Open source · MIT · v0.7 is out

Turn sources into
your podcast

Vozonda reads them, voices make waves, you listen. The open-source alternative to NotebookLM, ElevenLabs GenFM and Jellypod: 14 show formats, 1 to 3 voices or a single narrator, watchlists from your feeds, and every episode lands in your own podcast feed.

Get started, it's free

No account · No tracking · Runs on your own hardware

Daylight saving time: keep it or drop it? Debate · Ryan & Serena · from one Wikipedia article · made on one home machine
8:39

Ryan Daylight saving time is fundamentally about making better use of the longer daylight available during summer.

Works with OllamavLLMLM Studiollama.cpp any podcast appNostrLightning
14show templates to choose from, like deep dive or debate
10languages it speaks, and it translates your sources into them
100 %locally powered: script, voices and feed on your own machine
0trackers, accounts or cookies needed

Features

Everything the AI podcast tools do,
on your terms

A tray for all your sources

Links, PDFs, images, YouTube, audio files, RSS and notes. Mark what carries the episode and what is background; long papers are condensed, not cut off.

Check the script first

The hosts stay with your sources, and you can read and fix every line before a single word is voiced.

Your own podcast feed

Private RSS with chapters and transcripts (Podcasting 2.0). Apple Podcasts, Pocket Casts, AntennaPod: it just plays.

Local first, private by design

Script model and voices run on your machine. Uploaded files never go to a cloud model. Works fully offline.

Nostr and Lightning built in

Publish episodes to Nostr, let listeners boost with sats, split value-for-value between hosts and you.

An API for your agents

HTTP API, webhooks and an MCP server: your agents can turn what they find into episodes on a schedule.

How it works

From a page you read to an episode you hear

One Wikipedia article, followed through the whole pipeline. The script lines and the audio below are from the real episode.

  1. 01

    Bring a source

    A link, a PDF, a screenshot or your own notes. Vozonda reads it once and keeps what matters, up to ten sources per episode.

    sourceen.wikipedia.org · CC BY-SA 4.0

    Daylight saving time

    Daylight saving time (DST) … is the practice of advancing clocks to (in theory) make better use of the longer daylight available during summer by having darkness fall at a later clock time. The typical implementation of DST is to adjust clocks ahead of standard time by one hour in spring or late winter, and to set clocks back by one hour in the autumn … The practice of DST was first tested and widely adopted in the early 20th century.

  2. 02

    Read the script first

    Your local model writes the conversation, in one of fourteen show formats. You read every line and fix what is wrong before anything is voiced.

    script · debatereview before voicing

    RyanDaylight saving time is fundamentally about making better use of the longer daylight available during summer.

    SerenaExactly, and for many, that extra hour of daylight after work is a genuine benefit.

    RyanBut let's look at the actual data. A 2017 meta-analysis of forty-four studies found that DST leads to electricity savings of only zero point three percent.

    SerenaNo way.

  3. 03

    Voices record it, with the engine you pick

    One to three voices, or a single narrator. Voice engines are plugins: Kokoro and Piper on your CPU, Qwen3-TTS on a GPU, Voxtral as a hosted option, more on the way. This sample is one engine; switching is a setting, and so is the language model that writes the script.

    qwen3-tts · Ryan & Serenalocal · 8:39

    Ryan Daylight saving time is fundamentally about making better use of the longer daylight available during summer.

  4. 04

    It lands in your own feed

    A private RSS feed with chapters and transcripts. Apple Podcasts, Pocket Casts, AntennaPod: it just plays. Publishing to Nostr is optional.

    your feedrss · podcasting 2.0
    Daylight saving time: keep it or drop it?debate · 8:39 · transcriptnew
    Three NASA missions, one questionbalanced · 8:11 · transcript
    The Tell-Tale Heart, retoldstoryteller · 9:23 · transcript

Run it yourself Hear more samples

Listen

Real episodes, no cherry-picking

Made on one machine at home with local models. Each script was read and, where needed, corrected by hand before it was voiced, exactly as the app lets you. Open "How it was made" under a sample for the models and settings. Sources are openly licensed and linked.

Samples are loading…

How it compares

Honest numbers, not claims

We put Vozonda next to NotebookLM, ElevenLabs GenFM, Jellypod, Wondercraft, Audioread, Open Notebook and Podcastfy, feature by feature, and let listeners judge it in a blind test. Every vote is public.

  • The only one that is open source, runs on your hardware and gives you your own podcast feed
  • Read and fix the script before it is voiced
  • API and MCP for your agents, Nostr and Lightning built in
  • Where others are ahead, we say so: longer, smoother episodes in the cloud tools

Self-host & cloud

Run it yourself, or let us run it

Same software either way. Nothing is held back for the cloud.

Vozonda Cloud

Pay per episode in sats · coming if you want it

  • Nothing to install, update or back up
  • Studio voices and a large script model, no GPU needed
  • Pay with Lightning: no subscription, no account
  • API and MCP endpoint for your agents
  • Your files processed, then deleted

One anonymous click. We build it when enough people ask.

FAQ

Questions

How can I run or install Vozonda?

Run the full self-hosted stack with Docker Compose using prebuilt images from GitHub Container Registry (ghcr.io/vozonda/vozonda), or install the backend orchestrator directly via PyPI (pip install vozonda or uv add vozonda) and npm.

Is it really free?

Yes. The software is MIT-licensed and complete. The cloud, if it comes, is for people who would rather not run a model server; it will not get features the open-source version lacks.

What hardware do I need?

For the default setup, a recent computer with about 16 GB RAM: voices run on the CPU and a small local model writes the script. An NVIDIA GPU gives you the studio voices and bigger models.

Can I trust what the hosts say?

The script is written from your sources only and you can read and fix it before it is voiced. Like any language model it can still get a detail wrong, which is exactly why the review step exists.

Which podcast apps work?

All of them. Vozonda writes a standard RSS feed with Podcasting 2.0 chapters and transcripts. Keep it private with a feed key, or publish it.

How does the blind test work?

You hear the same minute of the same source from both tools, in random order and at the same loudness, and pick one. Results and raw votes are public and update every five minutes.

Give it a star, give it a listen

Vozonda is built in the open. Star the repo, try it at home, tell us what to fix.