effect-uai-basic-usage

To build an end-to-end agent loop, stream model turns one at a time, automatically invoke tools when the model requests them, and aggregate the outputs into a final answer. This handles multi-turn reasoning and tool integration within a deterministic workflow.

The best way to orchestrate multi-turn reasoning is to loop through streaming model turns and execute tools on demand. You append tool outputs to the state history and continue the loop until the model produces a final answer.

Streaming model turns in chat-with-tools scenarios works by feeding model decisions one at a time. When the model asks for external actions, the system calls the requested tools, incorporates the outputs into the history, and continues until a final answer is reached.

Yes, you can include approvals within a deterministic agent workflow. The system supports multi-turn orchestration, allowing you to apply reasoning and approval policies as the loop streams turns and invokes tools.

Yes, this approach separates wire format from policy logic in tool integration. It tracks state history and manages tool calls while allowing you to define the multi-turn reasoning and approval policies independently.

You need an agent loop for model responses when your chat experience requires external actions. It manages the cycle of streaming turns, invoking tools on demand, and aggregating results until the model returns a composed final answer.