VoxBee v0.9 adds a local API server — a small HTTP server, running on your Mac, that exposes VoxBee's transcription and text-to-speech over the network. It speaks the same routes the OpenAI API does, so any tool or script built for the OpenAI SDK works by changing one thing: the base URL.
Point Your OpenAI SDK at Localhost
The server listens on http://127.0.0.1:5111 by default. To use VoxBee instead of OpenAI's cloud, set the base URL and you're done:
from openai import OpenAI
client = OpenAI(
base_url="http://127.0.0.1:5111/v1",
api_key="YOUR_VOXBEE_TOKEN",
)
# Transcribe — runs on your local Whisper/Parakeet model
with open("meeting.m4a", "rb") as f:
text = client.audio.transcriptions.create(model="whisper-1", file=f)
# Speak — synthesizes on-device with the System voice (no API key needed)
speech = client.audio.speech.create(model="tts-1", voice="system", input="Build complete.")Existing OpenAI-SDK code doesn't need rewriting — it needs re-pointing.
The Routes
The server exposes OpenAI-compatible endpoints plus a few VoxBee-native ones for discovery:
POST /v1/audio/transcriptions— speech-to-text, OpenAI-compatible.POST /v1/audio/speech— text-to-speech, OpenAI-compatible.GET /v1/models— list the STT models you have installed.GET /v1/voices— list the TTS voices available to you.GET /v1/infoandGET /health— app info and a liveness check.
The native /v1/models and /v1/voices routes are the part the OpenAI cloud can't give you — they reflect your installed models and voices, not a fixed catalog.
Everything Runs On-Device
This is the whole point. Transcription runs on your local Whisper or Parakeet model. The System voice synthesizes speech locally with no API key. There's no cloud round-trip and no per-minute billing — the audio never leaves your Mac unless you've deliberately configured a cloud provider with your own key. For privacy-sensitive work, that means you get a drop-in OpenAI replacement that physically can't ship your audio to a third party.
Try Every Endpoint Without Leaving the App
The new API tab toggles the server on and off, manages your bearer token, and embeds an interactive API explorer (powered by Scalar) right inside VoxBee. You can fire a real transcription or speech request at every endpoint without opening a terminal. Prefer your browser? Hit "Open in browser" to get the full-window explorer at http://127.0.0.1:5111/docs.
Loopback by Default, Token Always
Security is built in, not bolted on. The server binds to loopback (127.0.0.1) so only your own machine can reach it, and every request needs a bearer token. There's an opt-in to bind to your LAN if you want to call VoxBee from another device, but the safe default is local-only. The port is configurable from the API tab if 5111 clashes with something.
Where This Goes Next
The same server also speaks the Model Context Protocol at POST /mcp, which turns these routes into native tools your coding agent can call. If you want Claude Code, Codex, or Cursor to "summarize this PDF and read it to me," that's the companion post.
Download VoxBee and try it free for 14 days. The local API server, the on-device transcription, and the System voice all work with no cloud account and no per-minute fees.