Running Claude Code Locally (No Limits, No Nonsense)

There’s something deeply satisfying about running your own stack locally.

No rate limits.
No “you’ve reached your session cap.”
No waiting in line behind someone else’s prompts.

Just raw, uninterrupted compute—on your terms.

Recently, I set up Claude Code locally, backed by a local model via Ollama. The result? A private, always-available coding assistant that doesn’t tap you on the shoulder after 20 prompts and say “come back later.”

Let’s walk through it.

Why Run Claude Code Locally?

Short answer: control.

Long answer:

🚫 No session limits
⚡ Lower latency (no round trips to the cloud)
🔒 Full privacy — your code stays on your machine
🧪 Perfect for experimentation and sandboxing
💸 Zero usage costs once you’re set up

If you’re doing real dev work (not just dabbling), this is a game changer.

Step 1: Install Ollama

Ollama makes running large language models locally dead simple.

curl -fsSL https://ollama.com/install.sh | sh

That’s it. No Docker gymnastics. No dependency rabbit holes.

Step 2: Pull a Model

You’ll need a model to power Claude Code locally.

Here’s a solid balance between performance and resource usage:

ollama pull qwen3.5:9b

Qwen 3.5 at 9B parameters runs well on most modern machines and is surprisingly capable for coding tasks.

If you’ve got more horsepower, you can go bigger. If you’re on a laptop, this is a sweet spot.

Step 3: Install Claude Code

Now install the CLI:

curl -fsSL https://claude.ai/install.sh | bash

This gives you the Claude Code interface, but instead of relying on cloud inference, we’re going to point it at your local model.

Step 4: Fix Your PATH

Make sure the binary is accessible:

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc && source ~/.bashrc

No PATH, no party.

Step 5: Create a Sandbox

Keep things clean and isolated:

mkdir -p sandbox
cd sandbox

You now have a playground for testing prompts, generating code, or building small tools.

Step 6: Launch Claude Code (Local Mode)

ollama launch claude --config

When prompted, select the model you pulled earlier (e.g., qwen3.5:9b).

And that’s it.

You’re now running a local AI coding assistant with zero external dependencies.

What It Feels Like

This is the part nobody talks about enough.

When you remove limits, your workflow changes:

You stop hesitating before asking a question
You iterate faster
You explore more aggressively
You treat the AI like a real pair programmer

It’s not just about privacy or cost—it’s about momentum.

Tradeoffs (Because Nothing Is Free)

Let’s be real:

Local models ≠ frontier models (yet)
You’ll need decent RAM/CPU (or GPU if you want speed)
Some responses may be less polished than cloud-hosted Claude

But for coding workflows, especially iterative ones?

It punches way above its weight.

Final Thoughts

Running Claude Code locally is one of those upgrades that quietly changes how you work.

It’s not flashy.
It’s not hyped.
But it removes friction—and that’s where real productivity lives.

If you’ve ever hit a session limit mid-flow and thought “this is ridiculous”…

This is your exit ramp.

Coming Next

I’ll be experimenting with:

Running larger models with GPU acceleration
Integrating this into my dev workflow (VS Code + terminal loops)
Using it to assist with infrastructure configs (Nginx, BIND, etc.)

If you end up setting this up, I’d love to hear how it changes your workflow.