I tried using Soniox on a @svat's recommendation, & it is pretty awesome.
Typing at the speed of speech, and it is quite forgiving. I thought certain "foreign" words would require enunciation. It is quite forgiving though.
(reading aloud Prof SKR's book on Bodhidharma)
Replying to @svat
Pretty impressive.. it even got the original em-dash in the text correctly (only a small pat for myself for conveying that in speech :p)
Thank you! this is great.
Serendipity!
Putting a "manager/ponytail/supervisor" node that monitors requests made by agents in sandboxes is how I was thinking of forcing/ensuring "correct behavior" .. in shower today.
Ever since i put in egress proxy into Kaas, I've been intrigued by this approach.
One of the biggest challenges I’ve grappled with lately is how to create determinism where it otherwise doesn’t exist. How do I get an agent to behave like I would, without me telling it to?
cra.mr/a-little-determinism
Replying to @svat
Pretty impressive.. it even got the original em-dash in the text correctly (only a small pat for myself for conveying that in speech :p)
Thank you! this is great.
Elevenlabs' Kannada Speech to Text is pretty good.
I'm considering using this to transcribe text when OCR fails.
My recent project - an easy to browse, and read library of late Prof S K Ramachandra Rao's published Kannada and English works:
files.btbytes.com/profskr/in…
And wrote up the process on my blog: maxim.usindic.us/2026/10/30/…
Quite easy to build your own like this today
cormacmccarthy.lib.utk.edu
Needs a camera and a few tokens. No need to scan the barcodes either.
Maybe I will.





















