addis_ai_sdk 0.2.0

Production-ready Flutter and Dart SDK for Addis AI chat, voice, speech-to-text, and translation APIs.

Addis AI Flutter SDK

Community-maintained Flutter and Dart SDK aligned with the official Addis AI JavaScript and Python SDKs (addisai 0.2.0).

It supports OpenAI-style chat, streaming, personas, function calling, durable voice clips, the voice catalog, speech-to-text, translation, legacy TTS, and the realtime WebSocket API.

Install

dependencies:
  addis_ai_sdk: ^0.2.0

Create API keys at Addis AI Platform. Never commit a key or ship a privileged key in a public client. For production consumer apps, call Addis AI through your backend or use a suitably restricted token.

import 'package:addis_ai_sdk/addis_ai_sdk.dart';

final addis = AddisAI(apiKey: const String.fromEnvironment('ADDIS_API_KEY'));

The client accepts apiBaseUrl, timeout, maxRetries, defaultHeaders, a custom http.Client, and default query parameters. It sends API keys through x-api-key; Supabase-style JWTs are sent through Authorization: Bearer. Transient failures are retried with jittered exponential backoff.

Chat

final completion = await addis.chat.completions.create(
  ChatCompletionCreateParams(
    messages: const [
      ChatCompletionMessage(
        role: ChatRole.user,
        content: 'ሰላም! አዲስ አበባን ግለጽልኝ።',
      ),
    ],
    language: Language.am,
    system: 'Answer concisely.',
    persona: 'A friendly Ethiopian travel guide',
    topP: 0.9,
    topK: 40,
  ),
);

print(completion.message.content);

Streaming uses standards-compliant SSE parsing:

await for (final chunk in addis.chat.completions.createStream(
  const ChatCompletionCreateParams(
    messages: [
      ChatCompletionMessage(role: ChatRole.user, content: 'Write a greeting'),
    ],
  ),
)) {
  print(chunk.content);
}

Use addis.chat.runTools(...) with RunnableTool for automatic local function-call execution. Attachments and voice commands use AddisFile.

Uploaded files are returned as UploadedAttachment values. Put them in later messages with ChatMessagePart.file(...) to continue a multimodal conversation without uploading the same image or document again.

Voice

The current voice API creates durable clips and uses idempotency keys so a retry is not billed twice.

final voices = await addis.voices.list(language: Language.am);
final clip = await addis.voice.generate(
  VoiceGenerateParams(
    voiceId: voices.first.id,
    text: 'ሰላም ዓለም',
    language: Language.am,
    outputFormat: OutputFormat.mp3_44100,
  ),
);

print(clip.audioUrl);
final bytes = await addis.voice.clips.download(clip.id);

Also available:

• addis.voice.estimate(...) and addis.voice.usage()
• addis.voice.clips.list/get/download/delete
• addis.voices.preview(...)
• addis.textToSpeechV2.convert(...) for an ElevenLabs-style migration API

The old textToSpeech, streamTextToSpeech, generateChat, and realtime methods remain available as a migration bridge.

Speech and translation

final transcription = await addis.speech.transcribe(
  audio: AddisFile(
    bytes: wavBytes,
    filename: 'speech.wav',
    contentType: 'audio/wav',
  ),
  language: SttLanguage.am,
);

final translation = await addis.translate.create(
  text: transcription.text,
  from: TranslateLanguage.am,
  to: TranslateLanguage.en,
);

Errors and lifecycle

All failures extend AddisAIException. HTTP errors are mapped to specific types including AuthenticationException, InsufficientCreditsException, PermissionDeniedException, IdempotencyConflictException, GenerationInProgressException, ValidationException, RateLimitException, and ServerException. API errors expose the status, code, details, request ID, and response headers.

Call addis.close() when the client is no longer needed.

For realtime sessions, await session.ready before calling sendAudio. Audio frames include the documented audio/pcm;rate=16000 MIME envelope.

Development

flutter pub get
dart format --set-exit-if-changed lib test example/lib
dart analyze lib test
flutter test
dart pub publish --dry-run

This project is MIT licensed and is not the official SDK published by Addis AI.