path
- access_token: Dropbox access token
Returns parsing metrics and element breakdown.
"""
import tempfile
from pathlib import Path
file_path = request.get("path")
access_token = request.get("access_token")
if not access_token or not file_path:
return {"error": "Missing path or access_token"}
try:
# Download file from Dropbox
async with httpx.AsyncClient(timeout=120.0) as client:
response = await client.post(
"https://content.dropboxapi.com/2/files/download",
headers={
"Authorization": f"Bearer {access_token}",
"Dropbox-API-Arg": f'{{"path": "{file_path}"}}'
}
)
if response.status_code != 200:
return {"error": f"Dropbox download failed: {response.text}"}
# Save to temp file
filename = Path(file_path).name
with tempfile.NamedTemporaryFile(delete=False, suffix=Path(filename).suffix) as tmp:
tmp.write(response.content)
tmp_path = tmp.name
# Run Docling parsing
try:
from src.ingestion.docling_loader import load_document_with_docling
from collections import Counter
doc = load_document_with_docling(tmp_path)
# Count element types
type_counts = Counter(el.element_type for el in doc.elements)
# Sample elements
samples = []
for el in doc.elements[:10]:
samples.append({
"type": el.element_type,
"text": el.text[:200] + "..." if len(el.text) > 200 else el.text,
"level": el.level
})
result = {
"status": doc.status,
"filename": doc.filename,
"format": doc.format,
"total_elements": len(doc.elements),
"total_chars": doc.chars,
"total_words": doc.words,
"page_count": doc.page_count,
"element_types": dict(type_counts),
"sample_elements": samples,
"error": doc.error
}
finally:
# Clean up temp file
import os
os.unlink(tmp_path)
return result
except Exception as e:
return {"error": str(e)}
@router.get("/eval/formats")
async def eval_formats():
"""Get supported document formats for Docling parsing."""
from src.ingestion.api import get_supported_formats
return get_supported_formats()
@router.post("/parse-docling")
async def parse_docling(request: dict):
"""
Parse files with Docling and return COMPLETE output.
Request:
- files: Array of {path, name} objects
- access_token: Dropbox access token
Returns array of parsed documents with ALL elements (not samples).
"""
import tempfile
import os
from pathlib import Path
from collections import Counter
files = request.get("files", [])
access_token = request.get("access_token")
if not access_token or not files:
return {"error": "Missing files or access_token"}
results = []
for file_info in files:
file_path = file_info.get("path")
file_name = file_info.get("name", Path(file_path).name if file_path else "unknown")
if not file_path:
results.append({
"filename": file_name,
"status": "ERROR",
"error": "Missing file path"
})
continue
try:
# Download file from Dropbox
async with httpx.AsyncClient(timeout=180.0) as client:
response = await client.post(
"https://content.dropboxapi.com/2/files/download",
headers={
"Authorization": f"Bearer {access_token}",
"Dropbox-API-Arg": f'{{"path": "{file_path}"}}'
}
)
if response.status_code != 200:
results.append({
"filename": file_name,
"status": "ERROR",
"error": f"Dropbox download failed: {response.text}"
})
continue
# Save to temp file
suffix = Path(file_name).suffix or Path(file_path).suffix
with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
tmp.write(response.content)
tmp_path = tmp.name
try:
from src.ingestion.docling_loader import load_document_with_docling
doc = load_document_with_docling(tmp_path)
# Count element types
type_counts = Counter(el.element_type for el in doc.elements)
# Return ALL elements (not just samples)
all_elements = []
for el in doc.elements:
all_elements.append({
"type": el.element_type,
"text": el.text,
"level": el.level,
"page": getattr(el, 'page', None),
"metadata": getattr(el, 'metadata', {})
})
results.append({
"filename": file_name,
"path": file_path,
"status": doc.status,
"format": doc.format,
"total_elements": len(doc.elements),
"total_chars": doc.chars,
"total_words": doc.words,
"page_count": doc.page_count,
"element_types": dict(type_counts),
"elements": all_elements,
"error": doc.error
})
finally:
os.unlink(tmp_path)
except Exception as e:
results.append({
"filename": file_name,
"status": "ERROR",
"error": str(e)
})
return {"results": results}
@router.post("/dropbox/file")
async def dropbox_file(request: dict):
"""
Proxy Dropbox file download to avoid CORS issues.
Supports text files (.txt, .md) and PDFs with text extraction.
"""
import io
path = request.get("path")
access_token = request.get("access_token")
if not access_token or not path:
return {"error": "Missing path or access_token"}
# Check if file is a PDF
is_pdf = path.lower().endswith('.pdf')
try:
async with httpx.AsyncClient(timeout=60.0) as client:
response = await client.post(
"https://content.dropboxapi.com/2/files/download",
headers={
"Authorization": f"Bearer {access_token}",
"Dropbox-API-Arg": f'{{"path": "{path}"}}'
}
)
if response.status_code == 200:
if is_pdf:
# Extract text from PDF
try:
from PyPDF2 import PdfReader
pdf_file = io.BytesIO(response.content)
reader = PdfReader(pdf_file)
text_parts = []
for page in reader.pages:
page_text = page.extract_text()
if page_text:
text_parts.append(page_text)
content = "\n\n".join(text_parts)
if not content.strip():
return {"error": "PDF contains no extractable text (may be scanned/image-based)"}
return {"content": content}
except Exception as pdf_err:
return {"error": f"PDF extraction failed: {str(pdf_err)}"}
else:
# Return text content directly
return {"content": response.text}
else:
return {"error": f"Dropbox API error: {response.text}", "status": response.status_code}
except Exception as e:
return {"error": str(e)}