Read Status before anything else. Mzizi-lang is a Phase 0 research
prototype. A lexer, a recovering parser, an agent diagnostic protocol, a
content-addressed IR and nine primitives exist and are tested. Nothing renders, nothing
lowers to Rust yet, contract bodies parse but are not evaluated, and the benchmark that
the whole thesis rests on has not run. These pages document a research programme in
progress, not a shipping framework.
The claim worth defending
Every web framework in existence was designed assuming a human is typing, reading docs, and holding context in their head. The charter’s bet is that this is the wrong assumption for where development is going:Mzizi’s single sharp edge: a Rust framework whose syntax, type system, and compiler feedback loop are designed for machine authorship, not just human ergonomics.And then a correction that turns out to matter more than the original claim. RFC-0001 was written by a frontier model optimising for its own experience; RFC-0002 withdrew that target:
The design target is small models. A 7B open-weight model running locally, not a frontier model with a million-token context. Frontier capability is improving on its own; the language cannot help there and does not need to.That inversion is load-bearing. It is why token efficiency is a first-order goal rather than a fading one, why
end component button echoes the name back at you, and why the grammar is
small and closed rather than familiar and open. A familiarity penalty is paid once. A large
decision space is a tax paid on every generation, and it scales inversely with model size.
Where to start
Status
What exists, what is tested, what is measured, and what is still just a design
document. The honest version, which is shorter than you would like.
Syntax
Nine named failure modes an agent hits writing Rust UI code, and the syntax decision
that answers each one.
The nine primitives
Real
.mz source, gated in CI, plus the packaging decision — why primitives are
copied files and not crates.The compiler
mz check, mz outline, mz hash, mz ir, and the NDJSON protocol written for a
reader holding zero file context.The IR
Content addressing: one architectural decision that answers seven of eight named
reading barriers.
The ecosystem
Which repository serves which domain, and which parts of it are real today.
What Mzizi is not
The charter is explicit about its non-goals, and they are worth reading before anyone assumes a larger claim:- Not a rendering engine. Phase 1 is interop with Dioxus’s existing renderer. Dioxus already spent years on “one codebase, four platforms”; rebuilding it buys the syntax/compiler layer nothing.
- Not a tensor runtime. ML inference is a declared capability backed by Candle, not a competing framework.
- Not a post-quantum cryptography project. Named as a future research thread and explicitly out of scope.
- Not a web-framework target. The compiled artifact is WASM (and native, for desktop). Astro is one thin, optional distribution surface among several — not a platform Mzizi targets.