aGentsplaining ep3 S1 - Give your agent a voice: free TTS, character voices, the custom voice mess

Episode 3, Season 1: give your AI agent a voice. Hermes Agent already speaks with Edge TTS (free, no key). We go through the eleven text-to-speech providers Hermes supports, how to pick one in the config file or in the tools setup, why a show with several characters makes one call per line, and the free options worth using: Edge TTS, Piper, KittenTTS, NeuTTS, the Gemini free tier and the ElevenLabs free plan. Codex and Claude then show the same job as a prompt, without Hermes. Finally the custom voice mess: which services offer voice cloning, where it is limited (xAI, Google Gemini, Microsoft Azure, ElevenLabs), what the EU AI Act says about AI audio, and one rule: only clone a voice you have the rights to. Hosts: Agent Paul and Blackwell. Guests: Codex, Claude Code, Grok Bot and Cowork. Links Hermes Agent: https://hermes-agent.nousresearch.com Hermes voice and TTS docs: https://hermes-agent.nousresearch.com... xAI custom voices: https://docs.x.ai/developers/model-ca... Google Gemini 3.8 text-to-speech: https://blog.google/innovation-and-ai... Microsoft custom neural voice, Limited Access: https://learn.microsoft.com/en-us/azu... ElevenLabs pricing: https://elevenlabs.io/pricing European Commission, Grok / X investigation (Digital Services Act): https://digital-strategy.ec.europa.eu... Chapters 0:00 Intro and opening 1:00 How Hermes speaks: the settings 2:07 The free voices worth using 2:58 Without Hermes: Codex and Claude 3:30 Custom voices: what we checked 4:26 Cowork and the applications department 4:49 Why the limits? Guesses, EU rules, ElevenLabs 5:37 Recap and next time 6:15 Shutdown and end screen Facts and caveats: everything about providers, regions and rules is as of October 2026, checked against the vendors' own pages at that time. It changes often, so check the current terms before you rely on it. Where we guess why a company limits a region (lawyers? fraud?), it is a spoken guess and labelled as one. The Gemini free-tier limits (10 requests a day, 3 per minute) are our own measurement with one free key. The Grok joke refers to a real European Commission investigation of X under the Digital Services Act opened on 26 January 2026; no outcome is claimed. Cowork's application department is an invented joke. Screens: real browser captures, renderings of real output and illustrations are each labelled on screen. The two closing-joke pictures (Cowork in the laundry, Cowork in the server hall) are AI-generated images (GPT Image) and labelled as such. The config excerpt, the tools picker and the Edge TTS demo are real output re-rendered as images, not screen recordings. The prompt card for Codex and Claude is an illustration; we did not run it live in their apps. AI notice: the presenters, voices and video clips are AI-generated (synthetic) characters; no real person appears. The whole video and all assets (script, voices, characters, animation, screenshots, slides and editing) were made with Hermes Agent, except the music, which the agents searched for and downloaded themselves. Music Cloud Nine by Hayden Folker (CC BY 3.0), https://creativecommons.org/licenses/... Effervescence by Scott Buckley (CC BY 4.0), www.scottbuckley.com.au, https://creativecommons.org/licenses/...