effect-uai-agentic-loop

A long-lived chat agent loop buffers incoming user messages in a queue and drains them between turns. This ensures rapid sequential inputs are processed systematically rather than interrupting active execution.

Debouncing rapid typing coalesces bursts of consecutive user messages into a single batch turn. This reduces processing churn by grouping fragmented inputs together, improving overall agent throughput and responsiveness.

Tool calls execute mid-turn and pass through to the next iteration automatically without checking for new user input. This allows the agent to complete tool exchanges seamlessly while maintaining deterministic turn progression.

The input queue agent loop supports WebSocket-based interactive environments, CLI agents, and external input streams. It defines external lifetime control to terminate sessions from outside triggers like WebSocket disconnects or Ctrl-C commands.

The best way to manage persistent conversation loops with external input streams is using queue-based input drainage combined with debounce window control. This approach provides controlled input flow and deterministic turn progression for long-lived agents.