~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
Python academiaanthropicarxivbravedeep-researchencryptionhome-automationhomeserverlocallocal-deep-researchlocal-llmmistralollamaopenaipubmedresearchresearch-toolretrieval-augmented-generationsearxngself-hosted
fix(embeddings): normalize OpenAI base_url before hostname check (#4310)
LearningCircuita387f65a846f
diff --git a/src/local_deep_research/embeddings/providers/implementations/openai.py b/src/local_deep_research/embeddings/providers/implementations/openai.pyindex 553ade1a99..f95c125455 100644--- a/src/local_deep_research/embeddings/providers/implementations/openai.py+++ b/src/local_deep_research/embeddings/providers/implementations/openai.py@@ -7,6 +7,7 @@ from langchain_core.embeddings import Embeddingsfrom loguru import loggerfrom ....config.thread_settings import get_setting_from_snapshot+from ....utilities.url_utils import normalize_urlfrom ..base import BaseEmbeddingProvider@@ -127,6 +128,12 @@ class OpenAIEmbeddingsProvider(BaseEmbeddingProvider):}if base_url:+ # Normalize first so a scheme-less entry like "api.openai.com"+ # parses to a hostname (urlparse otherwise returns hostname=None+ # for bare hosts, which would silently drop the ctx-length guard+ # for the real OpenAI endpoint). Mirrors the LLM-side OpenAI+ # provider, which already normalizes via the same helper.+ base_url = normalize_url(base_url)params["openai_api_base"] = base_url# Disable client-side context length checks only for non-OpenAI# hosts (LM Studio, vLLM, llama.cpp, etc.) which may lack tiktoken@@ -214,7 +221,7 @@ class OpenAIEmbeddingsProvider(BaseEmbeddingProvider):client_kwargs: Dict[str, Any] = {"api_key": api_key}if base_url:- client_kwargs["base_url"] = base_url+ client_kwargs["base_url"] = normalize_url(base_url)client = OpenAI(**client_kwargs)models_response = client.models.list()diff --git a/tests/embeddings/test_openai_embeddings.py b/tests/embeddings/test_openai_embeddings.pyindex 7272434142..3198e620ba 100644--- a/tests/embeddings/test_openai_embeddings.py+++ b/tests/embeddings/test_openai_embeddings.py@@ -201,6 +201,39 @@ class TestOpenAIEmbeddingsProviderCreateEmbeddings:)assert "check_embedding_ctx_length" not in call_kwargs+ def test_create_embeddings_with_schemeless_openai_base_url_keeps_ctx_check(+ self,+ ):+ """A scheme-less base_url like ``api.openai.com`` must still be+ recognized as the real OpenAI endpoint. ``urlparse`` on a bare+ host returns ``hostname=None``, so without URL normalization the+ ctx-length guard would silently be dropped for the cloud API."""+ from local_deep_research.embeddings.providers.implementations.openai import (+ OpenAIEmbeddingsProvider,+ )++ mock_embeddings = MagicMock()++ with patch(+ "local_deep_research.embeddings.providers.implementations.openai.get_setting_from_snapshot",+ return_value=None,+ ):+ with patch(+ "langchain_openai.OpenAIEmbeddings",+ return_value=mock_embeddings,+ ) as mock_class:+ OpenAIEmbeddingsProvider.create_embeddings(+ api_key="test-key",+ base_url="api.openai.com",+ )++ call_kwargs = mock_class.call_args[1]+ # normalize_url prepends https:// for external hosts.+ assert (+ call_kwargs["openai_api_base"] == "https://api.openai.com"+ )+ assert "check_embedding_ctx_length" not in call_kwargs+def test_create_embeddings_with_dimensions(self):"""Test creating embeddings with custom dimensions for v3 model."""from local_deep_research.embeddings.providers.implementations.openai import (