11 open source AI projects that developers will love

The most creative and impactful ideas in software often appear as open source. Sharing code invites a global community of contributors to work together to drive a project forward. Artificial intelligence is no exception. While AI’s need for computational firepower might constrain individual achievement, a distributed team can get the job done. AI is one technical arena where the sum of all efforts is truly greater than the parts.

In the interest of nurturing this vital sector, here are 11 open source AI projects that make a developer’s job a little easier. Some fine-tune the process of training models, while others help you find or access the data sources you need. There are productivity hacks, performance optimizers, SQL wranglers, and more.

Together, these projects let us see how innovations in AI are turning the dream of a smart, sentient computer into an everyday part of our lives and work. Take them as a starting point for your own projects, or as a blueprint for how to create something amazing when inspiration strikes.

Upscayl

Sometimes, an image just needs a bit more detail to look good on a page. Upscayl (see the GitHub project here) increases image resolution for the crispness and detail you seek. If you’ve got the right hardware, it’s a good way to enhance digital artwork or add detail to a photograph. Just remember that the AI is pretty much hallucinating these details. That means Upscayl is ideal for enhancing fictional images created by a digital artist, but it’s not as good for images that require absolute accuracy, such as documenting evidence at a crime scene.

Nyro

Developers spend a fair amount of time interacting with the computer’s operating system via the command line. While they are easy to overlook, all those seconds add up. Nyro is an open source project written on top of Electron that handles mundane tasks like taking screenshots, resizing windows, and synchronizing data between applications. Automating everyday tasks like these can save you many small fractions of time, which ultimately adds up to a big productivity boost.

Geppetto

Some development teams do most of their work in Slack channels, so the posts end up being pretty solid first-generation documentation. Geppetto is a Slackbot that connects your channels with several different LLMs (OpenAI, Anthropic, and Gemini), which can clean up and enhance your musings. Geppetto will even send a request to Dall-E if you want art to add life to your documentation.

E2B sandboxes

The earliest LLMs answered questions and maybe generated a bit of art using all the knowledge in their training set. But what if they were free to roam the Internet and use all the same tools that humans use? E2B is an agent sandbox that lets LLMs connect with many of the same tools that we humans use every day. That means web browsers, GitHub code repositories, and command-line tools like linters. LLMs can then use the power of these tools to do useful things like manage cloud infrastructure, so humans don’t have to.

Dataline

Not everyone wants to upload all their data to some distant AI GPU for training. Dataline uses an LLM to generate SQL commands that suck the data out of the database. Then, the code creates a data science report using a local connection to the local data. It’s a hybrid approach that merges classic data science algorithms for analysis with LLMs that guide them.

Swirl Connect

Sometimes, you want to start working with a data set but you don’t want to go to the trouble of extracting and reformatting it. If the data set is large, these processes can be very time-consuming. Swirl Connect (see the GitHub project here) links many standard databases with most standard LLMs and RAG search indices. All the data you need is in one place, and you can just focus on the training.

DSPy

The emergence of LLMs has created a whole new job specialization in prompt engineering. Unlike the algorithms that developers use, prompt engineers fiddle with words and write long instructions that wheedle and nudge an LLM to produce just the right result. This is a role that requires the gift of gab and the ability to use Jedi mind tricks on LLMs. DSPy is a tool that wants to bring a more systematic approach to LLM training. Instead of words and phrases, DSPy connects modules and optimizers and arranges them in a pipeline for the LLM. Developers using DSPy can spend less time worrying about linguistic nuance and more time working with code.

Guardrails

One of the challenges of generative AI is keeping the AI from straying off course. The engineers of Portkey Gateway found a way to integrate more controls into the generative AI pipeline. Asynchronous functions, known as guardrails, can track the evolution of AI-generated answers and “vote” at various stages of the pipeline. With each vote, an answer is refined. The end result should be fewer hallucinations and more correct answers.

Unsloth

Training a foundational LLM on a new set of data is often expensive. Unsloth is a tool designed to optimize such training for some of the most common open source models. By some accounts, the open source version of the tool is two to five times faster than model training without Unsloth, and the professional version is as much as 30 times faster. Carefully handwritten kernel code is applied in a way that lowers memory usage while maintaining or even increasing accuracy.

Wren AI for SQL

Most data in the world is stored in vast tables, often accessible with SQL. Alas, few people know how to write great SQL queries. Even good programmers struggle with writing fast and efficient SQL queries. Wren AI is a natural language front end to SQL. You ask your questions in plain English and the AI translates them into SQL, saving everyone a bit of time and grief.

AnythingLLM

Many people these days have a massive pile of digital documents tucked away somewhere for future reference. The challenge is finding that perfect quote or data point when you need it. AnythingLLM organizes your pile of documents into something useful. You just feed your documents into any LLM or RAG system and then query it for the answers you need. The tool runs on Linux, macOS, or Windows machines, and responses can be in a variety of formats including speech-to-text.