Channel and speech hooks¶

pynchy_create_channel¶

Provide a communication channel. Pynchy collects non-None channels and host configuration selects the active connections.

@hookimpl
def pynchy_create_channel(self, context: ChannelPluginContext) -> Channel | None:
    token = os.environ.get("TELEGRAM_BOT_TOKEN")
    if not token:
        return None
    return TelegramChannel(bot_token=token, on_message=context.on_message_callback)

ChannelPluginContext supplies callbacks for inbound messages, chat metadata, outbound text, reactions, structured answers, approval decisions, and workspace lookups. Return a Channel that implements connect, send_event, is_connected, owns_jid, disconnect, reconnect, and prepare_shutdown.

Channel plugins run persistently in the host process with filesystem and network access. Treat them as the highest-risk plugin category; see Security architecture.

pynchy_speech_synthesizer¶

Provide a host-side synthesizer for final spoken replies:

@hookimpl
def pynchy_speech_synthesizer(self) -> SpeechSynthesizer | None:
    return MySpeechSynthesizer()

Pynchy collects valid providers and uses the first. The provider exposes name, async synthesize(text, output_path), and async health(). Channels receive the selected provider in ChannelPluginContext.speech_synthesizer and must tolerate None instead of choosing their own fallback. See Voice and speech.