Moodify

A Spotify bot that rebuilds one playlist for me every two days, based on what I actually listened to. Runs on GitHub Actions, no server, costs nothing.

I used AI for multiple parts of the project, but the ideas, architectural design, and choices that influenced the project's progress were all my own.

How it works

every 2 days:  history -> mood -> candidates -> curator -> playlist
               then measure last cycle and tune the strategy

every hour:    recently_played -> deduplicate -> counter

State lives in a secret Gist, not this repo, since it holds my whole listening history. Each run pulls it at the start and pushes it at the end. That is all gist_state.py does.

The carry-over spiral

Some tracks get carried over so the playlist is not all strangers. My first version asked an AI how many, but the code rejected every increase, so the number could only fall. Few plays, lower carry_over, nothing familiar left, even fewer plays. It hit zero and stayed there for four cycles.

Now it compares the carried tracks against the fresh ones in the same playlist. Carried got played more, +1. Less, -1. Floor of 2 so it cannot bottom out again. No AI call.

Free tier

Groq allows about 8000 tokens a minute and three calls back to back blew past it, so there is a rolling window that waits when a request will not fit. Prompts got smaller too, track lists are just "Title - Artist" now instead of full dictionaries.

When Groq retired llama-3.3-70b-versatile the bot stopped dead because I had the model name hardcoded in three places. It walks a chain now.

Demo

There is a working demo in demo/, served straight from GitHub Pages. It runs on made up data in demo_state.json, is not wired to any Spotify account, and the buttons do nothing. Turn on Pages for the repo and it is at /demo/.

Dashboard

index.html is a static page that reads the Gist and draws it. Cycle history, mood log, the strategy it settled on and why, and its own score against the real one.

There is a text box on it. Whatever I type goes into the next prompt above everything else. "That was too gloomy, make the next one more upbeat" works.

Setup

Fork it, the cycle workflow commits back to the repo.

Spotify app at developer.spotify.com/dashboard. Redirect URI exactly http://127.0.0.1:5000/callback, the IP, not localhost. Tick Web API.

Refresh token. Save as get_token.py and run:

import json
from spotipy.oauth2 import SpotifyOAuth

SCOPE = ("user-read-recently-played user-top-read user-library-read "
         "playlist-modify-public playlist-modify-private playlist-read-private")

auth = SpotifyOAuth(
    client_id="YOUR_CLIENT_ID",
    client_secret="YOUR_CLIENT_SECRET",
    redirect_uri="http://127.0.0.1:5000/callback",
    scope=SCOPE,
    cache_path=".spotify_cache",
)
auth.get_access_token(as_dict=False)
print(json.load(open(".spotify_cache"))["refresh_token"])

A browser opens, you approve, you land on a page that fails to load. Copy the whole URL from the address bar and paste it back into the terminal. Delete the script after.

Groq key from console.groq.com.

Gist. A secret gist with one file bot_state.json containing {}. The id is in the URL. Then a classic token with only the gist scope, fine grained ones cannot touch gists.

Secrets. Settings, Secrets and variables, Actions. Six of them: SPOTIFY_CLIENT_ID, SPOTIFY_CLIENT_SECRET, SPOTIFY_REFRESH_TOKEN, GROQ_API_KEY, GIST_ID, GIST_TOKEN.

Write permission. Settings, Actions, General, Workflow permissions, Read and write. Without it the commit step 403s.

Playlist name. In both workflow files, and they have to match:

PLAYLIST_NAME: "Your Playlist Name"
PLAYLIST_SIZE: "20"

Then Actions, Moodify Cycle, Run workflow.

For the dashboard, point GIST_ID and BOT_REPO in index.html at your own and turn on Pages. Keep in mind a secret gist is not a private one, so publishing a working panel means anyone with the URL can read your listening stats. Leave the placeholder in if you only want the demo public.

Locally

pip install -r requirements.txt
cp env.example .env
python bot.py cycle
python bot.py poll

Settings are all environment variables, see env.example.

When it breaks

Green workflow but no cycle, poll saying "No new plays." — token is missing scopes.

No Groq model was usable — bad or rate limited key, the log says what it tried.

Short playlist — not much history to choose from yet.

"Connection error" on the dashboard — wrong gist id.

Licence

Do whatever you want with it.