简体中文

Aurio

Your personal AI radio.

Context-aware · library-native · locally hosted.

Quick Start · API Relay · Screenshots · Architecture · Docs

Recommended API relay — OpenAI-compatible endpoint maintained by the author. Plug into Settings → Brain · AI → API Key, or use .env below.

_{Electron desktop · browser PWA · 420×760 player · dark / light themes}

_{Works with   Claude · Codex · OpenAI-compatible APIs  ·  Navidrome · NetEase · QQ Music  ·  macOS Calendar · ICS · OpenWeather  ·  UPnP / DLNA}

---

At a glance

Not a playlist app. Not a chatbot.
A local AI DJ that pulls real tracks from your libraries, reads your day, and speaks between songs.

Library-native Navidrome · NetEase · QQ

Context-aware Calendar · weather · taste corpus

Locally hosted Loopback API · disk-cached TTS

---

See it in action

_{Standby → On-air → Chat → Settings → Brain}

---

Why Aurio

Streaming apps optimize for engagement. Playlists demand curation. Aurio is a third path — a radio host that runs on your machine, knows your day, and picks from your libraries.

	Algorithmic streaming	Aurio
Music source	Platform catalog	Your NAS + NetEase + QQ
Personality	None	Editable taste corpus + DJ persona
Context	Opaque	Calendar · weather · time-of-day
Voice	None	System / Tencent / Fish TTS
Privacy	Cloud-first	Local server · loopback API by default

Design principles

Principle	What it means
Local-first	Brain, queue, TTS cache, and settings live on your machine — cloud is optional
Context-native	Every segment assembles persona, taste, weather, calendar, and play history before the AI speaks
Library-grounded	Tracks resolve against real libraries, not hallucinated titles

---

A day on the air

Aurio doesn't wait for you to press play. Scheduled beats keep the show alive:

Time	Beat	What happens
07:00	plan	Day plan segment — sets the arc for what's ahead
09:00	morning	Morning open — weather, calendar, first picks
10:00–23:00	mood	Hourly mood check — append new segments to the queue
Anytime	open	Radio engine refills when the queue runs low
On demand	chat	"Play something jazzy" — enqueue, steer mood, or talk-only

---

What you get

Intelligence

	Capability	Detail
🧠	AI brain	Claude / Codex CLI, or any OpenAI-compatible API
📅	Context engine	Weather, macOS Calendar, ICS feeds in every segment
💬	Chat to steer	"Play some Jay Chou" — enqueue, mood shift, or talk-only
⏰	Scheduled show	07:00 plan · 09:00 morning · hourly mood 10–23

Music & broadcast

	Capability	Detail
🎵	Multi-source music	Navidrome · NetEase (QR) · QQ — search, queue, lyrics
🎙️	Voice	macOS say · Windows SAPI · Tencent · Fish — disk-cached
📻	Radio engine	Auto-refills queue via WebSocket when tracks run low
🔊	UPnP cast	DLNA speakers on your LAN

Platform

	Capability	Detail
🛡️	Secure default	Control API loopback-only; media proxies LAN-ready for casting
🖥️	Cross-platform	Electron desktop + browser PWA from one server

---

API relay

Don't want to wrangle CLI logins? Use the author-maintained relay:

token.baogutang.top

AI_PROVIDER=api
AI_API_KIND=openai
AI_API_BASE_URL=https://token.baogutang.top/v1   # use the exact base URL on the portal
AI_API_MODEL=your-model-id
AI_API_KEY=your-key-from-portal

Or configure in-app: Settings → Brain · AI → API Key.
CLI mode (Claude / Codex) still works with zero API key — the relay is optional.

---

Screenshots

  

     

UI notes

• Dot-matrix clock standby with live service strip (NetEase · Navidrome · QQ)
• Spectrum + synced lyrics; drag-and-drop Up Next queue
• Glass-morphism sheets with Framer Motion spring physics
• Monospace Nerd Font UI · accent #ff6a3d / #5ad19a · dark / light themes

---

Quick start

Requires Node.js 20+ · macOS, Windows, or Linux

Prefer packaged builds? Grab the latest macOS, Windows, or Linux installer from Releases.

Step	Command
1	git clone https://github.com/baogutang/aurio.git && cd aurio && npm install
2	cp .env.example .env — every key is optional
3	npm run server → open http://localhost:8080
4	npm start — Electron desktop (optional)

First launch opens an onboarding wizard (AI → music → voice). Reconfigure anytime in Settings.

npm run dist:mac        # macOS .dmg + .zip
npm run dist:win        # Windows NSIS + portable
cd web && npm run dev   # frontend HMR

---

Architecture

Trigger → context.js → brain/ → music/ → tts/ → WebSocket → React player

Segment pipeline (sequence)

sequenceDiagram
    participant T as Trigger
    participant C as Context
    participant B as Brain
    participant M as Music
    participant V as TTS
    participant P as Player

    T->>C: assemble(trigger)
    C->>B: think(prompt)
    B-->>C: {say, play[], intent, ...}
    C->>M: resolveQueue(play[])
    M-->>C: tracks with URLs
    C->>V: cachedSynthesis(say)
    V-->>P: WebSocket broadcast

Loading

Each beat returns { say, play[], reason, segue, intent, placement, mood }.
→ docs/architecture.md

Tech stack

Layer	Stack
Desktop	Electron 33
Frontend	React 18 · Vite · Tailwind · Framer Motion
Server	Node.js 20 · Express · WebSocket
Brain	Claude / Codex CLI · OpenAI-compatible API
Music	Navidrome (Subsonic) · NetEase API · QQ Music
Voice	macOS say · Tencent Cloud · Fish Audio
Cast	UPnP / DLNA via native SSDP discovery

---

Configuration

Copy .env.example → .env. In-app changes persist to data/settings.json.

Variable	Purpose
PORT	Server port (default 8080)
AURIO_ALLOW_LAN	Open control API to LAN (default false)
AI_PROVIDER	claude · codex · cli · api
AI_API_*	Hosted model / relay
NAVIDROME_*	NAS music library
NETEASE_COOKIE	Auto-filled after QR login
VOICE_PROVIDER	system · tencent · fish

---

Usage

Goal	Action
Hands-free listening	Let cron beats run (morning open, hourly mood)
Start the show	Tap Play — radio engine refills the queue
Request a vibe	Chat: "Play some Jay Chou" · "Change the mood"
Switch source	Tap Source — combined / NetEase / Navidrome / QQ
Cast	Settings → Cast → DLNA device
Tune taste	Edit user/taste.md, routines.md, mood-rules.md

curl -X POST http://localhost:8080/api/chat \
  -H 'Content-Type: application/json' \
  -d '{"text": "something mellow"}'

→ examples/api.md

---

Documentation

Topic	Link
Architecture	docs/architecture.md
Frontend spec	docs/FRONTEND_SPEC.md
Security model	SECURITY.md
Changelog	CHANGELOG.md
Contributing	CONTRIBUTING.md
Social preview	PNG · SVG source

Regenerate README media (server must be running):

node scripts/capture-readme-assets.mjs

---

Development

npm run server       # backend only
npm test             # vitest (19 tests)
cd web && npm run build

---

FAQ

Do I need an API key?

No. Default brain uses your local Claude or Codex CLI. API mode — including the author relay — is optional.

Brain shows unavailable

Verify claude --version or codex --version in terminal. For API mode, check Base URL + key under Settings → Brain · AI.

Without Navidrome?

Yes. NetEase and QQ search work out of the box. NetEase playback needs QR login in Settings.

Browser only?

Yes — npm run server serves the PWA at http://localhost:8080.

---

License

MIT © 2026 Aurio contributors

---

Built by baogutang

AI relay → token.baogutang.top