VVectorNodeinvectronode.hashnode.dev·Aug 18 · 5 min readKeep OpenAI-Compatible API Keys Out of Browser CodeA browser cannot keep a provider API key secret. If JavaScript can read the key, users, extensions, developer tools, and injected scripts can read it too. Obfuscation does not change that boundary. Ne00
VVectorNodeinvectronode.hashnode.dev·Aug 17 · 6 min readReuse HTTP Connections for OpenAI-Compatible APIs in PythonA one-off API script can create a client, send one request, and exit. A long-running worker should not rebuild its HTTP connection state for every prompt. Repeated client creation can repeat DNS looku00
VVectorNodeinvectronode.hashnode.dev·Aug 16 · 11 min readHow to Build a Compatibility Harness for OpenAI-Compatible APIs in PythonAn OpenAI-compatible endpoint can accept the same URL pattern, Bearer header, and chat request while still behaving differently after the first successful call. One provider may return a string in cho00
VVectorNodeinvectronode.hashnode.dev·Aug 15 · 11 min readHow to Rotate an OpenAI-Compatible API Key Without Dropping RequestsAPI keys should be replaceable. In practice, teams often discover that a key is visible in a log, a screenshot, a CI artifact, or an old developer laptop only after the service has grown around it. Th00
VVectorNodeinvectronode.hashnode.dev·Aug 14 · 11 min readHow to Retry OpenAI-Compatible API Requests Without Making Failures WorseYour API call worked in a quickstart. Then production returned a mix of 401, model_not_found, 429, and timeout errors. A common first fix is to wrap every request in a retry loop. That can make the in00