← Back to the Vault
AI Tools · Automation

Build your own local AI command center

You don't need a paid AI OS. A voice-driven command center is four open parts, and it runs on your own machine. Here is the whole thing, and how to start.

No vendor lock-in, no monthly fee. You point it at a local model and run it offline. Swap any part for something you prefer. These are the four pieces.

Brain
Local model
Any model you point it at
Memory
Obsidian
Results saved as markdown
Voice
Open-source
Local speech in and out
Engine
Claude Code
Routes the work to skills

How it works

  1. 1You speak. Local speech-to-text turns it into text, on your machine.
  2. 2Claude Code takes the request and pulls only the skill the moment needs, a small SKILL.md, not a giant prompt.
  3. 3The result lands as a note in an Obsidian vault. A linked graph answers fast, with no database to run.
  4. 4It speaks back. Private, offline-capable, and close to free to run.

The voice, fully local

SYSTRAN/faster-whisper
Speech-to-text. Fast, local, built on CTranslate2.
hexgrad/kokoro
Text-to-speech. An open-weight 82M-parameter voice model.

Start here

  1. 1Get a local model running and talk to it through Claude Code.
  2. 2Point Claude Code at an Obsidian folder so every answer is saved as a note.
  3. 3Add faster-whisper for voice in, then kokoro for voice out.
  4. 4Wrap each job as a small SKILL.md so the model loads only what it needs.

That is the whole skeleton. From there you reskin it, swap the model, or fork a copy per project.

Want the next drop first?

New builds and skills land in the Vault every week.

Follow for the weekly drops