isa9000

Turn your organization's documents into short, fact-checked podcasts.

A private audio knowledge base that runs inside your own network, with your own branding.

Documentation · Audio generation UI · CLI and Python · Presentation website and publishing · How quality is enforced

If isa9000 is useful to you, please ⭐ star the repository: it helps other teams find it.

Developed by Isabela and Alex @ girlea.ro, from the engine behind devstories.eu (more than 60 audio courses). Runs on Microsoft Azure (Microsoft Foundry) or on Google Cloud (the Gemini API or Vertex AI).

Quick start

pip install "isa9000[audio-generation-ui]"
isa9000 init                           # writes isa9000.toml: add your Azure or Google credentials
isa9000 audio-generation-ui            # create podcasts in your browser at http://127.0.0.1:8790
isa9000 presentation-website --serve   # build the presentation website, preview it at http://127.0.0.1:8800

Python 3.11 or newer, on Linux or macOS. Servers and CI only need pip install isa9000.

Azure and Google Cloud

isa9000 runs on Microsoft Azure by default, or on Google Cloud. Choose one with provider in isa9000.toml.

To set up either one, see Configuration.

How it works

1. Create. isa9000 reads the policies, release notes, wikis, PDFs or web pages you point it at, plans a short series of episodes, writes each one from those sources only, fact-checks every claim, rewrites episodes that a listener review scores below 9/10 and draws cover art. Podcasts that pass the quality gate are recorded straight away; the rest wait for a person to review them.
2. Publish. isa9000 presentation-website builds a static presentation website from your podcasts. Host it on any internal web server or storage bucket, by hand or with GitHub Actions.
3. Listen. People search, play, read transcripts and share episodes from a phone or a desktop.

Business users create podcasts in the audio generation UI on their own computer and download the results; developers use the command line, Python and CI. Everything listeners hear or see, from the narrator to the logo, is a setting.

Privacy

Only what the models need leaves your machine: your request, the content of your sources (PDFs are sent as files), the scripts to record and short clips of each recording for its check. It goes to the model endpoints you configure: the Gemini API, Vertex AI in your own project, or Microsoft Foundry. The presentation website is static files with no server code or tracking. Report vulnerabilities privately: see SECURITY.md.

License

Apache-2.0: see LICENSE, NOTICE and THIRD_PARTY_NOTICES.md. Contributions are welcome: see CONTRIBUTING.md. Using isa9000 at your organization? Tell us through girlea.ro.