This is a modern chat UI built with Next.js App Router and the AI SDK. It streams model responses with reasoning and sources, supports web search, and lets users switch models.

• Streaming responses with the AI SDK streamText
• Reasoning and sources panes (toggleable in the UI)
• Model picker (e.g. openai/gpt-4o, deepseek/deepseek-r1)
• Web search toggle that routes to perplexity/sonar
• Dark/light theme with system preference

• Next.js 15 (App Router)
• React 19
• AI SDK 5 (ai, @ai-sdk/react)
• Radix UI + small UI primitives in components/ui

• Node.js 18.17+ (Node 20+ recommended)
• npm, yarn, pnpm, or bun

pnpm install
# or: npm install / yarn / bun install

Create a local env file and set the keys you plan to use.

cp .env.example .env.local

You can run via an AI Gateway (recommended) or connect directly to providers.

Set these variables:

AI_GATEWAY_URL=    # e.g. your AI Gateway base URL
AI_GATEWAY_API_KEY=

Notes:

• The app uses model IDs like openai/gpt-4o, deepseek/deepseek-r1, and perplexity/sonar.
• Ensure your gateway is configured to route these model IDs to the corresponding providers.

Set the keys for the models you intend to use:

OPENAI_API_KEY=
DEEPSEEK_API_KEY=
PERPLEXITY_API_KEY=
# Optional advanced overrides if needed:
# OPENAI_BASE_URL=
# DEEPSEEK_API_BASE=
# PERPLEXITY_BASE_URL=

Security tips:

• Do not prefix server-only secrets with NEXT_PUBLIC_.
• Keep real secrets out of git. Use .env.local for local dev; it’s gitignored.
• If any secret was committed previously (e.g. in .env), rotate it.

pnpm dev
# or: npm run dev / yarn dev / bun dev

Open http://localhost:3000.

pnpm dev     # start dev server (Turbopack)
pnpm build   # production build
pnpm start   # start production server
pnpm lint    # run ESLint

• Client: components/ai-chat.tsx uses useChat from @ai-sdk/react. It lets users select a model and toggle web search. Submissions are sent to the API with { model, webSearch }.
• Server: app/api/chat/route.ts calls streamText and streams tokens back. When web search is on, it uses the perplexity/sonar model; otherwise it uses the selected model. The response includes reasoning and sources, which the UI renders.

You can add more models by editing the models array in components/ai-chat.tsx.

• Deploy on Vercel or any Node host. Ensure the same env vars are set in the hosting environment.

• 401 or 403 errors: check that your AI_GATEWAY_API_KEY or provider API keys are set and valid.
• Model not found: ensure the gateway or provider supports the model ID you selected.
• No streaming: verify your host supports streaming responses and that the route isn’t buffered by a proxy.

MIT (or your preferred license)