SaaS fixes (#6578)

Co-authored-by: Claude Opus 4.8 <[email protected]>
Co-authored-by: James Brunton <[email protected]>
Co-authored-by: Reece Browne <[email protected]>
Co-authored-by: ConnorYoh <[email protected]>
Co-authored-by: Reece <[email protected]>
Co-authored-by: EthanHealy01 <[email protected]>
Co-authored-by: Ludy <[email protected]>

engine/tests/conftest.py

@@ -5,7 +5,7 @@ from pathlib import Path
 
 import pytest
 
-from stirling.config import AppSettings, RagBackend, load_settings
+from stirling.config import AppSettings, DocumentsBackend, load_settings
 from stirling.services import build_runtime
 from stirling.services.runtime import AppRuntime
 
@@ -23,10 +23,13 @@ def build_app_settings() -> AppSettings:
         fast_model_name="test",
         smart_model_max_tokens=8192,
         fast_model_max_tokens=2048,
-        rag_backend=RagBackend.SQLITE,
+        model_max_concurrency=32,
+        documents_backend=DocumentsBackend.SQLITE,
         rag_embedding_model="voyageai:voyage-4",
-        rag_store_path=Path(":memory:"),
-        rag_pgvector_dsn="",
+        documents_sqlite_path=Path(":memory:"),
+        documents_pgvector_dsn="",
+        documents_pgvector_pool_min_size=1,
+        documents_pgvector_pool_max_size=10,
         rag_chunk_size=512,
         rag_chunk_overlap=64,
         rag_default_top_k=5,
@@ -41,6 +44,7 @@ def build_app_settings() -> AppSettings:
         contradiction_canonicaliser_batch_size=500,
         max_pages=200,
         max_characters=200_000,
+        require_user_id=False,
         posthog_enabled=False,
         posthog_api_key="",
         posthog_host="https://eu.i.posthog.com",

engine/tests/pdf_comment/test_routes.py

@@ -16,7 +16,7 @@ from fastapi.testclient import TestClient
 
 from stirling.api import app
 from stirling.api.dependencies import get_pdf_comment_agent
-from stirling.config import AppSettings, RagBackend, load_settings
+from stirling.config import AppSettings, DocumentsBackend, load_settings
 from stirling.contracts.pdf_comments import (
     PdfCommentInstruction,
     PdfCommentRequest,
@@ -35,10 +35,13 @@ class StubSettingsProvider:
             fast_model_name="test",
             smart_model_max_tokens=8192,
             fast_model_max_tokens=2048,
-            rag_backend=RagBackend.SQLITE,
+            model_max_concurrency=32,
+            documents_backend=DocumentsBackend.SQLITE,
             rag_embedding_model="test-embed",
-            rag_store_path=Path(":memory:"),
-            rag_pgvector_dsn="",
+            documents_sqlite_path=Path(":memory:"),
+            documents_pgvector_dsn="",
+            documents_pgvector_pool_min_size=1,
+            documents_pgvector_pool_max_size=10,
             rag_chunk_size=512,
             rag_chunk_overlap=64,
             rag_default_top_k=5,
@@ -49,6 +52,7 @@ class StubSettingsProvider:
             chunked_reasoner_notes_char_budget=250_000,
             max_pages=100,
             max_characters=100_000,
+            require_user_id=False,
             posthog_enabled=False,
             posthog_api_key="",
             posthog_host="https://eu.i.posthog.com",

engine/tests/test_model_concurrency.py

@@ -0,0 +1,58 @@
+from __future__ import annotations
+
+import asyncio
+
+import pytest
+from pydantic_ai.messages import ModelMessage, ModelResponse, TextPart
+from pydantic_ai.models import Model, ModelRequestParameters
+from pydantic_ai.settings import ModelSettings
+
+from stirling.services.runtime import ConcurrencyLimitedModel
+
+
+class TrackingModel(Model):
+    """Records the high-water mark of concurrent in-flight requests."""
+
+    def __init__(self) -> None:
+        super().__init__()
+        self.active = 0
+        self.max_active = 0
+
+    @property
+    def model_name(self) -> str:
+        return "tracking"
+
+    @property
+    def system(self) -> str:
+        return "test"
+
+    async def request(
+        self,
+        messages: list[ModelMessage],
+        model_settings: ModelSettings | None,
+        model_request_parameters: ModelRequestParameters,
+    ) -> ModelResponse:
+        self.active += 1
+        self.max_active = max(self.max_active, self.active)
+        # Yield twice so every gathered task gets a chance to be in flight
+        # together before any of them completes.
+        await asyncio.sleep(0)
+        await asyncio.sleep(0)
+        self.active -= 1
+        return ModelResponse(parts=[TextPart(content="ok")])
+
+
+@pytest.mark.anyio
+async def test_shared_semaphore_caps_concurrency_across_models() -> None:
+    inner = TrackingModel()
+    semaphore = asyncio.Semaphore(2)
+    fast = ConcurrencyLimitedModel(inner, semaphore)
+    smart = ConcurrencyLimitedModel(inner, semaphore)
+    params = ModelRequestParameters()
+
+    await asyncio.gather(
+        *(fast.request([], None, params) for _ in range(5)),
+        *(smart.request([], None, params) for _ in range(5)),
+    )
+
+    assert inner.max_active == 2

engine/tests/test_stirling_api.py

@@ -214,6 +214,26 @@ def test_pdf_edit_route() -> None:
     assert response.json()["outcome"] == "cannot_do"
 
 
+def test_routes_require_user_id_when_enforced() -> None:
+    """With STIRLING_REQUIRE_USER_ID on, an identity-less request is rejected
+    at the boundary before any handler runs; supplying X-User-Id is accepted.
+    The other route tests in this module run with the flag off and cover the
+    identity-less self-hosted path."""
+    app.dependency_overrides[load_settings] = lambda: build_app_settings().model_copy(update={"require_user_id": True})
+    try:
+        anonymous = client.post("/api/v1/pdf/edit", json={"userMessage": "rotate this"})
+        identified = client.post(
+            "/api/v1/pdf/edit",
+            json={"userMessage": "rotate this"},
+            headers={"X-User-Id": "alice"},
+        )
+    finally:
+        app.dependency_overrides[load_settings] = build_app_settings
+
+    assert anonymous.status_code == 401
+    assert identified.status_code == 200
+
+
 def test_pdf_questions_route() -> None:
     response = client.post(
         "/api/v1/pdf/questions",

engine/tests/test_stirling_contracts.py

@@ -77,17 +77,20 @@ def test_pdf_question_answer_defaults_evidence_list() -> None:
 def test_app_settings_accepts_model_configuration() -> None:
     from pathlib import Path
 
-    from stirling.config import RagBackend
+    from stirling.config import DocumentsBackend
 
     settings = AppSettings(
         smart_model_name="claude-sonnet-4-5-20250929",
         fast_model_name="claude-haiku-4-5-20251001",
         smart_model_max_tokens=8192,
         fast_model_max_tokens=2048,
-        rag_backend=RagBackend.SQLITE,
+        model_max_concurrency=32,
+        documents_backend=DocumentsBackend.SQLITE,
         rag_embedding_model="voyageai:voyage-4",
-        rag_store_path=Path(":memory:"),
-        rag_pgvector_dsn="",
+        documents_sqlite_path=Path(":memory:"),
+        documents_pgvector_dsn="",
+        documents_pgvector_pool_min_size=1,
+        documents_pgvector_pool_max_size=10,
         rag_chunk_size=512,
         rag_chunk_overlap=64,
         rag_default_top_k=5,
@@ -98,6 +101,7 @@ def test_app_settings_accepts_model_configuration() -> None:
         chunked_reasoner_notes_char_budget=250_000,
         max_pages=200,
         max_characters=200_000,
+        require_user_id=False,
         posthog_enabled=False,
         posthog_api_key="",
         posthog_host="https://eu.i.posthog.com",