diff --git a/surfsense_backend/tests/conftest.py b/surfsense_backend/tests/conftest.py
index effde2714..868664ca8 100644
--- a/surfsense_backend/tests/conftest.py
+++ b/surfsense_backend/tests/conftest.py
@@ -2,11 +2,14 @@
 
 from __future__ import annotations
 
-import contextlib
+import os
 from collections.abc import AsyncGenerator
+from pathlib import Path
 
+import asyncpg
 import httpx
 import pytest
+from dotenv import load_dotenv
 
 from tests.utils.helpers import (
     BACKEND_URL,
@@ -14,9 +17,38 @@ from tests.utils.helpers import (
     delete_document,
     get_auth_token,
     get_search_space_id,
-    poll_document_status,
 )
 
+load_dotenv(Path(__file__).resolve().parent.parent / ".env")
+
+DATABASE_URL = os.environ.get(
+    "TEST_DATABASE_URL",
+    os.environ.get("DATABASE_URL", ""),
+).replace("postgresql+asyncpg://", "postgresql://")
+
+
+async def _force_delete_documents_db(
+    search_space_id: int,
+) -> int:
+    """
+    Bypass the API and delete documents directly from the database.
+
+    This handles stuck documents in pending/processing state that the API
+    refuses to delete (409 Conflict). Chunks are cascade-deleted by the
+    foreign key constraint.
+
+    Returns the number of deleted rows.
+    """
+    conn = await asyncpg.connect(DATABASE_URL)
+    try:
+        result = await conn.execute(
+            "DELETE FROM documents WHERE search_space_id = $1",
+            search_space_id,
+        )
+        return int(result.split()[-1])
+    finally:
+        await conn.close()
+
 
 @pytest.fixture(scope="session")
 def backend_url() -> str:
@@ -43,68 +75,18 @@ async def search_space_id(backend_url: str, auth_token: str) -> int:
 
 @pytest.fixture(scope="session", autouse=True)
 async def _purge_test_search_space(
-    backend_url: str,
-    auth_token: str,
     search_space_id: int,
 ):
     """
     Delete all documents in the test search space before the session starts.
 
-    Ensures no stale data from a previous run interferes with the current
-    session.  Paginates through all documents and waits for any in-flight
-    documents to reach a terminal state before deleting.
+    Uses direct database access to bypass the API's 409 protection on
+    pending/processing documents. This ensures stuck documents from
+    previous crashed runs are always cleaned up.
     """
-    hdrs = auth_headers(auth_token)
-    async with httpx.AsyncClient(base_url=backend_url, timeout=60.0) as client:
-        all_docs: list[dict] = []
-        page = 0
-        page_size = 200
-
-        while True:
-            resp = await client.get(
-                "/api/v1/documents",
-                headers=hdrs,
-                params={
-                    "search_space_id": search_space_id,
-                    "page": page,
-                    "page_size": page_size,
-                },
-            )
-            if resp.status_code != 200:
-                break
-
-            body = resp.json()
-            all_docs.extend(body.get("items", []))
-
-            if not body.get("has_more", False):
-                break
-            page += 1
-
-        if not all_docs:
-            yield
-            return
-
-        in_flight = [
-            doc["id"]
-            for doc in all_docs
-            if doc.get("status", {}).get("state") in ("pending", "processing")
-        ]
-        if in_flight:
-            with contextlib.suppress(Exception):
-                await poll_document_status(
-                    client,
-                    hdrs,
-                    in_flight,
-                    search_space_id=search_space_id,
-                    timeout=120.0,
-                )
-
-        for doc in all_docs:
-            with contextlib.suppress(Exception):
-                await client.delete(
-                    f"/api/v1/documents/{doc['id']}",
-                    headers=hdrs,
-                )
+    deleted = await _force_delete_documents_db(search_space_id)
+    if deleted:
+        print(f"\n[purge] Deleted {deleted} stale document(s) from search space {search_space_id}")
 
     yield
 
@@ -134,13 +116,30 @@ def cleanup_doc_ids() -> list[int]:
 async def _cleanup_documents(
     client: httpx.AsyncClient,
     headers: dict[str, str],
+    search_space_id: int,
     cleanup_doc_ids: list[int],
 ):
     """
-    Runs after every test.  Deletes any document IDs that were appended to
-    the ``cleanup_doc_ids`` list during the test body.
+    Runs after every test. Tries the API first for clean deletes, then
+    falls back to direct DB access for any stuck documents.
     """
     yield
+
+    remaining_ids: list[int] = []
     for doc_id in cleanup_doc_ids:
-        with contextlib.suppress(Exception):
-            await delete_document(client, headers, doc_id)
+        try:
+            resp = await delete_document(client, headers, doc_id)
+            if resp.status_code == 409:
+                remaining_ids.append(doc_id)
+        except Exception:
+            remaining_ids.append(doc_id)
+
+    if remaining_ids:
+        conn = await asyncpg.connect(DATABASE_URL)
+        try:
+            await conn.execute(
+                "DELETE FROM documents WHERE id = ANY($1::int[])",
+                remaining_ids,
+            )
+        finally:
+            await conn.close()