About Demodokos

Demodokos Foundry

Local AI audio generation with a production studio around it.

Built by Compunect GmbH for Windows, Foundry brings music, speech, voice creation, stem separation, timeline editing, effects, and export into one application.

Foundry Music v4 V4 Speech Local Generation Windows 10 and 11

Generation is one part of making audio.

A useful result still needs direction, repair, arrangement, mixing, and export. Foundry keeps those steps beside the generation engines so music and voice can move through one local workflow.

Creative work stays local No per-generation credits Production tools included

Music, voice, and the tools between them.

Foundry combines four areas that normally require separate applications or services.

Foundry Music v4

The native engine offers six optimized music models: Small, Medium, Large, X-Large, Large Fast, and X-Large Fast. They cover GPUs with 6 GB to 12+ GB VRAM and support music and lyric workflows in 50+ languages.

V4 Speech model family

Natural text to speech in 10 languages with Large and Medium models, multi-speaker documents, and stable voice identity across long-form work. Explore local AI voice.

Directed voices

More than 40 emotions and speaking styles, each with five intensity levels, work across 60 bundled speaker presets and locally cloned voices. Explore voice cloning.

Production and finishing

Six-stem separation, section repair, a multi-track timeline, DSP effects, project tools, and automation take generated material through to export.

Your machine does the generation.

Foundry uses your Windows GPU instead of sending creative work to a generation service.

Your material stays close

Prompts, scripts, reference voices, and generated audio are processed on your computer.

Iteration is not metered

Foundry does not charge a credit for each audition, correction, or variation.

Online access is specific

Downloads, updates, account access, and license verification use the internet. Generation remains local.

Our privacy policy explains what the website and application transmit.

The current release in brief.

Version 2.1 introduced the native Foundry Music v4 engine, improved model delivery, lower VRAM use, and broader GPU support.

Six music models

Quality and Fast variants for GPUs from 6 GB to 12+ GB VRAM.

About 11 GB peak VRAM

The highest-quality music model previously needed more than 22 GB.

Six-stem output

The improved separation engine provides expanded stem output.

CUDA and Vulkan

NVIDIA with CUDA is recommended. AMD support through Vulkan remains experimental.

Who makes Demodokos Foundry?

Foundry is developed and supported by Compunect GmbH, a software company registered in the European Union.

Compunect GmbH

Compunect GmbH builds local-first audio technology and supports Foundry directly. Product questions can be sent to support@demodokos.com.

Why Demodokos?

In Homer's Odyssey, Demodokos is the blind bard whose songs make distant events present to his listeners. The name connects voice, music, imagination, and the instruments used to express them.

Questions, feedback, or a rough edge?

Product support is handled by the people behind Foundry. You can also find tutorials, demos, release updates, and other users through our public channels.

Echoes articles and release notes · Foundry tutorials