We don't publish
your competitive advantage.

IFOpenAIEmbeddings calls sporadically fail with APIConnectionError and ProxyError involving SSL/TLS connection closed.

THENSet the OPENAI_PROXY environment variable to your corporate proxy URL (e.g., 'http://your-proxy:8080') before instantiating OpenAIEmbeddings. If not behind a proxy, retrying the request may resolve transient network issues.

IFAzureChatOpenAI ignores proxy environment variables or direct proxy parameter, causing requests to time out when behind a corporate proxy.

THENManually create an httpx.Client and httpx.AsyncClient with the desired proxy URL, then pass them as http_client and http_async_client parameters to AzureChatOpenAI. This workaround ensures all outgoing requests use the specified proxy.

IFOllama LLM returns HTTP 500 error when a global proxy (e.g., Privoxy) is enabled.

THENExplicitly set the `base_url` parameter of the Ollama client to `http://site_1:11434` (or the correct URL) to bypass proxy interference. Ensure the proxy does not intercept site_2 traffic.

IFAzureChatOpenAI fails to honor the openai_proxy parameter or HTTP_PROXY environment variable, causing requests to bypass corporate proxy and resulting in timeouts.

THENManually create httpx.Client and httpx.AsyncClient instances with the desired proxy URL and pass them as http_client and http_async_client parameters to AzureChatOpenAI. Ensure both synchronous and asynchronous clients are provided to support all request paths.

IFUsing Ollama LLM via LlamaIndex with a global proxy (e.g., Privoxy) active causes HTTP 500 error when connecting to site_1:11434.

THENExplicitly set the `base_url` parameter to 'http://site_1:11434' in the Ollama model instantiation, or configure the proxy to bypass site_2 addresses.

IFWhen using GitHub Copilot provider via LiteLLM Proxy, the error 'missing Editor-Version header for IDE auth' appears despite providing the header in the request.

THENSet the required 'Editor-Version' and 'Copilot-Integration-Id' headers in the model configuration's `extra_headers` parameter rather than relying on request-level headers, because the proxy may strip or fail to forward them.

IFWhen using Vercel AI Gateway with an Anthropic model that supports thinking (e.g., claude-sonnet-4.5), the proxy returns a 500 error because the 'thinking' parameter is not supported by the gateway.

THENSet 'litellm.drop_params=True' in the proxy configuration (under 'litellm_settings') to automatically drop unsupported parameters like 'thinking'. Alternatively, if you need the thinking parameter to pass through, send 'allowed_openai_params': ['thinking'] in the request. The error message provides explicit guidance for both approaches.

IFWhen using the OpenAI site_1 SDK with a LiteLLM model configured for GitHub Copilot, the IDE auth header error occurs because the SDK does not automatically forward the required headers.

THENPass 'Editor-Version' and 'Copilot-Integration-Id' inside a `ModelSettings` object's `extra_headers` field when creating the Agent, ensuring the headers are included in all requests.

IFOpenAIEmbeddings calls sporadically fail with APIConnectionError due to proxy/network issues.

THENSet the OPENAI_PROXY environment variable to your proxy URL, or configure a custom httpx client with proxy settings. Additionally, increase the request_timeout parameter to allow more time for the connection.

IFAzureChatOpenAI ignores the proxy set via openai_proxy parameter or environment variables, causing requests to timeout when behind a corporate proxy.

THENInstead of relying on the openai_proxy parameter, create an httpx.Client and httpx.AsyncClient with the proxy URL, and pass them via the http_client and http_async_client arguments to AzureChatOpenAI.