at recurse-center, i’ve seen many people use a.i. efficiently. it doesn’t harm their learning; they’re careful about how they use it; and they’re perfect examples for learning how to leverage it.


i don’t want to pay for a claude code subscription. i know it’s powerful; but i don’t like the idea of control over my computer being available over the internet.

i’m going to do a little bit of digging to see if there are things i can run locally, and what a good workflow there would look like.


the most ideal way i found was to find a model (or models) on ollama, and use it using the cli. the way to customize outputs is to make a modelfile (https://docs.ollama.com/modelfile).

in SYSTEM, you define a ‘role’ and give it more specifications.


i set up three kinds of model-files;

code assistant:

thinker:

dialectic thinker:

make a new mode with:

ollama create {name} -f {modelfile location}

then run with:

ollama run {name}


ollama serves a local server: https://blog.postman.com/how-to-connect-to-local-ollama/


the next thing george mandis suggested was to use an agent. an agent can control your computer, and make changes on your behalf. i was skeptical; downloaded opencode; and set all permissions to explicitly ask.

the good thing is that everything only runs if i serve ollama with ollama serve. i can always stop the local server if i’m hesitant.