Open systems research / Est. 2026
Hyphae Research Foundation cultivates open infrastructure for local, durable computation: a verifiable data engine and a governed laboratory for transformer research.
The initial commons
A local-first Rust data engine unifying SQL, native structures, lexical and vector search, recovery, and verifiable proofs in one binary.
Examine the data engineA PyTorch framework for controlled research on identity-initialized residual learning, with preregistered experiments, budgets, immutable manifests, and reproducible reports.
Examine the research frameworkA correctness-first CPU runtime for supported ternary networks, with exact integer accumulation, family-gated model loading, bounded RAM layouts, and phase-aware kernels.
Examine the inference runtimeResearch ledger
We distinguish stable releases, measured experiments, active development, and known limits. A number without its protocol is not a result.
Open the ledger33 million observations across data, search, and durable commit surfaces.
Eight paired seeds, immutable manifests, and a preregistered practical-effect threshold.
Rejected optimizations, retractions, limits, and the method used to decide.
From the manifesto / § 04
“A system is not trustworthy because it says the right thing. It is trustworthy when its boundaries, history, and evidence can be examined.”Read all eight commitments
Build the institution
We are forming the governance, rights, security, and contribution structures that let technical work outlive its original authors. The unfinished parts are listed in public, not hidden behind a foundation label.
Review formation status