Open Source AI

Best Open-Source AI Project

Build something original with open-source or open-weight AI at your Hacktoberfest Hack Day. Your project could be an agent skill, an application powered by a small or large language model, or an original or meaningfully improved model harness. These are starting points within one prize category; you can combine them or take a different approach.

Every in-person Hacktoberfest Hack Day runs this challenge. Your host will tell you when project work and submissions open, when they close, and how demos will work.

What your project needs

• Make open-source or open-weight AI an important part of how the project works. Be ready to show what the AI does in your demo.
• Publish the project in a public GitHub repository with an open-source license. Include a README that explains what you built, how to run or try it, and which model and key dependencies you used.
• If you build an agent skill, follow the Agent Skills open standard.
• If you enter a model harness, show your original implementation or meaningful changes to an existing open-source harness. A renamed copy alone does not meet this requirement.

An open-weight model makes trained weights available to download or run under its provider’s terms. Open-weight does not automatically mean the whole model is open source. Name the model and link to its license or terms in your README. For this challenge, a small language model has 10 billion parameters or fewer; a large language model has more than 10 billion.

Useful resources

Choose and run a model

• Hugging Face Transformers quickstart — load a pretrained model, run inference, or explore fine-tuning.
• Ollama quickstart — get started with a local model and build with its API.
• Gemma model overview — examples of open-weight models at different sizes.
• Hugging Face model cards — check a model’s intended uses, limitations, and license before building with it.

Make your work reusable

• Agent Skills specification — the required format if your entry is an agent skill.
• OpenCode agent skills guide — an example of how an open-source agent harness discovers and loads reusable skills.
• Open Source Initiative license list and GitHub’s licensing guide — choose a license and add it to your repository. A public repository without a license does not meet this challenge’s open-source requirement.

These tools are examples, not required technologies. Check each model’s and dependency’s terms for your use case.

DigitalOcean for your Hack Day

DigitalOcean supports this prize category. You can use it to host an app, API, or model workflow, but using DigitalOcean is not required to enter Best Open-Source AI Project.

Attendees registered for an in-person Hacktoberfest Fest can find the $25 DigitalOcean credit in My Hacktoberfest promos. Sign in, read the offer terms, and claim close to your Hack Day: current credit links last three full days once activated and expire after October 31, 2026. The credit cannot be used for Serverless Inference proprietary models. Check the terms shown with your offer before activating it.

• App Platform quickstart — deploy an app or API from a Git repository or container image.
• Droplets quickstart — use a virtual machine when you need control over your environment.
• GradientAI documentation — explore hosted inference and agent workflows. Check model availability, product pricing, and credit eligibility before deploying.

Submit your project

One team member should submit through your event’s OrganizerHQ Challenges page before the host’s deadline. Add the public GitHub repository, project name, description, technologies used, and a demo URL if you have one. Select Best Open-Source AI Project under “Which challenge(s) are you submitting to?” You can select any other challenge your project is eligible for in the same submission.

Before submitting, check that someone outside your team can open the repository, find its license, understand the AI contribution, and follow the README to try the project. Your host may ask you to demonstrate the working project and explain what your team built during the event.

How the winner is chosen

The host selects one winning team. A strong entry meets the requirements, solves a clear problem or creates a useful experience, shows substantial work by the team, and has a working demo with an understandable explanation. Judges should be able to verify the source, license, model, and relevant dependencies. Pull-request counts do not decide this prize.

Prize: Every member of the winning team receives a DEV Badge on their DEV account. Depending on availability and the event’s onboarding instructions, MLH may also provide MLH+DEV branded swag bags. Ask your host which physical items, if any, are confirmed for your event.