Skip to main content
AI Chat Interface

​
Overview

Launch includes a production-ready AI chat feature with streaming responses and chat history persistence. The architecture is modular, making it easy to extend with additional providers and custom system prompts.

​
Prerequisites

  • Backend running with AI env vars set (copy from apps/api/example.env first)
  • EXPO_PUBLIC_API_URL set in apps/mobile/.env (from apps/mobile/example.env)
This feature is gated by the mobile feature registry flag: ai. See Feature Registry.

​
Key Features

Provider Abstraction

OpenAI is enabled by default. Anthropic support is wired in the backend and can be enabled in the mobile model list with a single config change.

Streaming Responses

Real-time streaming for a responsive chat experience with stop generation support.

System Prompts

Predefined personas and custom system prompts to shape AI behavior.

Chat Persistence

Conversation history stored in PostgreSQL via tRPC procedures.

​
Architecture

The AI feature is built with extensibility in mind, separating concerns into distinct layers.

​
Core Components

AI Provider Abstraction

Location: lib/ai/providers/ A unified interface for working with different AI providers. OpenAI is enabled by default; add Anthropic to the providers list to expose Claude models in the UI.

Custom Hooks

Location: features/chat/hooks/ Business logic extracted into reusable hooks: useChat for messaging, useChatHistory for history management, and useChatPersistence for database sync.

Chat Context

Location: features/chat/context/ Global state management for chat functionality, reducing prop drilling and simplifying component interactions.

​
Steps

  1. Copy the example env (if not already):
  1. Add API keys to your backend .env:
  1. Ensure the AI feature flag is enabled:
apps/mobile/features/feature-registry.tsx → featureFlags.ai = true
  1. (Optional) Enable Anthropic models in the mobile picker:
apps/mobile/lib/ai/providers/index.ts → add anthropicProvider to the providers array.
  1. Open the chat screen at /ai-chat.

​
How It Works

  • Mobile streams tokens from the API via POST /api/ai/stream (SSE)
  • Chat history is persisted via tRPC procedures under chat.*
  • System prompts are defined in apps/mobile/lib/ai/prompts/index.ts

​
Key Files

  • API streaming: apps/api/src/routes/ai-stream.ts
  • AI providers: apps/api/src/services/openai.ts and apps/api/src/services/anthropic.ts
  • Chat persistence: apps/api/src/routers/chat.ts
  • Mobile streaming client: apps/mobile/lib/api/streaming.ts
  • Chat UI and hooks: apps/mobile/features/chat/

​
Customizing System Prompts

System prompts define the AI’s persona and behavior:
Conversations are persisted in Prisma under the Chat and ChatMessage models in apps/api/prisma/schema.prisma.

​
API Endpoints

The chat feature exposes the following tRPC procedures:

​
Test Checklist

  • Chat screen loads and streams responses
  • New chats are saved and appear in history
  • Switching models updates the active provider

​
Troubleshooting

If streaming fails, verify API keys and server logs. For general issues, start with Troubleshooting.

​
Remove / Disable

To disable AI while you configure providers, set: apps/mobile/features/feature-registry.tsx → featureFlags.ai = false For production removal guidance, see Removing Features.

​
Next Steps