CLAUDE.md to use local AI model

Russ

I just followed this guy's tutorial:
https://towardsdatascience.com/run-claude-code-for-free-with-local-and-cloud-models-from-ollama/

I'm using claude code with ollama running gpt-oss:20b on my PC with my nice Nvidia GPU to speed it up.

I found this CLAUDE.md file works well. It tells the gpt-oss model how to use the Write tool to write to files:

My CLAUDE.md file:

## Important rules
- If the user wants you to write a file, do it. Use your Write tool to create the file. The Write tool has a required parameter called "file_path".

My interactive claude session:

 ▐▛███▜▌   Claude Code v2.1.29
▝▜█████▛▘  gpt-oss:20b · API Usage Billing
  ▘▘ ▝▝    ~/appdev/raspi

  /model to try Opus 4.5

❯ create a python script called hello.py in the current directory that prints "Hello World".

● Write(hello.py)
  ⎿  Wrote 2 lines to hello.py
      1 print("Hello World")

This is a command that runs a non-interactive claude session. You could run it via the TRIGGERcmd agent

claude -p -d --model gpt-oss:20b --permission-mode acceptEdits "create a python script called hello.py in the current directory that prints Hello World"

I created a this myclaude.bat file:

set ANTHROPIC_BASE_URL=http://127.0.01:11434
set ANTHROPIC_AUTH_TOKEN=ollama
cd c:\myfolder
claude -p -d --model gpt-oss:20b --permission-mode acceptEdits %1

My Command field in the GUI editor looks like this:

start /w c:\myfolder\myclaude.bat

My commands.json entry looks like this:

 {
  "trigger": "claude code",
  "command": "start /w c:\\myfolder\\myclaude.bat",
  "offCommand": "",
  "ground": "foreground",
  "voice": "",
  "voiceReply": "",
  "allowParams": "true",
  "quoteParams": "true",
  "mcpToolDescription": ""
 }