I am building Àmì as a local-first orthography engine for Hausa, Igbo, and Nigerian Pidgin. I offer diacritic assistance, whole-word suggestions, fuzzy matching, a small source-checked Hausa morphology family, and descriptive Pidgin phrase suggestions.
I keep ami as the ASCII repository slug. I write the product name as Àmì everywhere else.
I have a responsive writing screen, a Rust engine compiled to WebAssembly, Unicode grapheme-aware matching, Julia word counts, and CUE-validated language tables. I keep the code and model data local to the browser tab. I do not replace text until I explicitly accept a suggestion.
I can type how far in the Pidgin demo and see how far?. I can type the three-word test I dey fi and see I dey fine.. I keep both as optional phrase suggestions, not corrections.
I use one pinned Wikipedia article per language and a small, manually reviewed entry slice. My 965-token, 469-row count table is not representative of Hausa, Igbo, or Nigerian Pidgin. I make no trained-language-model or complete-dictionary claim.
I include one source-checked Hausa pair, gida and gidaje, as an optional morphology neighbor. I have not built full morphology for any language. I keep Pidgin descriptive, show only a few demo phrases, and never mark Pidgin text wrong or underline it as an error.
I have not added open parallel text, native-speaker evaluation, browser performance measurements, a public GitHub release, or a v0.1.0 tag. I leave each item open instead of implying completion.
I pin Node 24.18.0, Rust 1.99.0, Julia 1.13.1, and CUE 0.17.0. I start the demo with one command:
node server.mjsI let the server build the WebAssembly file when the artifact is missing or stale. I do not use npm packages or a runtime service. I keep my local toolchain cache outside the repository.
I run my pinned Rust, Julia, CUE, WebAssembly, and Node tests with one command:
scripts/test-all.shI verify the pinned source checksums and Julia output with node scripts/verify-corpus.mjs. I need network access to verify my pinned sources; I do not need an external service to use the running demo.
- My design note
- My architecture notes
- My corpus notes
- My data notice
- My pinned corpus manifest
- My launch kit drafts
I license my original code and project documentation under Apache-2.0. I keep corpus-derived data under the source-specific terms and attribution in my data notice.