Speechify for Hermes Agent

Give Hermes Agent a Speechify voice.

This plugin adds Speechify as a text-to-speech provider. Once it is on, Hermes speaks its replies with Speechify's Simba 3 models: in CLI voice mode, in Hermes Desktop, and as voice notes on Telegram, Discord, WhatsApp and the other gateway platforms.

Low latency. Audio streams from POST /v1/audio/stream, and in voice mode Hermes starts playing while the rest of the sentence is still being made (Hermes 0.21.6 or newer).
Voice bubbles that just work. Ogg output is Opus, which Telegram and Discord play inline.
Every voice your key can use, including your own cloned voices.
No extra Python packages. It uses requests, which Hermes already ships.

Install

You need a Speechify API key. Create one at console.speechify.ai.

hermes plugins install Lokeshm24/hermes-speechify
hermes plugins enable speechify

Hermes asks for SPEECHIFY_API_KEY during install and saves it to ~/.hermes/.env. To set it yourself:

echo 'SPEECHIFY_API_KEY=your-key-here' >> ~/.hermes/.env

Turn it on

In ~/.hermes/config.yaml:

tts:
  provider: speechify

That's all you need. Hermes now speaks with the geffen_32 voice on simba-3.2. Restart Hermes (or the gateway) after changing the config.

Pick a voice

List every voice your key can use:

hermes speechify voices
hermes speechify voices --locale en-GB
hermes speechify voices --model simba-3.2
hermes speechify voices --search oliver

Hear one before you choose it:

hermes speechify say "Hi, I'm your new voice." --voice geffen_32

This prints the path of the audio file it wrote.

Then set it:

tts:
  provider: speechify
  speechify:
    voice: geffen_32

Your own cloned voices show up in the list too, marked "(your clone)".

All settings

Everything goes under tts.speechify and is optional.

tts:
  provider: speechify
  output_format: mp3          # mp3 (default), ogg, opus or wav
  speed: 1.0                  # 0.5 to 4.0, sent as SSML <prosody rate>
  speechify:
    voice: geffen_32          # any id from `hermes speechify voices`
    model: simba-3.2          # simba-3.2 (English) or simba-3.0 (multilingual)
    language: en-US           # optional; a non-English value defaults the model to simba-3.0
    text_normalization: true  # optional; spell out numbers, dates, units
    loudness_normalization: false
    timeout: 60               # seconds
    max_text_length: 4000     # Hermes splits longer replies into chunks (API max 20000)

Hermes's shared tts.voice and tts.model keys also work, and win over the tts.speechify ones.

Models

Model	Languages	Notes
`simba-3.2` (default)	English	Lowest latency, most expressive
`simba-3.0`	English, German, Spanish, French, Italian, Portuguese	Use it for non-English voices

simba-3.2 refuses non-English voices. If you set language to something that is not English, the plugin uses simba-3.0 unless you name a model yourself.

Audio formats

`output_format`	What you get
`mp3`	MP3, 24 kHz, 128 kbps
`ogg` / `opus`	Opus in Ogg, 24 kHz. Plays as a voice bubble with no conversion
`wav`	16-bit PCM WAV, 24 kHz
`flac`	Not offered by the API. You get WAV instead

For live voice mode the plugin streams raw 24 kHz PCM to Hermes's speaker.

Using Hermes on Telegram, Discord or WhatsApp? Set output_format: ogg. Speechify then sends Opus directly, so replies arrive as voice bubbles even when ffmpeg is not installed. With mp3, Hermes needs ffmpeg to convert them.

How it talks to Speechify

POST https://api.speechify.ai/v1/audio/stream to make speech.
GET https://api.speechify.ai/v1/voices to list voices (only when you ask for them).
Every request sends Speechify-Version: 2026-09-30, so the plugin behaves the same no matter which API version your workspace has stored.

What leaves your machine: the text Hermes speaks, your API key, and the voice and model you picked. They go to api.speechify.ai (or the base_url you set) and nowhere else. Speechify bills synthesis to your workspace by character. There is no telemetry.

Troubleshooting

You see	Do this
`SPEECHIFY_API_KEY is not set`	Add the key to `~/.hermes/.env`
`HTTP 401 unauthorized`	The key is wrong or revoked. Make a new one
`HTTP 403 insufficient_scope`	The key cannot use text-to-speech. Make a key with audio access
`HTTP 404 voice_not_found`	Run `hermes speechify voices` and pick an id from the list
`HTTP 400` naming the model	`simba-3.2` is English only. Use `simba-3.0` for other languages
`HTTP 402`	The workspace is out of credits
Hermes still uses Edge	Check `hermes plugins list` shows `speechify` enabled, then restart

Every error includes Speechify's request id. Include it when you contact Speechify support.

Development

git clone https://github.com/NousResearch/hermes-agent   # for the real base class
uv venv && uv pip install -e . pytest ruff
HERMES_SRC=./hermes-agent bash scripts/install-hermes-contract.sh
uv run pytest                                  # unit tests, no network
SPEECHIFY_API_KEY=... uv run pytest -m live    # real API calls
hermes plugins validate .                      # the catalog admission check

License

MIT. See LICENSE.

speechify

What it adds

Environment variables it needs 1

README