With every tech firm and software developer slapping artificial intelligence onto their products, it’s easy to get burned out of these so-called AI features. But as much as I despise the AI bubble, I have to admit that machine-learning tools have their own niche uses. Take locally-deployed LLMs and image generators, for instance. While they may seem like standalone services that can only generate soulless text and images, you can use them to bolster the capabilities of a handful of apps and services.

Karakeep

Auto-generate summaries for your bookmarks

Previously called Hoarder, Karakeep is a powerful bookmarking application that can help you build a digital archive from web pages, clipboard notes, PDF files, and even images. Besides its OCR capabilities and compatibility with RSS feeds, Karakeep can save entire web pages to prevent link rot from making your favorite posts inaccessible.

If you’ve got an LLM running on a local server, pairing it with Karakeep will allow it to generate automatic tags for your web links and images, making it a lot easier to organize your bookmark collection. Plus, an LLM-powered Karakeep server can also generate short summaries for your bookmarks, making it a lot easier to sift through your digital library if you have a habit of pinning tons of posts, links, and documents like I do.

VS Code

You’ll need a powerful GPU though

Capable of running on practically every OS out there, VS Code is a lightweight yet powerful coding editor. Between its built-in debugger, terminal, and Git integration, VS Code is already a solid option for most coders, but it’s the Extension Marketplace that elevates its utility to the next level.

While there are dozens of cool extensions that can enhance your coding workflows, AI-powered add-ons are surprisingly useful. Continue, Granite, and a handful of AI-centric extensions let you harness local LLMs to optimize your code, use inline suggestions, and help out with debugging your programs. The only caveat is that you’ll need a lot of VRAM for the large parameter models, while the lower-end models aren’t good enough for coding tasks.

GIMP

Bet you didn’t see this one coming

I have several issues with Adobe’s treatment of customers, but its AI tools aren’t one of them. Photoshop, in particular, has a plethora of cool AI-powered utilities that make editing images a lot easier. Unfortunately, I parted ways with Adobe months ago, and GIMP has been a decent alternative to Photoshop – even more so after I encountered the gimp-stable-boy repository while doomscrolling on GitHub.

Although it can’t replicate all of Photoshop’s AI tools, this neat repository can invoke AUTOMATIC1111’s web API to combine Stable Diffusion with GIMP. AI-powered inpainting works really well when you pair it with a decent model, and you can even use your locally-hosted Stable Diffusion instance to enhance the rectangle selection tool’s detection capabilities. Like VS Code’s extensions, you’ll want to use powerful models for the best results, so it’s not very useful for folks with weaker GPUs.

Paperless-AI and Paperless-GPT

Make Paperless-ngx even more powerful

If you find document management as soul-draining as I do, you should look into Paperless-ngx. Besides supporting most of the popular document formats, Paperless-ngx can turn any old image into a searchable document using its OCR provisions. You can configure Paperless-ngx to automatically pull files from your emails and store them alongside your other documents. Meanwhile, its powerful search capabilities and the ability to bulk-edit tags make organizing documents a lot easier.

The best part? Paperless-ngx has a couple of companion applications that pair well with locally-hosted LLMs. The first one, Paperless-AI, can analyze your document collection and generate the right tags depending on the contents of each file. It also improves upon Paperless-ngx’s search functionality, as it can retrieve documents from their context instead of simply relying on keywords. Then there’s Paperless-GPT, which includes AI-powered OCR capabilities on top of the automatic tagging and LLM-enhanced search facilities of Paperless-AI. I’ve found Paperless-GPT a lot more useful than its AI counterpart, though its interface can feel rather rudimentary at times.

Home Assistant

A self-hosted home assistant inside HASS

Home Assistant is a common name in the self-hosting landscape, and it more than lives up to its reputation. Compatibility-wise, HASS supports most of the popular smart devices and IoT paraphernalia, and you can integrate even more gadgets with it using the all-powerful Home Assistant Community Store. Its automation provisions are also top-notch, and you can also design sleek dashboards for all your smart home gizmos.

By default, HASS includes a decent voice and text assistant, but the real fun begins once you connect it to a self-hosted LLM. Even on a low-parameter model, you can expect precise answers for your queries, though it’s a good idea to use 7b and high-end models for the best results. If you’re in a tinkering mood, you can build a voice assistant by pairing your HASS instance with text-to-speech and speech-to-text models.

Some more ways to use LLMs and image generators in your workflows

For folks who love reading RSS feeds, you can use RSSBox with an LLM to automatically translate web pages into different languages. And as ridiculous as it may sound, it’s possible to bring responsive NPCs to Skyrim with the Mantella mod. While it requires powerful LLMs and a little bit of elbow grease to set up, there’s nothing quite like conversing with NPCs, who not only respond to weird questions, but also retain their original personalities from the game.