Agentic Podcast Generator

Automatically generate, narrate, and publish a daily data-driven podcast using Claude, ElevenLabs, and Spotify — with zero human intervention.

A fully automated pipeline that fetches live data from any source, generates a spoken briefing script using Claude Sonnet (with real-time web search for context), synthesizes audio with ElevenLabs TTS, and publishes to Spotify every morning via headless browser automation running on an AWS EC2 server.

Built as a general framework. The included examples cover weather and prediction markets, but the pattern works for any domain: financial data, sports, research, product analytics, news — anything that produces structured daily data.

Architecture

Your Data Source      →  Fetch structured daily data (any API)
      ↓
Claude Sonnet         →  Generate spoken script + web search for context
      ↓
ElevenLabs TTS        →  Synthesize audio (your voice or built-in)
      ↓
pydub                 →  Stitch intro jingle + speech
      ↓
Playwright            →  Automated Spotify upload (headless Chromium)
      ↓
Spotify               →  Published episode, every morning at 7am

Quick start (local)

git clone https://github.com/sfrbuilds/agentic-podcast
cd agentic-podcast
pip install -r requirements.txt
brew install ffmpeg        # macOS — needed for audio stitching

cp .env.example .env       # fill in your API keys
python3 generate_episode.py

Out of the box this runs the weather briefing example — no data API key needed (uses Open-Meteo, free and open). To switch to a different data source, change one import line at the top of generate_episode.py.

The script pauses twice for human review — once after generating the script, once after generating the audio — before uploading. Safe to run interactively while you're iterating. Once you're happy with the output, deploy to EC2 and let the cron job take over.

Setup

API keys

Copy .env.example to .env and fill in:

Variable	Where to get it	Required
`ANTHROPIC_API_KEY`	console.anthropic.com	Yes
`ELEVENLABS_API_KEY`	elevenlabs.io	Yes
`ELEVENLABS_VOICE_ID`	See voice setup below	Yes
`SPOTIFY_SHOW_ID`	From your Spotify for Creators URL	For upload
`DATA_API_KEY`	Your data source API key	Depends on source

Spotify show setup

Create a show at creators.spotify.com
Copy your show ID from the URL: creators.spotify.com/pod/show/YOUR_SHOW_ID/home
Set SPOTIFY_SHOW_ID=YOUR_SHOW_ID in .env
Run the one-time session login:
```
playwright install chromium
python3 save_spotify_session.py
```
A browser opens — log in, press Enter. Session persists for weeks.

Voice setup

Option A: Built-in ElevenLabs voice (easiest)

No extra setup. Pick from ElevenLabs' library and paste the voice ID into .env.

Good options for a professional briefing style:

Voice	ID	Character
Brian	`nPczCjzI2devNBz1zQrb`	American, deep, resonant (default)
Daniel	`onwK4e9ZLuTAKqWW03F9`	British, broadcaster style
Adam	`pNInz6obpgDQGcFmaJgB`	American, authoritative

Option B: Clone your own voice

Go to elevenlabs.io/app/voice-lab
Click Add Voice → Instant Voice Clone
Upload 1-3 minutes of clean audio of yourself speaking naturally (no background noise)
Copy the resulting voice ID → paste into .env as ELEVENLABS_VOICE_ID

Instant Voice Clone requires ElevenLabs Creator plan ($22/mo). Recommended voice settings are already tuned in the script for cloned voices.

Adapting to your data source

Each data source lives in data_sources/ as a self-contained Python file. To switch domains, change one import line at the top of generate_episode.py:

# Default — weather briefing (no API key needed)
from data_sources.example_weather import fetch_data, PODCAST_NAME, PROMPT_TEMPLATE

# Swap to prediction markets
# from data_sources.example_markets import fetch_data, PODCAST_NAME, PROMPT_TEMPLATE

# Or write your own in data_sources/my_domain.py
# from data_sources.my_domain import fetch_data, PODCAST_NAME, PROMPT_TEMPLATE

To write your own, create a file in data_sources/ with three exports:

# data_sources/my_domain.py

PODCAST_NAME = "My Daily Briefing"

PROMPT_TEMPLATE = """\
You are producing "{podcast_name}" — a 90-second daily briefing.
Today: {{today}} | Episode: #{{ep_num}}
Data: {{data}}
Return JSON with: title, script, description, items_referenced.
"""  # Note: use {{double braces}} for today/data — single braces are for podcast_name only

def fetch_data() -> dict:
    # Return any dict — it gets passed to Claude as context
    return requests.get("https://your-api.com/data").json()

See data_sources/README.md for more examples and the full interface spec.

Generating an intro jingle

Add a short 2-3 second audio signature before each episode using the ElevenLabs Sound Effects API:

import requests, os

prompts = [
    "two quick keyboard clicks then a single soft chime, minimal and clean",
    "short punchy news intro, three ascending electronic tones",
    "single clean electric piano chord, bright, short decay",
]

for i, prompt in enumerate(prompts, 1):
    r = requests.post(
        "https://api.elevenlabs.io/v1/sound-generation",
        headers={"xi-api-key": os.environ["ELEVENLABS_API_KEY"]},
        json={"text": prompt, "duration_seconds": 2.0, "prompt_influence": 0.5},
    )
    with open(f"jingle_{i}.mp3", "wb") as f:
        f.write(r.content)

# Pick your favourite, then:
# cp jingle_1.mp3 intro_jingle.mp3

Save your chosen jingle as intro_jingle.mp3 in the project directory. It auto-stitches before every episode.

Production deployment on AWS EC2

Running generate_episode.py locally works for testing, but for a truly automated daily podcast you need an always-on server with a cron job. Here's the full path from zero to a daily episode publishing itself every morning.

1. Launch an EC2 instance

Sign in to console.aws.amazon.com → EC2 → Launch instance
Name: podcast-server (or whatever you like)
AMI: Amazon Linux 2023 (free tier eligible, works well)
Instance type: t3.micro (~$8/mo) — plenty for this workload
Key pair: Create a new key pair, download the .pem file, save it somewhere safe (e.g. ~/Downloads/podcast-key.pem)
Security group: Allow inbound SSH (port 22) from your IP only
Storage: Default 8GB is fine
Click Launch instance

Once it's running, note the Public IPv4 address from the EC2 console.

# Fix key permissions (required by SSH)
chmod 400 ~/Downloads/podcast-key.pem

# Connect
ssh -i ~/Downloads/podcast-key.pem ec2-user@YOUR_EC2_IP

2. Install dependencies

# Update system packages
sudo dnf update -y

# Python packages
pip3 install anthropic requests python-dotenv pydub

# Playwright + Chromium (for Spotify upload)
pip3 install playwright
playwright install chromium
playwright install-deps chromium

# Verify Chromium works headlessly (important — not all servers support it)
python3 -c "
from playwright.sync_api import sync_playwright
with sync_playwright() as p:
    b = p.chromium.launch(headless=True)
    print('Chromium OK:', b.version)
    b.close()
"

# ffmpeg (for audio stitching)
sudo dnf install ffmpeg -y

Note: If the Chromium test fails with a missing library error, run playwright install-deps chromium again and check the error message. On Amazon Linux you may also need sudo dnf install atk cups-libs gtk3 -y.

3. Upload the project

# From your Mac — upload the whole project to the server
scp -i ~/Downloads/podcast-key.pem -r \
  /path/to/agentic-podcast \
  ec2-user@YOUR_EC2_IP:~/agentic-podcast

4. Configure credentials on the server

# SSH into server
ssh -i ~/Downloads/podcast-key.pem ec2-user@YOUR_EC2_IP

cd ~/agentic-podcast
cp .env.example .env
nano .env   # fill in ANTHROPIC_API_KEY, ELEVENLABS_API_KEY, etc.

5. Set up the Spotify session (one-time, from your Mac)

The Spotify upload uses a saved browser session. You generate it once on your Mac (where you can see the browser), then copy it to the server. You will need to repeat this every few weeks when the session expires.

# On your Mac — run the login helper
cd /path/to/agentic-podcast
python3 save_spotify_session.py
# A browser opens. Log into Spotify for Creators. Press Enter when done.

# Copy the session file to the server
scp -i ~/Downloads/podcast-key.pem \
  spotify_session.json \
  ec2-user@YOUR_EC2_IP:~/agentic-podcast/spotify_session.json

6. Test a full run on the server

# SSH in and do a dry run with --auto to skip the interactive prompts
ssh -i ~/Downloads/podcast-key.pem ec2-user@YOUR_EC2_IP

cd ~/agentic-podcast
mkdir -p logs
python3 generate_episode.py --auto

Check that an episode appears in your Spotify for Creators dashboard before setting up the cron job.

7. Set up the cron job

# On the server — open crontab
crontab -e

# Add these two lines (runs at 7am Eastern, every day):
CRON_TZ=America/New_York
0 7 * * * cd ~/agentic-podcast && python3 generate_episode.py --auto >> ~/agentic-podcast/logs/podcast.log 2>&1

Verify it was saved:

crontab -l

8. Monitor

# Check today's log
tail -50 ~/agentic-podcast/logs/podcast.log

# Check the episode counter (increments each day)
cat ~/agentic-podcast/episode_count.txt

# Watch logs live (useful after first scheduled run)
tail -f ~/agentic-podcast/logs/podcast.log

9. Get email alerts on failure (optional but recommended)

Add this to your crontab so you get an email if the episode fails:

MAILTO=you@youremail.com
CRON_TZ=America/New_York
0 7 * * * cd ~/agentic-podcast && python3 generate_episode.py --auto >> ~/agentic-podcast/logs/podcast.log 2>&1 || echo "Episode failed — check ~/agentic-podcast/logs/podcast.log" | mail -s "Podcast failed" you@youremail.com

Or add a simple check at the end of the cron command that pings you on Slack/Discord via webhook if the exit code is non-zero.

10. Updating the code

When you make changes locally and want to deploy them to the server:

# From your Mac — sync only changed files
scp -i ~/Downloads/podcast-key.pem \
  generate_episode.py data_sources/my_domain.py \
  ec2-user@YOUR_EC2_IP:~/agentic-podcast/

# Or sync the whole project (excludes .env and session files)
rsync -av --exclude='.env' --exclude='spotify_session.json' \
  --exclude='*.mp3' --exclude='logs/' \
  -e "ssh -i ~/Downloads/podcast-key.pem" \
  /path/to/agentic-podcast/ \
  ec2-user@YOUR_EC2_IP:~/agentic-podcast/

Session refresh

When the Spotify session expires (every few weeks), the log will show:

Session fully expired. Run: python3 save_spotify_session.py

Fix it in 2 minutes from your Mac:

python3 save_spotify_session.py   # log in when browser opens

scp -i ~/Downloads/podcast-key.pem \
  spotify_session.json \
  ec2-user@YOUR_EC2_IP:~/agentic-podcast/spotify_session.json

Cost summary

Item	Cost
EC2 t3.micro	~$8/mo
Claude Sonnet (~0.08/episode × 30)	~$2.50/mo
ElevenLabs Creator plan	$22/mo
Web search (~3 searches/episode)	~$0.90/mo
Total	~$33/mo

That's a fully automated daily podcast for about $1/day.

Project structure

agentic-podcast/
├── generate_episode.py          # Main pipeline — domain-agnostic core
├── spotify_upload.py            # Playwright Spotify upload module
├── save_spotify_session.py      # One-time Spotify browser login (run on Mac)
├── data_sources/
│   ├── example_weather.py       # Default example — weather briefing (no API key)
│   ├── example_markets.py       # Example — prediction markets via Veynor API
│   └── README.md                # How to write your own data source
├── intro_jingle.mp3             # Intro sound (not tracked in git)
├── spotify_session.json         # Browser session cache (not tracked)
├── episode_count.txt            # Persistent episode counter (not tracked)
├── recent_topics.json           # Repeat-prevention memory (not tracked)
├── logs/                        # Cron job output (not tracked)
├── .env                         # Your credentials (not tracked)
├── .env.example                 # Credential template
├── requirements.txt
└── README.md

How the script generation works

Claude Sonnet receives your structured data and a prompt describing the format. Before writing the script, it runs 2-3 web searches to understand why the data looks the way it does — adding real-world context that makes the episode sound like a human who read the news, not a ticker readout.

The web search integration uses Anthropic's built-in web_search tool ($10/1000 searches). You can disable it to reduce latency and cost:

# In .env
USE_WEB_SEARCH=false

Cost estimate (API only, per episode)

Service	Usage	Cost
Claude Sonnet	~2,000 input + ~600 output tokens	~$0.03
Web search	2-3 searches	~$0.03
ElevenLabs	~250 words (~1,500 chars)	~$0.03
Total		~$0.09/episode

Monthly at daily cadence: ~$2.50/month in API costs (not counting ElevenLabs plan or EC2).

License

MIT. Fork it, build your own automated voice, ship something.

Built by @sfrbuilds

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Agentic Podcast Generator

Architecture

Quick start (local)

Setup

API keys

Spotify show setup

Voice setup

Option A: Built-in ElevenLabs voice (easiest)

Option B: Clone your own voice

Adapting to your data source

Generating an intro jingle

Production deployment on AWS EC2

1. Launch an EC2 instance

2. Install dependencies

3. Upload the project

4. Configure credentials on the server

5. Set up the Spotify session (one-time, from your Mac)

6. Test a full run on the server

7. Set up the cron job

8. Monitor

9. Get email alerts on failure (optional but recommended)

10. Updating the code

Session refresh

Cost summary

Project structure

How the script generation works

Cost estimate (API only, per episode)

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
data_sources		data_sources
.env.example		.env.example
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
generate_episode.py		generate_episode.py
requirements.txt		requirements.txt
save_spotify_session.py		save_spotify_session.py
spotify_upload.py		spotify_upload.py

Folders and files

Latest commit

History

Repository files navigation

Agentic Podcast Generator

Architecture

Quick start (local)

Setup

API keys

Spotify show setup

Voice setup

Option A: Built-in ElevenLabs voice (easiest)

Option B: Clone your own voice

Adapting to your data source

Generating an intro jingle

Production deployment on AWS EC2

1. Launch an EC2 instance

2. Install dependencies

3. Upload the project

4. Configure credentials on the server

5. Set up the Spotify session (one-time, from your Mac)

6. Test a full run on the server

7. Set up the cron job

8. Monitor

9. Get email alerts on failure (optional but recommended)

10. Updating the code

Session refresh

Cost summary

Project structure

How the script generation works

Cost estimate (API only, per episode)

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages