nodecode-ttsbuddy

TTS Buddy: speaks text into an MP3 through the hosted TTS Buddy API, lists the voices offline, and downloads a job that was already submitted. A Nodecode rewrite of ngelik/ttsbuddy-cli by ngelik.

Get it

Tell the agent "add the ttsbuddy cell", or run

nodecode add nc://ttsbuddy

Set it up

Make an agent key on your TTS Buddy account, then either paste it into the cell's api_key in /settings, or export it:

export TTSBUDDY_API_KEY=...

Pick a voice with ttsbuddy_voices (or /ttsbuddy voices French female) and keep its id in voice. The original plugin needed the ttsbuddy command-line tool installed; this cell needs nothing but the key, because it posts to the same endpoint the tool posted to.

The section, as ~/.nodecode/config.jsonc carries it:

Setting	Default	What it does
`api_key`	(unset)	the agent key, written inline as a secret
`api_key_env`	`TTSBUDDY_API_KEY`	the variable read when `api_key` is empty
`voice`	`st_m1`	the voice every call speaks in unless it names one
`speed`	(unset)	the pace, 0.5 to 1.5, as a number
`language`	(unset)	the language a Supertonic `st_*` voice speaks; Kokoro voices take none
`directory`	`voice/`	where the MP3s are kept, under the home
`api_base_url`	`https://www.ttsbuddy.com/v1/agent-tts`	the agent endpoint; change it only for a stand-in
`poll_timeout_seconds`	`600`	how long one job is followed, 30 to 1800

Every synthesis is billed to the TTS Buddy account the key belongs to, and uses its minutes. Nothing is sent until a call names a voice and a key.

Use it

The model says a sentence and gets back the path of the MP3 it may hand you:

ttsbuddy_speak(text, voice?, speed?, language?)
ttsbuddy_voices(query?, language?, limit?)
ttsbuddy_recover(job_id, path?)

ttsbuddy_speak submits the text, follows the job while it runs, and downloads the finished MP3 to a new file — it never overwrites one. At most a paragraph or two per call: the file is played, not read.

ttsbuddy_voices is the curated catalogue the CLI carried, so it costs no request and no key: the Kokoro voices, each speaking its own language, and the ten Supertonic voices that take a language.

ttsbuddy_recover takes the exact job id an earlier answer named — after a timeout, say — and downloads that job to a fresh path. It never submits speech, so nothing is billed twice. A job that is still running keeps running.

From the operator's side, /ttsbuddy shows the settings, /ttsbuddy voices <words> searches the catalogue and /ttsbuddy say <text> speaks a line.

A local timeout stops the waiting, not the hosted job: if the answer named a job id, ttsbuddy_recover with that id is the way back.