Streaming

The bot streams replies by default to feel more human. It maintains the typing indicator and sends natural “bursts” rather than editing messages.

How it works - Uses OpenAI Responses streaming to receive text deltas - Buffers tokens and splits on sentence/paragraph boundaries - Sends the first burst ASAP, then roughly every 2 lines, with ~1 msg/sec pacing and slight jitter - Respects Discord’s 2000 character limit per message - Falls back to non-streaming if an error occurs

Tuning - Persona YAML streaming: controls the pacing: - rate_hz: messages/s - min_first: characters before the first send - min_next: characters before subsequent sends

Disabling streaming - CLI: --no-stream (or --stream=false on Python 3.9+)

Notes - Code blocks: for heavy code output, consider raising min_next. - Rate limits: basic 429 backoff is built in.