Vercel AI Chatbot

Routes chat requests through Vercel AI Gateway to multiple LLM providers (OpenAI, Anthropic, Google, etc.) with automatic provider failover and streaming token-by-token responses back to the client. Uses the Vercel AI SDK's `generateText` and `streamText` APIs which abstract provider-specific APIs into a unified interface, with streaming handled via Server-Sent Events (SSE) from the `/api/chat` route.

Implements bidirectional chat state management using the `useChat` hook from @ai-sdk/react, which maintains optimistic UI updates while streaming responses from the server. The hook automatically handles message queuing, loading states, and error recovery without manual state management, synchronizing client-side chat state with server-persisted messages via the `/api/chat` route.

Allows users to upvote/downvote AI responses via the `/api/votes` endpoint, storing feedback in the database for model improvement and quality monitoring. Votes are associated with specific messages and can be used to identify problematic responses or train reward models. The UI includes thumbs-up/down buttons on each message.

Uses shadcn/ui (Radix UI primitives + Tailwind CSS) for all UI components, providing a consistent, accessible design system with dark mode support. Components are copied into the project (not npm-installed), allowing customization without forking. Tailwind configuration enables responsive design and theme customization via CSS variables.

Enforces strict TypeScript typing from database schema (via Drizzle) through API routes to React components, catching type mismatches at compile time. Database types are automatically generated from Drizzle schema definitions, API responses are typed via Zod schemas, and React components use strict prop types. This eliminates entire classes of runtime errors.

Includes Playwright test suite for automated browser testing of chat flows, authentication, and UI interactions. Tests run in headless mode and can be executed in CI/CD pipelines. The test suite covers critical user journeys like sending messages, uploading files, and sharing conversations.

Stores all chat messages, conversations, and metadata in PostgreSQL using Drizzle ORM for type-safe queries. The data layer abstracts database operations through query functions in `lib/db` that handle message insertion, retrieval, and conversation management. Messages are persisted server-side after streaming completes, enabling chat resumption and history browsing across sessions.

Implements schema-based function calling where the AI model can invoke predefined tools (weather lookup, document creation, suggestion generation) by returning structured function calls. The `/api/chat` route defines tool schemas using Vercel AI SDK's `tool()` API, executes the tool server-side, and returns results back to the model for context-aware responses. Supports multi-turn tool use where the model can chain multiple tool calls.

Accepts file uploads (images, documents) through the `MultimodalInput` component, converts them to base64 or URLs via Vercel Blob storage, and passes them to the AI model as multimodal context. The `/api/upload` route handles file storage, while the chat component embeds file references in message history for vision-capable models to process.

Implements user authentication using NextAuth 5.0 with support for both guest (anonymous) and registered user flows. The middleware in `lib/auth.ts` protects routes, while login/registration pages handle credential-based auth. Guest users get temporary session IDs for chat history, while registered users have persistent accounts with email verification and password reset capabilities.

Allows the AI to create and update code snippets, documents, or other artifacts through tool calls, which are stored as separate entities with version history. The `Document and Artifact System` tracks changes via `document_versions` table, enabling users to view diffs and revert to previous versions. Artifacts are rendered in a side panel with syntax highlighting for code and rich editing for documents.

Implements conversation-level access control where users can mark chats as private (only visible to owner) or public (shareable via URL). The database schema includes `visibility` column on conversations, and middleware enforces access checks before rendering chat content. Public chats generate shareable URLs that bypass authentication for read-only access.

Stores in-flight streaming state in Redis to enable resumption if the connection drops mid-stream. When a user reconnects, the system retrieves the partial response from Redis and continues streaming from where it left off, rather than restarting the entire request. This is implemented via Redis keys keyed by chat ID and message ID.

Enforces per-user rate limits on API calls and gates premium features based on user entitlements (free vs paid tiers). Rate limiting is implemented via middleware that checks request counts against time windows, while entitlements are stored in the user object and checked before allowing access to premium models or tools.