dabit3/react-native-ai

1. This repo is a three-part monorepo: an Expo/React Native mobile app in app/, a TypeScript backend in server/, and a small Node CLI in cli/. The architecture is intentionally split so the mobile client stays UI-focused while the server owns API keys, provider integration, and file handling.

2. The mobile app is the primary user-facing surface. It is an Expo-based React Native project with native iOS/Android shells, shared TypeScript source, and bundled assets/fonts. Entry points are app/index.js and app/App.tsx, with the app rendering from src/main.tsx.

3. The app UI is organized around a few screens in app/src/screens: chat.tsx for text/assistant chat, images.tsx for image generation, assistant.tsx for assistant-style flows, and settings.tsx for runtime configuration. Shared UI pieces live in app/src/components, including model-specific icons and a model picker modal.

4. App-wide state and behavior are centralized in app/src/context.tsx, app/src/theme.ts, app/constants.ts, and app/src/utils.ts. These files govern selected model, theme selection, and mapping between user choices and backend routes, which is important for the architecture because the app does not hardcode provider logic into the screens themselves.

5. The mobile app is designed to support multiple LLM providers rather than a single backend. The README and component set show explicit support for OpenAI, Anthropic, Gemini, Cohere, and Mistral-branded UI affordances, with streaming/real-time response handling as a core interaction pattern.

6. The backend in server/src is the provider proxy and orchestration layer. It is split by capability into chat/, images/, and files/, with routers such as chatRouter.ts, imagesRouter.ts, and fileRouter.ts acting as the external API boundary consumed by the mobile app.

7. Chat is the most structured backend subsystem. The server/src/chat folder contains provider adapters (gpt.ts, claude.ts, cohere.ts, gemini.ts) plus orchestration helpers like createAssistant.ts, addMessageToThread.ts, getThreadMessages.ts, runResponse.ts, and runStatus.ts. This suggests a thread/run model for assistant-style conversations with streaming status updates back to the app.

8. Image generation is handled separately from chat, with server/src/images/gemini.ts implementing Gemini-based image generation and imagesRouter.ts exposing it. The README indicates this path also supports natural-language-to-image workflows and image input handling from the app’s image screen.

9. File uploads are a distinct backend boundary. server/src/files/upload-file.ts and helpers/saveFileToOpenai.ts indicate the server accepts uploads, stores them locally under server/uploads/, and can forward them to OpenAI-compatible storage when needed. This keeps file credentials and transport logic off the device.

10. The backend is environment-driven. Both app/.env.example and server/.env.example show that runtime configuration is externalized, with the server specifically requiring provider keys such as GEMINI_API_KEY. The repo’s architecture assumes the server is the trusted layer for secrets and provider access, while the app talks to it over HTTP.

11. The CLI in cli/ is a project bootstrap/configuration tool rather than an app runtime. The README shows it is invoked with npx rn-ai to generate a new project and set up environment configuration, making it part of the developer experience and not the production runtime path.

12. Native platform files under app/ios and app/android are mostly generated shells around the React Native runtime. They matter architecturally only insofar as they provide the deployment targets for the shared app code and host the app icons, splash screens, and platform manifests.