I by no means anticipated Docker to deal with doc administration and AI chatbots, however right here we’re

Over time, I’ve found a number of Docker containers/apps I can’t work without. FOSS apps are constantly high-quality, however some apps kick it up a notch. I beloved the app a lot and received to interested by related apps. Apps you would possibly jokingly say “do stuff you would not anticipate from a Docker container” as a result of they’re premium-level, feature-packed packages which are (often) free, self-hostable, and containerized. Here are just a few of one of the best ones I’ve discovered.

Paperless-ngx

For making a searchable doc database

We’ll begin with one thing close to and pricey to my work: doc administration. Paperless-ngx can deal with something I throw its approach. It aligns with my curiosity in constructing and sustaining a Knowledge Base in a seamless approach whereas additionally permitting me to have a hub for all my paperwork.

Paperless-ngx is open-source and is supposed to construct a searchable database of bodily paperwork. The concept is you may scan or {photograph} your paperwork and add them into the database. I’m utilizing it to handle bills, preserve observe of music concepts and lyrics, in addition to constructing a database of manuals and chord charts.

Paperless-ngx makes use of OCR (optical character recognition) to learn and categorize the paperwork you add. To set up it, I used the Docker Compose YML file, however made just a few changes to get it to my liking.

Once I began the container with docker compose up -d, I created a login and received began. The interface could be very easy and intuitive. The concept is you simply drag and drop your paperwork on the correct facet, and it will begin constructing the index from there. You also can change OCR settings, AI settings, and add an software brand.

For my preliminary check, I used it to arrange and again up paperwork associated to an album venture I’ve been kicking round for the previous few years.

I set my OCR output to PDF and saved all the things in the usual configuration. I made positive all AI capabilities have been switched off, since I did not wish to use them for this explicit venture. There are going to be some proprietary paperwork in right here that I do not want to share with LLMs. I added my brand (a flying cheeseburger named Clyde) and began importing information.

Setting up information for the doc is fairly simple. When you click on on the doc or picture, it opens up a sub-menu the place you may add the suitable knowledge (serial quantity, creator, and so forth) and tag it. This is an instance of my “brand:”

From there, I started importing applicable paperwork: the track listing, lyrics, HTML information, chord charts/photographs, and different knowledge related to the album venture. Once all the things was loaded, I used to be capable of actually begin engaged on the venture (a venture that I’d been pushing aside for a decade at this level).

I wasn’t anticipating to have the power to construct such a database with a containerized app. It’s fairly easy and helped me begin a really bold venture that I’ve been pushing aside for some time. I plan to make use of this sooner or later for extra than simply my bold music venture, and I’m pleasantly stunned by how good it’s.

Open Web UI and Ollama

A match made in Docker heaven

I do know, I do know. I’ve made it crystal clear that I’m not a fan of LLMs prior to now (largely due to the environmental influence, dodgy info, disdain for AI artwork, and the best way the weirdos answerable for it market them). That hasn’t modified. But I do have an open thoughts about tech and through my analysis, I could not cross up the chance to speak about utilizing Docker to self-host an area LLM.

I’ve really been curious about Ollama for some time now, so I took this as an indication to put in and take it for a check run. It makes use of Open WebUI.

Ollama has been obtainable as a Docker picture since 2023. I attempted it as a result of I wish to self-host issues and since it could actually run offline with out sending my knowledge anyplace else. Once put in, I chosen llama 3.2:1b (I plan to attempt 3b and mistral as nicely) because the mannequin to run, because it’s fairly light-weight and my check machine might deal with it.

I arrange the Docker Compose file after which pulled the mannequin instantly in with docker exec -it ollama ollama run llama3.2:1b.

The interface is fairly simple. The AI’s a bit gradual, however that is to be anticipated. I began by asking it about its limitations and capabilities. Then I made it create puns and a brain-teaser. It wasn’t superb at puns, I’m sorry to say. It resorted to dad jokes fairly quick, however I did chuckle on the espresso cup being mugged. Peak LLM humor.

The brain-teaser was a horrible failure. I answered it, the mannequin received confused and began coding…one thing. Let us by no means converse of it once more.

I by no means would’ve anticipated Docker to deal with an AI chatbot, nevertheless it works, and I can see loads of potential makes use of sooner or later.

Docker continues to amaze and shock me

Sometimes you discover a good app in an sudden place. Docker is filled with these sorts of surprises. While I’ve solely explored just a few right here, I’m nonetheless discovering new and helpful software program. I’ve discovered Docker to be a strong companion to my work, hobbies, and the simply plain absurd tasks that I love to do in my free time.