https://arxiv.org/abs/2606.28344. The end of web parsing. The beginning of scalable pixel-native search. link: https://pixelrag.ai/
fix(serve): bound n_docs to prevent unbounded-search OOM/DoS (#122)
Shubham Srivastava3688a5eaf838
diff --git a/serve/src/pixelrag_serve/api.py b/serve/src/pixelrag_serve/api.pyindex a0a3008..0271ff9 100644--- a/serve/src/pixelrag_serve/api.py+++ b/serve/src/pixelrag_serve/api.py@@ -52,7 +52,7 @@ from fastapi import FastAPI, HTTPException, Requestfrom fastapi.middleware.cors import CORSMiddlewarefrom fastapi.responses import FileResponse, Responsefrom PIL import Image-from pydantic import BaseModel+from pydantic import BaseModel, Field_request_id_ctx: contextvars.ContextVar[str] = contextvars.ContextVar("request_id",@@ -172,7 +172,9 @@ class Query(BaseModel):class SearchRequest(BaseModel):queries: list[Query]- n_docs: int = 10+ # Bounded: fetch_k = n_docs * 10 feeds FAISS search k, so an unbounded value+ # is a trivial OOM/DoS on the public endpoint. Largest real caller uses 10.+ n_docs: int = Field(default=10, ge=1, le=1000)nprobe: int | None = None # override default nprobemin_tile_height: int | None = None # filter out small/blank chunksinstruction: str | None = None # override query embedding instruction