Paying monthly for Claude made sense until I actually looked at what I was getting versus what I was giving up. The cost adds up, and I’m sure that more people are starting to see this too. If you’ve had that same realization, the tools below are worth knowing about. They cover the main features that Claude has without the major downsides and takedowns.

You can run models right on your computer

It’s the fastest way to drop monthly fees

I’ve tried the Claude subscription, and it’s not worth it to me. I use Llama.cpp because it’s the fastest and lightest way to run an LLM on your own machine. It’s an inference engine built from scratch in C and C++ to be as lean as possible. Since it compiles down to a single portable binary, you don’t need to wrestle with heavy dependencies, virtual environments, or frameworks like PyTorch.

It talks directly to your CPU and GPU and gets out of the way. Aside from being easy to set up, it doesn’t have monthly fees, has no per-token API costs, and none of your data ever touches a remote server.

The biggest challenge with moving from the cloud to your own hardware is memory. There are plenty of quantized models you can use to run on it. I’ve had both Firefox and Chrome on while running Llama and only noticed my fan was working harder, but it didn’t really affect my PC as a whole. So this is an easy replacement for Claude if you’re willing to set it up.

Keep your code private while you working

This extension plugs local models right into your editor

Credit: Nick Lewis / How To Geek

Cloud-based coding tools like Claude or GitHub Copilot are genuinely useful, but they come with serious privacy issues. Every time you use them, your source code, architecture decisions, and business logic are being sent to someone else’s servers. For teams with strict compliance requirements, or anyone who cares about keeping their intellectual property off external systems, that’s a real problem.

Continue.dev is an open-source extension for VS Code and JetBrains that lets you connect locally hosted AI models directly to your editor. Everything runs on your own hardware, so your code never touches a third-party server. No data transfers, no wondering where your code ends up.

Hooking it all up is pretty easy since Continue.dev doesn’t come with its own backend, so you can attach it to whatever local inference engine you’re already using. So you can use Ollama, LM Studio, or a llama.cpp server. Configuration happens in a single file where you define which models handle which tasks.

Connect your local model to the rest of your apps

A visual editor lets you automate tasks without a wall of code

Credit: n8n

Running a massive language model on your own hardware is impressive, but an AI is only as useful as what it can actually do. Luckily, n8n is an open-source tool that connects your local models to everything else you run. So your LLMs get to talk with your other apps and move data around automatically.

Something like a paid Claude subscription keeps you locked into its platform and charges you extra just to connect to outside services. n8n is free and hooks your local inference server directly into your email, databases, and third-party tools. It does all this through a visual, drag-and-drop editor, so you don’t have to write a wall of code just to get things talking to each other.

The best part about n8n is that you can set workflows to fire based on very specific things happening. You can even have it filter out spam and newsletters before the email ever reaches your AI, so you’re not burning compute on junk.

Bring your local AI straight into your browser

You can chat with webpages and search the web locally

Credit: n8n

One of the biggest issues with running AI models locally is that they’re cut off from the internet. Page Assist fixes that. It’s an open-source browser extension that connects your locally hosted AI directly to your browser as a sidebar. Instead of switching back and forth between a terminal window and your browser, your local model is just right there while you work.

The most useful thing it does is let you chat with whatever webpage you have open. Normally, you’d have to copy the whole thing and paste it into a prompt yourself. Page Assist scrapes the page and feeds the contents straight to your model.

It works on many pages because it pulls the page’s rendered HTML, converts it to clean Markdown, and drops the whole thing into the model’s context at once. This is great if you’re running a model with a large context window, since it can take in an entire article or technical document in one shot without missing anything.

The second way is better for longer content or when you’re working with less memory. It breaks the page into chunks, indexes them using a local embedding model, and then pulls only the sections that are actually relevant to your question.

Page Assist also comes with a built-in web search and a personal knowledge base where you can upload your own PDFs, CSVs, and Word documents and chat with them right from the sidebar. There’s experimental support for the Model Context Protocol if you want to push the tool integrations further, but you don’t need to.

The extension works with pretty much any local AI backend, whether that’s Ollama, LM Studio, llama.cpp, or even Chrome’s built-in Gemini Nano. You just point it at your local API endpoint, and you’re good to go.

Since everything stays in your browser’s local storage, none of your browsing data goes anywhere. Your local model gets to read the live web, and nothing leaks out to an external server.

An afternoon of setup gives you total control over your data

None of these tools is a one-click replacement. Getting everything working together takes some setup time, and you’ll need hardware capable of running models without grinding to a halt. The tradeoff is that once it’s running, there are no monthly fees, no data leaving your machine, and no vendor deciding what you can or can’t do with the tool. If that sounds worth an afternoon of configuration, it probably is.