There are decisions that feel tactical, and then there are decisions that rearrange the direction of everything you build after them.
Creating the Hyphae Research Foundation belongs to the second category.
For a long time I worked under the assumption that the natural path for serious technical work is to productize it. You build something valuable, you wrap a company around it, you find a way to charge, you try to scale. That is the default script most of us inherit.
I followed parts of that script. Celiums existed. Real work was being done. Performance was good. There was interest. The infrastructure I was building — especially the data engine — was already becoming foundational to how I thought about computation, training, and recovery.
And then I reached a quiet but uncomfortable realization:
Some systems should not be owned.
Not because they lack commercial potential.
But because their long-term value depends on remaining free, inspectable, and independent of any single company's incentives.
That realization is what led me to create Hyphae Research.
The weight of keeping something free
It is easy to talk about open source when the project is small.
It becomes harder when the work has taken years, when the architecture is non-trivial, and when you are the only person who has carried every layer of it.
I have been the sole founder and the only engineer on this stack. Every design decision, every recovery path, every proof mechanism, every research experiment — it has all gone through my hands. There is a particular kind of attachment that forms when you build alone for that long. You start to feel the system as an extension of your own thinking.
The temptation in that situation is to protect it. To keep control. To turn it into something that can generate revenue and justify the years already spent.
I chose the opposite.
I decided that the more important outcome is that the work outlives me, and that it remains usable by people who will never pay a license fee or depend on a company's roadmap. That is not a romantic statement. It is a practical one. Infrastructure that becomes critical should not be subject to acquisition, policy changes, or the quiet death of a startup.
Making this decision required accepting a different kind of success metric. Not growth charts. Not valuation. But whether the systems remain coherent, local-first, and verifiable years from now.
What is actually being opened
The foundation currently holds three projects. They are not demos. They are the concrete results of the work I have been doing:
Hyphae is a local-first Rust data engine. It unifies relational SQL, native structures, lexical search, vector search, recovery, and verifiable proofs inside a single process that does not require a network. The design prioritizes durability and inspectability over convenience features.
Hyphae Transformer is a controlled research framework built on PyTorch. It exists to make certain classes of experiments reproducible by construction: preregistered runs, immutable manifests, budget constraints, and reports that can be audited rather than just claimed.
Hyphae BitNet is a correctness-first CPU runtime for supported ternary networks. The emphasis is on exact integer accumulation, bounded memory layouts, and predictable behavior rather than maximum theoretical throughput.
Together they form the beginning of an open infrastructure for local and durable computation. Not another cloud abstraction. Not another managed service. Something that can still function when the network is gone, and that can prove what it claims.
Why this feels like the right decision
I have spent enough time around both research and production systems to know how quickly important work can disappear. A company pivots. Funding dries up. Priorities shift. The people who understood the original design leave. What remains is often a partially maintained artifact that no longer reflects the original intent.
By placing this work under a foundation — currently under interim stewardship by Celiums Solutions LLC, with the explicit goal of remaining non-profit — I am trying to reduce that risk. The decision is not about rejecting commercial activity entirely. It is about refusing to let the core infrastructure become dependent on it.
This is the best decision I have made as a builder so far, not because it is the easiest, but because it is the one that aligns most closely with the actual nature of the work. Systems that aim for durability and verifiability should themselves be durable and verifiable in their governance.
I do not know yet how large the community around this will become. I only know that I am no longer willing to keep the work private while pretending that "later" is a responsible timeline for opening it.
An invitation without the usual language
I am not going to end this with a polished call to action.
If the ideas above resonate — if you care about local-first systems, about evidence over marketing claims, about research infrastructure that can be audited, or about the quiet work of making computation more durable — then the repositories are open.
You can start here:
https://hyphaeresearch.org/
I built this alone up to this point. I do not intend to keep it that way forever. But I also will not pretend that community happens automatically just because the code is public. It takes people who are willing to read carefully, disagree productively, and stay long enough to understand the constraints.
If that describes you, I would be glad to have you involved.
This decision cost me the simpler path.
I still believe it is the correct one.
Top comments (0)