Evidence-backed facts
- vLLM documents named and automatic tool calling through its OpenAI-compatible Chat Completions server for supported models and parsers. informative
- The server emits tool-call arguments, while the client is responsible for executing tools, returning results, and enforcing application policy. informative
- Server startup flags, model templates, parser selection, API authentication, tool credentials, and network controls must be configured independently. informative
- Tool-call syntax support does not prove a model follows schemas reliably, a server is secured, an external action occurred, or a production endpoint exists. informative