I’m connecting a DeepSeek model to a client that can run tools such as database lookups and internal search. Some models return more than one tool call in a single assistant turn, but I’m not sure what to expect from DeepSeek or how consistently different model and API versions behave.

For those who have implemented this, can DeepSeek request multiple tools at once, or should the client expect one call, execute it, and then ask the model what to do next? I’m trying to understand the right way to handle multiple tool calls without dropping arguments or producing duplicate actions.