Announcements

Addis Scribe transcribes Amharic from a microphone or an uploaded file. Choose Standard or Turbo (3× speed) at the same price. Stream live audio over a WebSocket with a one-use session ticket, or upload a file of up to 25 MB and 180 seconds and receive partial transcripts while it is processed. Only the final transcript is billed, at 3.5 ETB per 1,000 characters. The SDKs add scribe.transcribe, scribe.stream, scribe.connect, and request recovery.

• Try Scribe in the playground

Scribe file uploads can return the start and end time of every word with timestamps=word, or a ready-made caption file with format=srt or format=vtt. Use them to subtitle video, highlight words during playback, or jump to the moment a phrase was said. SDK 0.5.0 adds timestamps: "word" and local SRT and VTT helpers.

Addis Voices 2 can now play speech while it is generated. Use an HTTP stream for individual requests or a persistent WebSocket for successive turns on one connection, with Amharic, Afaan Oromo, and Tigrinya voices. Streaming uses the same 5 ETB per generated minute pricing and wallet as saved clips, and every streamed turn is saved as a durable clip. Use voice.stream in the SDKs.

• Try streaming in the playground

Addis Voices 2 introduces a production voice catalog, previews, cost estimates, durable generated clips, idempotent requests, and minute-based billing at 5 ETB per generated minute. Use am-hamen for the minimal Amharic example, then query the catalog before saving a voice for production.

The official addisai packages are now the recommended way to integrate Addis AI. The documentation leads with Node.js and Python examples, while raw REST remains available for unsupported runtimes and interoperability.

• Install the official SDKs
• View the package on npm
• View the package on PyPI

Chat requests now document system instructions, personas, and structured function calls. System instructions control behavior, personas define identity, and messages preserve conversation state.

• System instructions and personas
• Function calling

The documentation has been updated in place around official SDK examples, production guidance, and complete capability best practices.