nodecode-sound-effects
ElevenLabs sound effects: a text description becomes an audio file the model saves and hands you. A Nodecode rewrite of NousResearch/hermes-official-plugins by NousResearch.
One tool, elevenlabs_sound_effect, turns "glass shattering on a concrete floor" or "90s hip-hop drum loop, 90 BPM, loopable" into a real audio file, up to 30 seconds, and keeps a library of every sound this profile has made.
Get it
Tell the agent "add the sound-effects cell", or run:
nodecode add nc://sound-effects
Set it up
Create an API key at https://elevenlabs.io/app/settings/api-keys with Sound Effects enabled (it is separate from Text to Speech), then either paste it into the cell's api_key in /settings, or export it:
export ELEVENLABS_API_KEY=...
The section, as ~/.nodecode/config.jsonc carries it:
| Setting | Default | What it does |
|---|---|---|
api_key |
(unset) | the key, written inline as a secret |
api_key_env |
ELEVENLABS_API_KEY |
the variable read when api_key is empty |
sound_model |
eleven_text_to_sound_v2 |
eleven_text_to_sound_v2 or eleven_text_to_sound_v3 |
output_format |
mp3_44100_128 |
mp3_44100_128, mp3_44100_192, opus_48000_128, opus_48000_192, wav_44100, wav_48000 (192 kbps MP3 is Creator plans and up, 44.1 kHz WAV Pro and up) |
api_base_url |
https://api.elevenlabs.io |
change only for a data-residency endpoint; https only, or http for a localhost stand-in |
directory |
sounds/ |
where the audio and the library are kept, under the home |
Without a key the tool still answers -- with the sentence that says which setting to fill -- and nothing is billed.
Use it
Ask for a sound: "the sound of a heavy wooden door creaking open in a stone hall", "a short retro coin pickup", "rain on a tin roof, loopable". The model calls the tool with:
| Argument | Values | Notes |
|---|---|---|
prompt |
up to 450 characters | what the sound is: source, action, material, space |
duration_seconds |
0.5 to 30 | pass it whenever a length is named; unset, ElevenLabs picks one at a flat, higher rate |
loop |
true/false | ambience, music beds, engine hums |
prompt_influence |
0 to 1 | higher follows the prompt more literally (0.3 unless sent) |
model |
v2 or v3 | omit for the section's model |
The answer names the saved file, its real playing time (read out of the bytes, not from the request: a 1 s loop can come back 1.5 s long), the format, the size and the characters ElevenLabs billed. Give the path to the operator: the file plays in any audio player and can be copied into a project or a game engine.
saved ~/.nodecode/sounds/sfx_heavy-wooden-door_20261108_101500_3f2a9c1b7d4e.mp3 prompt: heavy wooden door creaking open in a stone hall 0.26 s long, loop, eleven_text_to_sound_v2, mp3_44100_128, 4.1 KB, 47 characters billed id 3f2a9c1b7d4e -- /sfx lists this profile's sounds
/sfx lists the library (newest first, with per-container totals), and /sfx <id or words> shows one sound with its prompt, settings and file. (help :sfx) carries the prompt-writing guidance: one sound per call, concrete descriptions, sequences with "then", music with genre, tempo and key, and when to set loop.
Nothing is overwritten: every file is named with its description, a timestamp and its id, and every generation appends one line to sounds/library.jsonl beside the audio. Post-processing is local and free -- trim, normalise, convert or fix a loop seam with ffmpeg on the machine -- so a sound is generated once and edited after, not regenerated.
Cost and privacy. Every call is a paid ElevenLabs generation on your account (about 10 credits per second when a length is set, a flat amount per generation otherwise; loops can be billed on the request and returned longer). What leaves the machine is the description and the parameters above, sent to api_base_url with your key, and nothing else from the conversation. The library line never leaves the home.
Tests
.github/test-cell.sh tools/nodecode-sound-effects
Every HTTP call is stubbed and the environment is a stub: the suite proves the request shape and its headers, that the bytes that come back are saved and their length read from the frames, header or granule, that /sfx lists and finds a sound, that the library survives a torn line, and that a missing key, a length in the prompt, an out-of-range value and a non-audio 200 all answer without a second billed call.
Credits
A rewrite of NousResearch's Hermes Agent plugin, MIT, at commit e7a218f26db88426ab3e7376f0e000b6e356f5e7: see NOTICE for what was rewritten and what was left out, and LICENSE for the original licence.
