Give your agent a voice: Fish Audio Drama 3 in Hermes Agent and OpenClaw

Fish Audio released Drama 3, a model that turns written text into spoken audio, on the 13th of October 2026, according to the company's own blog. Both agents that people run on their own servers, OpenClaw and Hermes Agent, already list it in their official documentation. That means a personal agent can now send a spoken daily briefing for under €2.50 a month, by the sums further down.

What Fish Audio announced

Drama 3 is a text to speech model. You give it words, it gives back a recording of a voice saying them. Fish Audio wrote that the voice can be directed with plain instructions placed in square brackets, such as asking for a whisper or a laugh.

The company said the model speaks 83 languages.

The weights are open. In plain words, the trained model itself can be downloaded and run on your own hardware. Fish Audio published it under its Community License, which the company describes as free for research and non commercial use, and free for commercial use when a business and its affiliates earn less than US$500,000 (€446,628) a year. Above that line, a separate licence from Fish Audio is required, according to the same post. The licence also bans cloning a person's voice without consent.

What a voice does inside your own agent

An agent with a voice can reply to you as audio. The OpenClaw documentation says Fish Audio works for voice notes, for its live Talk mode and for phone calls, across the chat channels OpenClaw connects to.

In daily life, that looks like this:

• A morning summary of your inbox and calendar, sent to your phone as a voice note.
• A long report read aloud while you drive or cook.
• Replies in a messaging app that arrive as a short spoken message.

The Hermes Agent documentation says the Fish Audio plugin brings speech in 80 or more languages, speech to text, the Fish Audio voice library, voice cloning and voice design, across the command line, messaging apps and the desktop app.

A spoken briefing at 7 in the morning only arrives if the agent is awake at 7. That is the job of a server that stays on through the night.

How to pick Fish Audio in Hermes Agent

Hermes Agent version 0.22 came out on the 13th of October 2026. Its release notes list Fish Audio on the new Voice shelf of the plugin catalogue, next to Speechify, Deepgram and Gemini Live.

The Hermes Agent documentation describes the Fish Audio plugin as the official one. The steps, in plain words:

1. Create an API key on the Fish Audio website. An API key is a private password that lets your agent use your Fish Audio account.
2. Install the Fish Audio plugin from the Hermes plugin catalogue, or from the settings of Hermes Desktop.
3. Sign the plugin in with your key. Hermes checks that the key works.
4. Choose the Drama 3 preview from the list of Fish Audio models.

How to pick Fish Audio in OpenClaw

The OpenClaw documentation has its own Fish Audio page. As of the 15th of October 2026, it lists a Drama 3 preview model among the Fish Audio voices. The steps, in plain words:

1. Create an API key on the Fish Audio website.
2. Add the Fish Audio plugin to OpenClaw and give it the key.
3. In the OpenClaw settings, set Fish Audio as the speech provider.
4. Pick the Drama 3 preview as the model, and a voice if you want a specific one.

Both agents connect with your own Fish Audio key, so every spoken reply counts against your own Fish Audio plan.

What it costs, in human scale

Fish Audio charges by the number of characters it turns into speech. A character is a single letter, digit, space or punctuation mark. All euro amounts below use the European Central Bank reference rate of the 14th of October 2026, 1 euro to 1.1195 US dollars.

Fish Audio said the normal price for Drama 3 is $50 (€44.66) per million characters. A launch price of $10 (€8.93) per million characters runs from the 13th of October to the 12th of November 2026.

Here is the math for a daily spoken briefing. A two minute briefing is about 300 spoken words. At about 6 characters per English word, counting the space, that is about 1,800 characters. Over 30 days, that is 54,000 characters.

• At the launch price: 54,000 divided by 1,000,000, times $10, gives $0.54 (€0.48) for the month.
• At the normal price: the same 54,000 characters times $50 per million give $2.70 (€2.41) for the month.

A heavier user who gets ten spoken replies a day of about 500 characters each reaches 150,000 characters a month. That comes to $1.50 (€1.34) at the launch price and $7.50 (€6.70) at the normal price.

The word and character counts above are estimates for typical English speech. Fish Audio's price per million characters is the fixed part of the sum.

For people on OneClickClaw

People who run their OpenClaw or Hermes Agent on OneClickClaw can add their own Fish Audio key the same way. Plans start at Starter, EUR 14.99 per month, and every plan opens with a seven day free trial.