The Bike4Mind open-core platform is up and public on Github!
We're opening up Bike4Mind open-core, under the Business Source License 1.1.
The most important thing to me in the advancement of AI is that we don't let it become a machine that makes you, me, and everyone else redundant. It should be a bicycle for the mind. That's literally what our name means. Steve Jobs used to talk about how a human on a bicycle is the most efficient creature on the planet, better than a condor, and that a computer could be a bicycle for the mind. A good tool takes a capable person and makes them far more capable, like how the bike lets the rider get a lot further on the same effort than they would on foot. But it doesn't remove the human rider who is still using those same legs to turn the pedals!
The second reason we built this is that serious AI tooling shouldn't only live behind the walls of two or three big labs. Their models are genuinely great and we support them on our platform, of course. But I don't love renting every last piece from OpenAI ("open" 🙄) and Anthropic and hoping the terms there stay good forever. We wanted something you can read, run, own, and host yourself. Something you can put on your own machines, point at your own model keys or local models, and keep running no matter what any one vendor decides to do next. After almost three years of building it privately, that's what we're putting out in the open.
Bike4Mind is a TypeScript monorepo AI workspace. You get chat, notebooks, and agents running across a lot of different models, so you're never locked to a single provider. Bring your own API keys or buy credits on our hosted instance -- one flavor of credits that gets you access to all the models and features. Countless MCPs/tools already supported and now you can contribute your own, too. There's knowledge base and data lake support, and even our own ink based CLI. It runs on AWS through SST with MongoDB, and you can stand the whole thing up on your own hardware with Docker.
Before we opened it, it had just about 3 years of closed source work and progress:
- 28,196 commits
- 7,315 pull requests (6,863 of them merged)
- 2,551 issues filed
- 30+ active human contributors
- First commit: August 3, 2023
The License
We're launching under the Business Source License 1.1, which is source-available and free to use in many many ways, just not full OSI "open source" on day one. This means you can read every line of the source, self-host it, run it for yourself or your company or your team, fork it, and build products on top of it and even sell those products. Running it internally for your own AI sovereignty, independent of any hyperscaler or model provider, is explicitly fine. That's a use we actually want to encourage.
The one thing you can't do: stand up your own competing hosted instance "as-a-service" and sell access to outside third parties. That's the one carve-out for us. That said, every released version automatically converts to Apache 2.0 exactly two years after it ships. So it's on a clock, written into the license, and we can't retroactively take it back.
These launches usually go bad the same way. A company starts permissive to win adoption, then quietly claws the freedoms back once everyone is hooked. Redis, Elastic, HashiCorp, and Mongo all did a retroactive relicense, and all of them got forked by people who felt burned by that. We're in a different spot. We have no prior open license to rug-pull, and the BSL-to-Apache clock only runs one way. So whether you trust us personally or not, you can trust that timer!
Self Hosting
There's a Docker Compose path and a SELF_HOST.md that walks you through it. Bring your own model keys, your own Mongo, and your own compute, and the whole thing runs on your infrastructure. The self-host stack is still being finalized, so expect a few rough edges for now, but the path is real. Self-hosting is genuine work, too. You own the boxes, you own the API bills, and you own the ops when something falls over at 2am. Some people will read SELF_HOST.md, nod, and decide they'd rather pay us to run it. That's a completely fine outcome, and to be straight, it's the main way open core will work for us as a business. But the self hosting option is there and available to everyone that does want it!
Our Instance
If you just want to use it without fighting your own infrastructure, there's our hosted version at bike4mind.com. Credits roll over month to month and cover all the different providers/models/features. They're what actually pay the model providers real money under the hood, so heavier use goes through credits faster, the way you'd expect. You don't have to pay anything to try it, either. New accounts start with 10,000 credits, roughly $10 of usage, on us. That's a launch special for now, so it's a good time to sign up and try it before you commit to anything.
TLDR: Read the code, run it, try to break it, host it on your own machines, or let us run it if ops isn't how you want to spend your nights. File issues, fork it if you want, and send PRs. It's out in the open and we're really excited to see how everyone uses it and contributes -- and we aren't done building either!