Introduction
Addis AI / Developer platform
Voice-first AI infrastructure,
for African languages.
Addis AI gives developers a unified infrastructure layer for text generation, speech recognition, natural voices, translation, multimodal reasoning, and realtime voice, built to process African languages as first-class inputs. Our native tokenization architecture and language-specific models preserve morphology, Ge’ez-script logic, cultural context, code-switching, and local accents across text and speech, so teams can build directly for the languages their users actually speak.
Why Addis AI?
Building AI for Africa requires language-specific training. Many general-purpose models have limited exposure to the morphology, scripts, and cultural context of African languages. Addis AI treats those languages as the core product.
Why language-specific infrastructure matters
- Morphology: Amharic verbs fuse subject, object, and tense into single words. This complex structure requires language-specific training. Our models parse it correctly, while major general-purpose AI models can misinterpret or hallucinate the underlying meaning.
- Culture: Our models understand African history, regional geography, and contemporary legal and regulatory contexts.
- Voice: We don't just do text. We built the most natural sounding TTS engines for Amharic and Afaan Oromo.
The Core Engines
Choose the capability your product needs, then follow the corresponding guide to start building.
Addis-፩-አሌፍ (NLP)
Our flagship Large Language Model. It is optimized for instruction following, reasoning, and content generation.
- Best for: Chatbots, RAG (Document Search), and Summarization.
- Languages: Amharic and Afaan Oromo are fully supported, and the model also understands English instructions.
- Context Window: 128k tokens.
አሌፍ-Audio (Voice)
A suite of high-fidelity audio models.
- Addis Voices 2New: The latest generation of Addis AI's Text-to-Speech technology, with multiple production-ready Amharic and Afaan Oromo voices designed for natural, expressive delivery. Use the catalog, previews, cost estimates, and durable clips to build conversational AI, narration, customer support, announcements, and other voice-first applications. Explore the complete Text-to-Speech guide.
አሌፍ-Audio-AM: Amharic Text-to-Speech with natural intonation.አሌፍ-Audio-OM: Afaan Oromo Text-to-Speech.addis-whisper: Speech-to-Text transcription that handles dialect variations.
The previous Voice 1 Text-to-Speech API remains available only for existing applications and is documented in the Legacy Text-to-Speech guide.
Realtime Audio
Powered by አሌፍ-1.2-realtime-audio.
- Latency: <300ms response time.
- Use Case: Live conversational agents (like Siri/Alexa) that need to interrupt and respond instantly.
Documentation Roadmap
We have structured the docs to get you into production fast.
🚀 Get Started
- Playground Guide: Learn how to prompt the models visually.
- Quick Start: Get your API Key, choose a model, and make your first request (Auth & Code included).
⚡ Capabilities
Deep dives into the specific endpoints and parameters.
- Text Generation: Build chat, summarization, RAG, structured-output, and function-calling applications.
- Text-to-Speech: Discover, preview, estimate, and generate completed audio clips with Addis Voices 2.
- Speech-to-Text: Transcribe Amharic and Afaan Oromo audio into text.
- Translation: Translate between Amharic, Afaan Oromo, and English in every direction.
- Multimodal: Reason over images, audio recordings, and documents through the chat API.
- Realtime API: Build low-latency, interruption-capable voice conversations through WebSockets.
🧩 Integration
Patterns for connecting Addis AI securely to your application.
- Web Applications: Keep API keys on your backend and expose only the routes your web client needs.
- Mobile Applications: Use a backend proxy for mobile requests and audio delivery.
- Server-side Integration: Structure production Node.js and Python services.
- Voice Interface: Orchestrate speech, text generation, and voice playback.
⚙️ Platform
- Rate Limits: Understanding tiers and quotas.
- Errors: Troubleshooting status codes like
429and500.
Community & Support
We are building this together.