Hey everyone,
We’re running an online hackathon around a simple problem: even a capable chatbot can forget everything about its users when the conversation ends.
We’ve been working on Walrus Memory, a persistent memory layer for AI agents. It lets a chatbot store and recall useful context across conversations instead of repeatedly asking users for the same information.
For this hackathon, we want to see what happens when builders combine that memory with models such as Llama, Mistral, Qwen, Gemma, DeepSeek, or anything else available through Hugging Face. You can build something new or add memory to an existing chatbot. It could be a support bot, tutor, personal assistant, companion, Discord bot, internal helpdesk, game NPC, or something completely different. You can deploy it as a Hugging Face Space, website, messaging bot, CLI, or anywhere else people can actually use it.
There is a specific prize category for projects whose primary model is not from OpenAI or Anthropic, and those projects can also win one of the main prizes. There are $2,500 in total rewards across all categories.
The hackathon is already running and submissions close on October 9. It is free to enter, and you don’t need any previous experience with Walrus Memory.
Full explanation, resources, rules and registration:
Happy to answer questions, especially if you’re considering building with an open or locally hosted model.