Logo Lanfrica

unmoha/African-Language-Accessibility-Toolkit

Domaine:

natural language processing

Type de record:

software
Créateur:
unm
Hôte:
Open Local AI is an early-stage Python library project for building offline-first language and accessibility tools around local AI providers # Open Local AI Open Local AI is an early-stage Python library project for building offline-first language and accessibility tools around local AI providers. It is intended for developers who need a stable application-facing interface while keeping translation engines and local models replaceable. ## Current status The repository currently contains the project foundation, canonical language metadata, and an Argos Translate provider adapter. Translation is available when a suitable Argos model is installed locally. The initial implementation scope is translation. The current provider-independent service uses adapters for verified local engines. Speech-to-text, text-to-speech, OCR, accessibility utilities, web interfaces, and remote providers are planned but are not implemented. ## Languages Open Local AI currently defines a canonical language vocabulary containing: - `am` - Amharic - `om` - Afaan Oromo - `ti` - Tigrinya - `en` - English Canonical recognition does not imply that a translation provider currently supports every language or language pair. Actual translation support will depend on verified provider and model availability. A language code alone will never imply that a provider supports a language or a language pair. ## Planned architecture Applications will use a translation service through a provider abstraction: ```text Application -> TranslationService -> TranslationProvider -> Concrete provider adapter -> Local model or engine ``` The core package will remain usable without any particular provider installed. Provider availability, supported language pairs, model installation, and network requirements will be checked explicitly. Unsupported requests will produce project-level errors rather than fallback or fabricated translations. ## Offline and privacy goals Local providers should process text without sending it to a remote service, but offline behavior will only be claimed after the provider execution path has been verified. Remote provi …