is there a way to say a quick message over the phone without having to start an agent?

Thanks for reaching out, Ed. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

theres a certain feature flag we check during inbound calls to decide if we should continue with the inbound call. when we decide not to take the inbound calls, i'd like to say a message to the user like "sorry we can't be reached" or something, but i dont want to have to create and start an agent and session in order to just say one message and end the call.

Thanks — I’m responding to your scenario about wanting to play a brief prerecorded/TTS message on inbound calls when a feature flag says “do not accept” without creating or starting an agent/session. You can do this. High-level options that avoid launching an agent: • Use a dedicated inbound handling flow (dispatch rule) that checks the feature flag first and routes the call to a “message-capture” or “playback” handler instead of an agent. That handler can play a prerecorded audio file or synthesize TTS, wait briefly, then hang up. • Play a prerecorded file or TTS directly on the call media path, record nothing else, and end the connection. This is a simple play-and-disconnect pattern — no agent session required. • If you already use voicemail detection or silence-based logic: after the flag check, open the call briefly, play the message, then either wait a short timeout for additional audio or immediately disconnect. Silence detection is optional here because you’re proactively playing the message. Practical questions so I can give exact steps: 1. Do you want to play a prerecorded audio file or use TTS? 2. Are you using LiveKit’s inbound webhook/dispatch rules today to handle incoming calls? 3. Do you need the play action logged/stored (e.g., for analytics), or is simply playing-and-hanging-up sufficient? If you tell me those details I’ll give a concise implementation path (example handlers/webhook logic and the minimal Media API calls you’d use) tailored to your setup. , with interactive elements

i want to use TTS, do i need to start an agent for that?

Thanks — I’m answering your question about whether you need to start an agent to play a TTS message on an inbound call when the feature flag says “do not accept.” Short answer: No — you do not need to start an agent or AgentSession to synthesize and play a TTS message. You can generate TTS as a standalone component and play it directly into the call, then hang up. How you’d typically implement this (high level): • On inbound call, check your feature flag. • If flag says “do not accept,” route the call to a simple playback handler instead of creating an agent. • Create a standalone TTS stream, push the message text into it, consume the synthesized audio frames, publish/play those frames onto the call’s audio path, then disconnect. Minimal practical options: • Standalone TTS (recommended for dynamic messages): instantiate a TTS stream, push text, stream audio frames into the call, end call. • Pre-synthesized file: generate a WAV beforehand (or at runtime), then play the file into the call and hang up. • Silence/timeout guard (optional): after playback you can wait briefly or use silence detection before disconnecting to ensure the caller heard the message. Things to confirm so I can give exact code/step-by-step instructions: 1. Do you want TTS generated live or are pre-synthesized files acceptable? 2. How are inbound calls currently handled — via a webhook/dispatch rule, or inside an existing call-processing service? 3. Do you need the message transcribed, logged, or stored, or is playing-and-hanging-up enough? If you want, I can produce a concise code example for your stack (e.g., Python snippet showing a standalone TTS stream feeding audio frames into the call) once you tell me whether you use live generation or audio files and how your inbound flow is triggered. Sources: Text-to-speech (TTS) models overview | LiveKit Documentation | Playing Audio | LiveKit Documentation, with interactive elements