Adriel's Lab > AI

[ glyph ]

AI

Local models, real work, nothing leaves the house.

The rule of this lab: the interesting question isn’t what a frontier model can do, it’s what a machine in your basement can do reliably, every week, on your own data. Everything below runs on local hardware.

You have thousands of hoarded files that you’ll never read. However, a local AI is your saving grace to read all that back to you. And that’s what matters the most; your machine has a soul that will read all of it, and tell you which of that tiny fraction matters.

If you have the racks, and you reject the corporate dominance of the AI space, then you have all the power to run a fully functioning system. The cloud models become an afterthought, or the super smart auditor on occasional standby.

Fundamentals:

  1. A corpus / private library is your starting point to giving it full context about who you are, or what it should know. Your data slowly grows and matures with you. Nothing you build later replaces it.
  2. Creating an AI pipeline (n8n)
  3. Home network monitoring: If you spend lots on computers that run nonstop at home, maintenance will eat up a good chunk of your day. Hardware health, file integrity, and whether each computer is doing its job right.
  4. A way to measure. Ask the cloud to suggest any LLM that fits best with your setup. Let it run tests to see how well it performs, and see for yourself. LLMs get pumped out on the internet frequently.
  5. Fully automated tasks: Anything you want it to do while you mind your own business, is the perfect reason to use it. Can’t keep up with your youtube feed? Need to compile a month’s worth of emails into a quick read? Got finances to sort and track? These are a few good reasons out of the hundreds. Just think, what’s noise to you, and think about how a machine will handle that.

Intermediate:

  1. Searching plain text over your library
  2. More parameters ≠ better performance. A balance between quantization and parameters with extra GPU space to cram more context is the best way to enjoy a local AI conversation.

Advanced:

  1. Fully autonomous home network of servers. You built your automations, and now it’s time to put them to the test.
  2. Agents with access to tools, will feel like the cloud experience you wanted to build.
  3. Orchestrator - The one brain that controls everything while you sleep.

Payoffs:

  1. Building the AI infrastructure won’t make you a faster builder overnight, it will make you more independent.
  2. A searchable library is more than enough for the average user.
  3. Cloud AI is cheap monthly against hardware that you own.