SEBAA Mohammed

Posted on Oct 21

Running Claude Code with Ollama and Local Models

#ai #java #ollama #claudecode

Claude Code + Ollama: Coding with a Local Model

Claude Code is Anthropic's agentic coding tool. It reads your project, edits files, and runs commands from the terminal. With Ollama, you can run it on a model hosted on your own machine instead of in the cloud.

Why local models?

Your code never leaves your machine
No API costs
Works offline

In this guide, you'll install Claude Code, run it inside IntelliJ IDEA, connect it to Ollama, and test it on a small Java project.

Step 1: Install Claude Code

Install it globally with npm:

npm install -g @anthropic-ai/claude-code

Then open a terminal and run:

claude

If the Claude Code prompt appears, the installation worked.

Step 2: Run It Inside IntelliJ IDEA

You don't need to leave your IDE. Open your project in IntelliJ, open the built-in terminal (View → Tool Windows → Terminal), and run:

claude

Claude Code now has access to your project files and works right next to your code.

Step 3: Connect It to Ollama

Ollama exposes an Anthropic-compatible API, so Claude Code can talk to a local model.

1. Install Ollama
Ollama lets you download and run language models on your own machine. Your code never leaves your computer and there is no per-request cost.

On https://ollama.com you can create a free account. The dashboard shows your free quota for cloud models. For the rest of this tutorial (100% local), it isn't required.

2. Pull a coding model:

ollama pull qwen3.5:9b

3. Point Claude Code at Ollama:

export ANTHROPIC_AUTH_TOKEN=ollama
export ANTHROPIC_BASE_URL=http://localhost:11434
claude --model qwen3.5:9b

On Windows PowerShell, use $env:ANTHROPIC_AUTH_TOKEN="ollama" and $env:ANTHROPIC_BASE_URL="http://localhost:11434" instead of export.

Alternatively, let Ollama handle the setup with one command, using --model

ollama launch claude --model qwen3.5:9b

Step 4: Build a Demo Project

Let's test the setup with a small Java project: a console task manager.

Create an empty Java project in IntelliJ, open the terminal, start Claude Code, and give it this prompt:

Create a simple Java console app: a task manager. Users can add,
list, complete and delete tasks. Save tasks to a text file so they
persist. Keep it to 2-3 classes.

Claude Code will plan the work, create the files, fix the errors, and ask for your permission before making changes. Approve the actions and watch it work.

Then run the app and try it out:

Notice that it doesn't only generate code. It also creates the files, runs the program, and fixes its own errors.

Cloud vs Local: Quick Comparison

	Cloud Claude	Local model (Ollama)
Privacy	Code is sent to the API	Stays on your machine
Cost	Subscription / API	Free
Speed	Fast	Depends on your hardware
Quality	Highest	Good for small tasks

Conclusion

Claude Code with Ollama is a practical way to try agentic coding without sending your code anywhere, as long as your machine can handle the model. Start with a small project like this one, then try it on your own code.

Thanks for reading!

Top comments (15)

AI Dive • Oct 21

2.1.300 reportedly cut 19,045 prompt tokens, handy for 9b models.

SEBAA Mohammed • Oct 22

Thanks for the tip! I can't confirm the exact number, but lower prompt overhead definitely helps small models since the system prompt eats into the context window.

Paul Irolla • Oct 21

Local is the right place for the cheap steps. A lot of a session is reads, so the split matters. Does the review step still go to the cloud model?

SEBAA Mohammed • Oct 22

Great question! In this demo everything ran locally, so I haven't tested a split yet. Local for reading and boilerplate, and a cloud model for the final review, sounds like the sweet spot. I am planning on trying it and cover it in a follow-up.

Paul Irolla • Oct 23

@mohammed_sebaa_2610 Local for the reads, cloud for the final review. A missed error costs the most at the review, so that is where the better model earns its price.