Runtime.MiRust.com

Runtime.MiRust.com is the project website for TinyRustLM: a custom Rust/WASM browser-local small language model runtime, static browser UI, local model format, adapter route, and proof pipeline.

Runtime.MiRust.com should not read like a placeholder CMS blog. It should explain the actual project: a zero-dependency Rust and browser runtime that loads local .slm model files, validates them, runs inference in WebAssembly, displays model provenance, and records the evidence chain around generated model artifacts.

Current implementation in one view

AreaCurrent source-backed status
Browser shellStatic HTML, custom CSS, handwritten JavaScript.
RuntimeNo-crate Rust compiled to WASM with explicit exported functions.
Model formatCustom .slm format with checksum, tensor directory, tokenizer metadata, and quantization metadata.
Quantizationf32, q8_0, and q4_0 fixture and TinyLM-16M smoke paths.
AdaptersADP1 raw adapters, ASP1 sparse adapters, and ALR1 low-rank adapters.
Provenance.slm.manifest, assembly receipts, adapter-family receipts, selector registries, and module-plan receipts.
TestingRust tests, browser harness checks, Node DevTools browser smokes, WASM ABI smoke, route-soak checks.
Product boundaryRuntime and evidence are real. Trained assistant quality and full GGUF import are not claimed yet.

What the name means

Runtime.MiRust.com is the public lane for the Rust/WASM local-model runtime work. GGUF remains an important roadmap-facing format boundary, but the source-backed implementation currently centers on TinyRustLM and the custom .slm route because that path is small enough to audit, validate, and prove in the browser before wider format support is added.

Why this exists

Most local AI demonstrations hide important boundaries behind frameworks, package managers, opaque model loaders, or remote inference calls. TinyRustLM takes the opposite route:

Main project lanes

  1. Browser-local runtime: local .slm files and WASM runtime, no external inference API.
  2. Proof-first model format: checksum-bearing local model files with explicit quality boundaries.
  3. Adapter ecology: raw, sparse, and low-rank delta packages validated by Rust before use.
  4. Model breeding pipeline: offline compatibility, lineage, operators, candidates, promotions, population review, parent-pool recipes, selector admission, and browser-routable bundles.
  5. Agent-readable site: semantic PHP-rendered pages, Markdown alternates, llms.txt, llms-full.txt, and no dependency-heavy front-end framework.

Audit-driven site improvements

The latest improvement round adds the missing layer that a technical project site needs before deeper UI work: deployment trust, namespace clarity, research routing, and explicit launch checks.

Start here