Forem: Sushan

Running Claude Code with Ollama models (Local / Cloud)

Sushan — Thu, 12 Mar 2026 06:05:45 +0000

Run Claude Code with Ollama (Local, Cloud, or Any Model)

This guide shows how to run Claude Code using Ollama, allowing you to use local models, cloud models, or any Ollama-supported model directly from your terminal.

Prerequisites

Make sure the following tools are installed:

Ollama
Claude Code

Install Ollama

If Ollama is not installed, you can install it using the commands below.

You can also follow this guide:

https://dev.to/sushan/how-to-connect-a-local-ai-model-to-vs-code-1g8d

Windows (PowerShell)

irm https://ollama.com/install.ps1 | iex

macOS / Linux

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

Verify installation:

ollama --version

Install Claude Code

Windows (PowerShell)

irm https://claude.ai/install.ps1 | iex

macOS / Linux

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

Verify installation:

claude --version

Running Claude Code with Ollama

Once both tools are installed, you can start Claude Code through Ollama.

The commands work the same on Windows, macOS, and Linux.

Option 1: Launch and Select a Model

Run the command:

ollama launch claude

This will open a model selection menu where you can choose a model using the arrow keys.

Option 2: Launch with a Specific Model

You can also specify the model directly.

Example:

ollama launch claude --model kimi-k2.5:cloud

Other examples:

ollama launch claude --model llama3
ollama launch claude --model qwen2.5
ollama launch claude --model kimi-k2.5:cloud

Replace the model name with any model available in your Ollama environment.

Grant Folder Access

When Claude Code starts, it will ask permission to access the current folder.

Select Yes to allow Claude to read and modify files in the directory.

Done

Claude Code will now start and connect to the selected model.

You can start interacting with your codebase immediately.

Official Documentation

For more details, see the official docs:

https://docs.ollama.com/integrations/claude-code

How to connect a local AI model(with Ollama) to VS Code. (Updated)

Sushan — Mon, 01 Dec 2025 14:36:16 +0000

You can try out the latest Ollama models on VS Code for free.

We are using Ollama, which is a free local AI model running application developed by the Llama community.

Installing and using Ollama

You can download Ollama from its website.

Now you'll be able to access ollama using your terminal.

Open your terminal.
Type ollama to verify if it's been installed.
Run a model you like(depending on your hardware), using the command:
ollama run qwen3:4b
This command will pull and run the model.
Change the model name to your preferred model and install it.
To view all the available models, go to ollama.com/search
ㅤ
If you want to run a high-end AI model, you can use Ollama Cloud for free.
Run them like this: ollama pull qwen3-coder:480b-cloud.
ㅤ
ㅤ

Integrating with VS Code

Make sure the Ollama server is running in the background.
Verification: Check this URL, localhost:11434, and see if Ollama is running.
If not: Run it using the command ollama serve.

ㅤ

Open VS Code -> Copilot chat sidebar.
Select the model dropdown -> Manage models -> Add Models(Top right) -> Select Ollama -> Hit Enter ("Ollama" prewritten) -> Type 'http://localhost:11434/' and Enter -> Select the desired models.

Ollama will be added to the list like this.
Make sure to make the model visible by clicking the eye 👁️ sign.

Demo:

If you still don't see your models in the models drop-down in the chatbar, make sure you're in Ask mode.

The option will disappear once you turn the Ollama server off.

And if it still doesn't work:

Get a free Gemini API key from https://aistudio.google.com/api-keys
Select Google model, add API key, and DONE.

Use Ollama with VS Code for free.

Sushan — Thu, 16 Oct 2025 13:43:10 +0000

You can try out the latest Ollama models on VS Code for free.
It also offers Ollama Cloud, which helps you run powerful models.

More: Ollama Cloud

First, pull the coding models so they can be accessed via VS Code:
ollama pull qwen3-coder:480b-cloud

Open the Copilot chat sidebar
Select the model dropdown -> Manage models -> Select Ollama
Select the desired models.
Open the model dropdown and choose the model.

But, Why is Serverless? When should you use it?

Sushan — Sat, 13 Sep 2025 15:22:18 +0000

Serverless sounds like "no servers," but of course, servers still exist in the process.
We all know that it's to help reduce the workload of managing a server for a developer.

But why do we need it? Why does it even exist?, And how did it come into existence?

The Beginning:

So in 2006, during the cloud revolution, companies like AWS started offering VMs(Virtual Machines) on the cloud (with services like EC2 and S3) that developers could rent. This was a low-cost and efficient alternative to purchasing your own physical servers.
It was Infrastructure as a Service(IaaS).

The Simpler Era:

But, even if devs could rent VMs, they didn't want the hassle of setting up and managing these servers, and that's when Platform as a Service(PaaS) showed up.
In 2008, GAE(Google App Engine) was released with the PaaS model. Around the same time, Heroku (est. 2007) debuted in 2009 as a widely admired PaaS platform, letting users quickly deploy their apps.

The Game Changer:

Around 2014, AWS launched AWS Lambda, marking a major shift in cloud computing. You no longer needed to provision or run servers. It enabled the execution of discrete code snippets (functions) triggered by events.
Strictly speaking, it was called “function-as-a-service”.

Lambda marked the birth of serverless computing, and soon other players started catching up. Developers could now write code, and the cloud providers would handle execution, scaling, and billing.
It was not only reliable but damn cheap too, allowing wider adoption among developers.

Now we have:
Microsoft Azure Functions,
Google Cloud Functions, and
Cloudflare Workers...

All of them appeared after AWS introduced Lambda, and they tried to decrease the cost of serverless further.
Serverless was cheap, reliable, and removed a ton of operational overhead.

And now you can see how much Lambda costs.
Source: Amazon AWS

Drawbacks of Serverless:

After all, no technology is complete. Serverless has many benefits, but it also comes with a few tradeoffs.

Cold Start: This is the main issue of serverless. When a function hasn’t been used recently, the cloud provider may spin it up to save resources, which helps bring down the cost of serverless. This isn't ideal for real-time applications like trading platforms, games, etc.
Limited Control: Serverless might not give you much control over your infrastructure due to platform restrictions.
Cost Predictability & Vendor Lock-in Concerns: If you need a predictable billing (for budgeting) or want to avoid being tied to just one cloud provider, serverless may introduce some challenges.

When Should You Use It?

Serverless is good when your service runs sporadically.

For example:

Event-driven tasks like file uploads, DB change triggers, etc.
APIs or microservices with variable traffic.
It's quite reliable for prototyping, MVPs, or for experimental features where you want to move quickly and minimize initial infrastructure cost and risk.

When not to use it:

Just because serverless seems to be cheap doesn't mean that it's the cheapest for every kind of task.

Long-running or compute-intensive tasks that can exceed the execution time, memory, or CPU limits of serverless platforms.
High, steady traffic where functions are always invoked heavily, which may even exceed the cost of alternatives.
Latency-sensitive applications because of cold starts.

The Birth of Python

Sushan — Sun, 07 Sep 2025 13:43:10 +0000

In the late 1980s, computers were buzzing all around, programmers were wrestling with clunky programming languages, and one guy in Amsterdam was getting a little frustrated. It was Guido van Rossum.

Guido van Rossum was employed at a research institute where he was using a programming language called ABC. ABC had some neat concepts as it was accessible for beginners and was sweet and clean, but it was also inflexible.

Try to extend it? No chance!

Try connecting it with system calls? Nope.

Guido appreciated the spirit of ABC, but he wanted something more useful, something with enough flexibility to work through real-life problems.

So, during the Christmas break in 1989, while most were all about unwrapping gifts or singing Christmas carols, Guido decided to treat himself with a new programming language.

Guido wanted it to be:

As easy to read as plain English,
As fun as ABC, but not as restrictive,
Powerful enough to connect to systems. And of course, because Guido had a ridiculous sense of humor, he named it after the British comedy ‘Monty Python’s Flying Circus.’ (Yup, Python has nothing to do with 🐍 snakes!).

At first, it was named Monty Python, but it didn’t sound quite right, so Guido shortened it to Python.

A few months later, this little holiday side project began to grow legs. By 1991, Python had an official release, 0.9.0, with core features like functions, exceptions, and even modules.

It quickly gained some following. Programmers were loving it. It was easy without being trivial. It was flexible without being unstructured. Other languages made you feel like you were solving complicated math puzzles. Python made you feel like you were well… talking to computers.

Eventually, what started as a Christmas experiment by Guido van Rossum became one of the most beloved programming languages on the planet. From powering Instagram to helping in NASA space missions and AI, Python slithered into almost every corner of technology.

In 2024, it became the most popular programming language worldwide. (Though some surveys still place JavaScript just ahead)

The best part? It never lost that human touch Guido was going for. Python wasn’t meant to make machines happy. It was meant to make programmers happy.

The Accidental Birth of NodeJS!

Sushan — Wed, 13 Aug 2025 15:21:14 +0000

Ever wonder how Node.js came to be? Here’s the story of a frustrated developer and a wild idea that changed backend dev forever.

It’s 2009.

Ryan Dahl is sitting at his desk, frustrated. He’s been wrestling with the same problem for days:
Why do web servers just stop and wait?

He's building a system to handle real-time file uploads, but the servers of the time are blocking — they freeze until each file operation finishes.
Ryan wants something faster, more fluid, something that can juggle thousands of connections at once.

First, he tries C.
Too low-level and too much pain.

Then he tries Ruby and Python.
Still blocking and still slow for what he needs.

Then, almost by accident, he stumbles upon something new:
Google’s V8 JavaScript engine, which was built to make Chrome blazing fast.

JavaScript on the server seemed to be an absurd idea at the time.
JS was the language for pop-ups, form validation, etc., not heavy backend work.

But he thinks V8 is fast.
So Ryan does something unusual — he embeds the V8 JavaScript engine inside C++ code, then writes an event loop and hooks it up to the system’s I/O.

Finally, the result? A platform where JavaScript can run outside the browser, talking directly to the operating system — and doing it without blocking.
He had never planned to create a backend revolution. What he needed was just a tool for his own problem.

But later, when he released Node.js, developers around the world saw its potential.
And what began as an experiment became one of the most widely used backend technologies in history.
Sometimes, the future is built by accident.

Basics of MCP

Sushan — Tue, 12 Aug 2025 13:41:04 +0000

Here are some helpful terms for MCP users.

Why are so many devs ditching WordPress?

Sushan — Thu, 07 Aug 2025 15:51:17 +0000

Because they’ve met: Headless CMS

But, what's the difference b/w Docker and VMs?

Sushan — Tue, 05 Aug 2025 16:21:21 +0000

But, what's the difference b/w Docker (a containerization platform) and VMs (Virtual Machines)?

Here's the difference b/w the architecture of

VMs
Docker (A containerization platform)
Docker (on non-Linux-based systems)

But, what does Qwen3-235B-A22B-Thinking-2507-FP8 mean?

Sushan — Sat, 02 Aug 2025 15:29:28 +0000

Here's the information it holds.

235B: 235 billion parameters.
A22B: 22B is active at any given time.
2507: Release date — July 2025.
FP8: Uses 8-bit floating point precision.

[Boost]

Sushan — Thu, 31 Jul 2025 15:02:08 +0000

Sushan

Jul 31 '25

Why use MCP over APIs?

#mcp #api #ai

Comments

1 min read

Why use MCP over APIs?

Sushan — Thu, 31 Jul 2025 15:01:17 +0000

Here's a summary of MCP and its advantages.