An AI-powered multi-agent system that creates personalized Spotify playlists through natural conversation. Built on the Fetch.ai uagents framework, accessible via ASI:One chat.

A user chats with the orchestrator on ASI:One. The orchestrator — powered by an LLM — decides which specialized agents to involve based on the request. Each agent handles one part of the playlist creation process:

User: "make me a chill playlist for studying"

Orchestrator (LLM-powered)
    |
    |-- decides: need user's taste --> SpotifyAgent
    |-- decides: analyze taste     --> TasteAgent
    |-- decides: parse request     --> ContextAgent
    |-- decides: find songs        --> DiscoveryAgent
    |-- decides: build playlist    --> PlaylistAgent
    |
    v
User gets a playlist + it's created on their Spotify account

The orchestrator doesn't follow a fixed pipeline — it reasons about each request:

• "make me a playlist based on my taste" -> runs all 5 agents
• "give me Frank Ocean and Tyler the Creator songs" -> skips Spotify/Taste, goes straight to Context -> Discovery -> Playlist
• "edm playlist" -> skips Spotify/Taste, searches by genre

• Python 3.12
• A Spotify Developer app (https://developer.spotify.com/dashboard)
• An OpenAI API key
• An Agentverse account (https://agentverse.ai)
• An ASI:One account (https://asi1.ai)

cp .env.example .env

Fill in your .env:

ORCHESTRATOR_SEED_PHRASE=<random string>
SPOTIFY_SEED_PHRASE=<random string>
TASTE_SEED_PHRASE=<random string>
CONTEXT_SEED_PHRASE=<random string>
DISCOVERY_SEED_PHRASE=<random string>
PLAYLIST_SEED_PHRASE=<random string>

SPOTIFY_CLIENT_ID=<from Spotify Developer Dashboard>
SPOTIFY_CLIENT_SECRET=<from Spotify Developer Dashboard>
SPOTIFY_REDIRECT_URI=http://127.0.0.1:9999/callback
OPENAI_API_KEY=<your key>

python3.12 -m venv .venv
source .venv/bin/activate
pip install -r agents/requirements.txt
pip install spotipy openai google-genai

.venv/bin/python get_spotify_token.py

This opens your browser, you log into Spotify, and the token + refresh token auto-save to .env. The refresh token doesn't expire, so future runs auto-refresh without the browser.

Each agent runs in its own terminal from the project root:

make orchestrator
make spotify
make taste
make context
make discovery
make playlist
make auth

1. Open each agent's inspector URL (printed in the terminal when it starts)
2. Click Connect on each one
3. Select Mailbox
4. On the orchestrator's inspector, click Go to Agent Profile
5. Click Chat with Agent

make me a chill playlist for studying

give me 20 hype workout songs

create a playlist of Frank Ocean and Tyler the Creator named "vibez"

edm playlist, 15 songs, named "rave mode"

generate a sad playlist based on my music taste

• Fetch.ai uagents — agent framework, Agentverse registration, ASI:One chat
• Spotify Web API — user profiles, track search, playlist creation
• OpenAI GPT-4o-mini — orchestrator routing, taste analysis, context parsing
• spotipy — Spotify SDK
• Python 3.12

• Agentic orchestration: The orchestrator uses an LLM to decide which agents to call, not a hardcoded pipeline. It can skip agents, and tracks call history to prevent loops.
• Per-user OAuth: Each user connects their own Spotify account via the auth server. Tokens are stored per user and auto-refresh.
• Failsafe design: Every agent falls back gracefully — expired tokens auto-refresh, API failures fall back to mock data, LLM failures fall back to keyword parsing.