Six Rust → WebAssembly demos. Zero hand-written JavaScript. Every <canvas> pixel comes from compiled Rust via presentar.

make serve              # → http://127.0.0.1:3000

Demo 6 ships the real trained aprender-shell-base.apr model — 3-gram Markov, ~380 commands, ~9 KB zstd-compressed — byte-for-byte identical to what interactive.paiml.com/shell-ml serves (SHA-256 068ac67a89693d2773adc4b850aca5dbb65102653dd27239c960b42e5a7e3974).

• Model file: wasm-demos/assets/aprender-shell-base.apr — embedded into the wasm binary via include_bytes!.
• Loader: wasm-demos/src/shell_model.rs — ShellAutocomplete::load_from_bytes parses the 32-byte APR header, decompresses the payload with pure-Rust ruzstd, bincode-deserializes the n-gram tables, and builds a prefix trie. Vendored from presentar::browser::shell_autocomplete because the published aprender-present-lib 0.34 has a broken include_bytes! path in an unrelated module that fails wasm builds (the same workaround m5-dash documents).
• Integration: wasm-demos/src/logic.rs — shell::lookup(prefix) returns the top-3 completions on every keystroke, painted to the canvas by shell_demo.rs.

git clone https://github.com/paiml/wasm-from-zero
cd wasm-from-zero
make install                                  # cargo install aprender-contracts-cli
rustup target add wasm32-unknown-unknown
make serve                                    # browser gallery
make demo                                     # all 9 demos on native
make ci                                       # fmt + clippy + test + 95% coverage + pv lint

• m1-canvas/, m1-app/, m2-elm-wasm/, m2-events/, m3-components/, m3-charts/, m4-bundle/, m4-tests/, m5-dash/ — the nine module crates the course builds in order.
• wasm-demos/ — the six-demo browser gallery. Paint logic in src/logic.rs, mount wrappers in src/<name>_demo.rs, HTML harness in web/<name>/, 60 native probar tests in tests/.
• contracts/ — three YAML contracts (wasm-rendering-v1, wasm-lifecycle-v1, wasm-panels-v1) gated by pv.
• lean/ — Lean 4 theorems discharging the universal claim at L5.

Dual-licensed MIT OR Apache-2.0.