Own your own models
Owning a language model end to end used to take a lab, a Python stack, and a cloud account. With Smedjan it takes one Rust binary and your own hardware — one person, or one company, can train it, run it, fine-tune it, and keep it. Whether you build one is your decision to make, and a model that lives in your repo cannot be revoked, throttled, or switched off when someone upstream changes the rules. That independence is yours to pick up — and the whole engine is right here, waiting to be forged.
The belief
The default path to a language model runs through someone else's framework, someone else's runtime, and someone else's cloud. That's fine for a demo and wrong for ownership. Smedjan is the opposite bet: a complete engine you can read, change, and run yourself — so a person or a company can build an agent that's genuinely theirs, on hardware they control.
Get involved
Write code
The repo has good first issues and a CONTRIBUTING guide. Kernels, mixers, docs, tests — all welcome, all MIT-licensed.
Report & discuss
File issues, propose features, ask questions. A clear, reproducible bug report is gold for a one-person project.
Sponsor
Fund hardware hours and independent development so the project stays free and unowned by any vendor.
Collaborate
Companies — especially in Sweden — wanting private or on-prem models can partner on features and support.
Conduct
Be respectful, assume good faith, keep it technical. Harassment of any kind isn't tolerated. The full Code of Conduct lives in the repository.