Local AI as a concrete sovereignty decision

The debate about local AI tends to revolve around the big questions: chips, data centers, geopolitics. For a company, though, its value often shows up on a far smaller scale, in everyday tasks that can be handled locally without losing quality.

One example I recently set up myself is dictating text on the computer. With AI, dictation is becoming almost a must. For years it was a losing proposition: whatever you said, you spent longer correcting afterwards than typing would have taken, especially with product names, technical terms and proper nouns. Punctuation had to be spoken as a command.

That has shifted. On my Mac, a local Whisper model handles speech recognition, a small local language model from the Qwen family cleans up filler words, punctuation and paragraphs, and the dictation tool VoiceInk runs the whole process and inserts the text at the cursor. For German dictation with specialist vocabulary, the result came very close to a good cloud service, without the recording ever leaving the device.

In a business context, local AI already becomes a sorting question: which work should the machine handle itself, which belongs on a controlled enterprise platform, which genuinely needs the cloud? For more complex agentic workflows, local AI often isn't enough yet. But for bounded tasks like this dictation example, it reaches the level of good cloud services, and it cuts running costs substantially.

More on the concrete setup: https://sixtyfour.solutions/praxis/lokales-diktat-ki-setup/