aidea

Integrates OpenAI, Anthropic, and Chinese LLM providers (Tongyi Qianwen, Wenxin Yiyan) through a provider-agnostic abstraction layer that normalizes API schemas and handles authentication tokens. Uses BLoC pattern for state management to decouple chat logic from UI, enabling seamless provider switching within conversations without losing context or message history.

Maintains chat room state with full message history, user/assistant role tracking, and context window optimization using local SQLite storage. The BLoC pattern manages conversation state transitions (loading, success, error) while the APIServer handles pagination and lazy-loading of historical messages to prevent memory bloat on mobile devices.

Centralizes all external API communication through a single APIServer component that abstracts provider-specific details (authentication, request/response formats, error handling). Each provider (OpenAI, Anthropic, Aliyun, Baidu) has a dedicated adapter that translates between the provider's API schema and AIdea's internal message format, enabling seamless provider switching and fallback logic without touching business logic layers.

Streams API responses token-by-token from providers supporting streaming (OpenAI, Anthropic, Stable Diffusion) and renders them progressively in the UI using Dart streams and Flutter's StreamBuilder widget. The chat interface updates in real-time as tokens arrive, creating a typewriter effect that improves perceived responsiveness compared to waiting for full response completion.

Detects network connectivity using the connectivity plugin and allows users to compose messages while offline, storing them in a local queue (SQLite) with 'pending' status. When connectivity is restored, the app automatically retries sending queued messages in order, updating message status from 'pending' to 'sent' or 'failed' based on API response.

Queries each AI provider's API to detect supported capabilities (vision, function calling, streaming, image generation) and gates UI features accordingly. For example, if a model doesn't support vision, the image upload button is hidden; if it doesn't support streaming, responses are fetched as complete blocks. Capability metadata is cached locally to avoid repeated API calls.

Enables users to send a single prompt to multiple AI models in parallel and display responses side-by-side, coordinating concurrent API calls through async/await patterns in Dart. The UI layer renders responses as they arrive using StreamBuilder widgets, allowing partial responses to display before all models complete, while the BLoC layer manages request/response lifecycle and error handling per model.

Captures audio input from device microphone, sends it to a speech-to-text provider (integrated via APIServer abstraction), and converts transcribed text into chat messages. Uses platform-specific audio recording APIs (iOS AVAudioEngine, Android AudioRecord) wrapped in Flutter plugins, with automatic audio format normalization (WAV/MP3) before transmission to ensure provider compatibility.

Integrates image generation models (DALL-E, Stable Diffusion, and Chinese alternatives) through the provider abstraction layer, accepting text prompts and returning generated images. The Creative Island feature manages image generation workflows, storing generated images locally with metadata (prompt, model, parameters) and enabling batch generation of multiple variations through sequential or parallel API calls.

Provides image editing capabilities (inpainting, outpainting, style transfer) integrated with image generation models through the Creative Island module. Users select regions of existing images or upload new ones, specify editing instructions, and the system sends the image + mask + prompt to providers supporting image-to-image operations (Stable Diffusion, DALL-E variants).

Implements a credit ledger system where users purchase credits that are consumed by API calls (different models/operations cost different amounts). The payment system integrates with platform-specific payment processors (Apple In-App Purchase, Google Play Billing) and tracks usage per user through a backend API, with local caching of credit balance to enable offline awareness of remaining quota.

Uses Flutter's built-in localization framework (intl package) combined with custom implementations to support 10+ languages including English, Chinese (Simplified/Traditional), Japanese, and others. Language selection is persisted locally and applied dynamically without app restart using BLoC state management to trigger UI rebuilds across all screens when language changes.

Implements the Business Logic Component (BLoC) pattern across the application to decouple UI from business logic, using Dart streams and events to manage state transitions. Each feature (chat, image generation, settings) has its own BLoC that listens to events, processes them through business logic, and emits states that the UI consumes through StreamBuilder or BlocBuilder widgets.

Leverages Flutter's cross-platform capabilities to compile a single codebase to iOS, Android, macOS, and Windows with platform-specific configurations for signing, entitlements, and app store submission. Uses Xcode for iOS/macOS builds and Android Studio/Gradle for Android, with CI/CD integration for automated builds and deployment to app stores.