Devaten
AI-Powered Database Optimization Solutions
Start new thread
Mikko Larkela

Self-Hosted Voice AI Workflow - From voice to action. Powered by on-prem AI.

by•
Meet Devaten— a fully on-premise AI voice automation tool, deployed via Docker Compose. It converts voice to text, runs the text through AI classification, and then notifies users or updates your source database — all without sending data to the cloud.

Add a comment

Replies

Best
Mikko Larkela
Maker
📌
👋 Hey Product Hunt! We’re excited to launch Devaten, a fully on-premise AI pipeline for voice-driven automation. Here’s what it does: 🎙️ Converts voice to text 🧠 Runs the text through AI classification 🔔 Sends notifications or updates your source database 📦 Fully self-hosted via Docker Compose — no cloud, no external APIs We built this for teams who need automation but can’t (or won’t) send sensitive data to the cloud. Think regulated industries, secure environments, or local-first workflows. We’d love your feedback — questions, use cases, or even critiques are all welcome! 🙌 Let us know how you might use it — or if there’s a feature you’d love to see next. — The Devaten Team 🚀
Domhnall O' Hanlon

This is a great idea @mikko_larkela will definitely have to try it out

Do you connect to your own models or does the container ship with the models? (Does this make the containers quite large?)

Mikko Larkela
Maker

@domhnall_o_hanlon  Glad you liked the idea 😊

the container itself doesn’t ship with all AI models preloaded, though OpenAI Whisper is already included in the base image.

The only additional model we’ve tested with so far is LLaMA 3 (8B). So after the Docker containers are up and running, you just need to run:

ollama pull llama3:8b

This approach helps keep the Docker images lighter (as shown in the docker system df output—total image size is around 12.5 GB), and gives flexibility to pull only the models you need.

Here is guide to try this:

https://github.com/devatengit/onpremises-common-agent/tree/main

Domhnall O' Hanlon
Mikko Larkela
Maker

@domhnall_o_hanlon 

A few things to note:

After the Docker Compose setup is up and running, and Ollama has been pulled,

You can start using the APIs. also.

You can send audio files and receive transcriptions and classifications via AI.

http://host:8081/devaten/audio/transcribe/?file

The current prompts are defined in the docker-compose file. There is a demo set up for "Patients" classification, which you can customize to fit your own use case.

ollama.request.prompt:
ollama.classification.prompt: