Hermes Agent
Let Hermes listen, transcribe, and reply with natural speech through AllModels.
Hermes Speech connects Hermes Agent to AllModels. It lets Hermes understand voice messages, reply out loud, find voices and speech models, and manage your AllModels balance through normal conversation.
Install and set up
Send this message to your Hermes agent:
Install this plugin and set up AllModels: https://github.com/allmodels-io/hermes-speech
Copy this message and send it to Hermes.
After installation, restart the Hermes gateway so it loads the plugin.
Talk to Hermes
Choose where you use Hermes:
- Start the classic terminal interface with
hermes --cli. - Enter
/voice onto turn on microphone input. - Enter
/voice ttsto make Hermes speak its replies. - Press
Ctrl+B, then start talking. - Pause when you are finished. Hermes detects the silence, sends what you said, and replies out loud.
Hermes keeps the conversation going and listens again after each reply. Press Ctrl+B during recording to stop, or enter /voice off to leave voice mode.
- Start the terminal UI with
hermes --tui. - Enter
/voice onto turn on microphone input. - Enter
/voice ttsto make Hermes speak its replies. - Press
Ctrl+B, then start talking. - Pause when you are finished. Hermes detects the silence, sends what you said, and replies out loud.
The TUI uses the same voice controls as the classic CLI. Press Ctrl+B during recording to stop, or enter /voice off to leave voice mode.
- Open Hermes Desktop, or launch it with
hermes desktop. - Open a chat and click the Start voice conversation waveform button beside the message box.
- Allow microphone access if your computer asks.
- Start talking. Hermes detects when you pause, sends what you said, and replies out loud.
- Continue speaking when Hermes finishes. Say “stop” or click End when you are done.
- Open your chat with the Hermes bot.
- Enter
/voice onto get spoken replies when you send voice notes. Use/voice ttsinstead if you want every reply spoken, including replies to typed messages. - Press and hold Telegram's microphone button, record your message, then release it to send the voice note.
- Hermes transcribes the note and responds with a playable voice message.
Enter /voice off whenever you want text-only replies again.
- Join the Discord voice channel where you want to talk to Hermes.
- In a text channel where the Hermes bot is present, enter
/voice join. Hermes joins the same voice channel you are in. - Start talking. Hermes listens, posts a transcript in that text channel, and speaks its reply in the voice channel.
- Enter
/voice leavein the text channel when you are done.
For ordinary Discord voice messages in a DM or text channel, send a voice message and enter /voice tts if you want Hermes to reply with audio.
What you can ask Hermes to do
Once the plugin is installed, ask for what you want in plain language:
| Try saying | What Hermes does |
|---|---|
| “Set up AllModels speech for me.” | Signs you in, then chooses compatible starting models and a voice. |
| “Find me a British woman’s voice.” | Searches the available voices and gives you matching choices. |
| “Preview that voice saying ‘Welcome back.’” | Makes a sample without changing your current voice. |
| “Switch my speech model to OpenAI 4o.” | Finds matching speech models and helps you choose the right one. |
| “Switch my transcription model to Soniox.” | Finds matching transcription models and lets you choose one. |
| “Speak a little faster.” | Adjusts how quickly Hermes speaks. You can also ask for a specific speed such as 1.1×. |
| “For transcription, expect a French accent and Canadian place names like Montréal and Québec.” | Saves that context to help future transcriptions. |
| “Check my AllModels balance.” | Shows your available paid and promotional credit. |
| “Top up my AllModels $5.” | Creates a secure $5 checkout link for you. |
| “Update Hermes Speech.” | Installs the latest stable plugin release. Restart the gateway afterward. |
What it does not do
- It does not decide when Hermes should speak. Choose the behavior you want with the
/voicecommands above. - It does not charge you automatically. A top-up request creates a secure checkout link for you to open and complete yourself.
- It does not update until you ask. Tell Hermes “Update Hermes Speech,” then restart the gateway when it finishes.
Sign up yourself if needed
If your agent isn't able to sign up correctly, you can use these /speech commands to sign up yourself:
/speech signup you@example.com
/speech verify 123456Hermes emails you a six-digit code. Replace 123456 with that code, then follow the prompts to choose your speech model, voice, and transcription model.
See the Hermes Speech repository for source code and release information.