How running an OSS AI Assistant on my workstation is super easy!

Talking to my notebook, without being creepy talking to myself - although probably being cringy, actually is pretty easy. I just need podman and the official container image of Mycroft AI. I could install it directly into my Linux, but I like it containerized, especially when playing around.

The command to run Mycroft AI on my Fedora Workstation is this:

podman run --rm -it -v /run/user/1000/pulse/native:/run/pulse/native -v /home/w4tsn/.config/pulse/cookie:/run/pulse/cookie -e PULSE_SERVER=/run/pulse/native --device /dev/snd -v /home/w4tsn/.mycroft:/root/.mycroft -v /home/w4tsn/.config/mycroft:/root/.config/mycroft -p localhost:8181:8181 --name mycroft --privileged docker.io/mycroftai/docker-mycroft

This runs the docker.io/mycroftai/docker-mycroft image in my terminal using podman. I’m mounting the PulseAudio devices and sockets, the /dev/snd device (for control of audio input and output devices) as well as configuration directories. The UI is only accessible on my machine on port 8181. I run it --privileged as I do not know the exact required Kernel Capabilities.

After startup to make use of Mycroft AI I needed an Account and bind this instance to it. A little bummer.

Mycroft AI makes use of TTS (text-to-speech) and STT (speech-to-text) engines. It’s default is Google, but as of today there are self-hosting OSS alternatives Mary TTS and Kaldi respectively. The Mozilla offerings can also be used as hosted alternatives to Google.

Running the whole thing in German (including a custom keyword) is a bit challenging though.

Any thoughts of your own?

Feel free to raise a discussion with me on Mastodon or drop me an email.