A local AI chat application powered by Ollama, built with Next.js 16, the Vercel AI SDK, SQLite, Drizzle ORM, and shadcn/ui.

Requires Ollama running locally. Pull any model with ollama pull <model> before starting.

Chat	Arena	Voice

• Chat: multi-turn conversations with markdown and syntax-highlighted code rendering, image attachments for vision models, persistent chat history, and per-chat model selection
• Arena: send the same prompt to two models side-by-side and compare responses in real time
• Voice: hands-free voice call interface with low-latency TTS streaming
• Screen share: capture the active screen when speaking to vision-capable models in voice mode
• Dark mode: system preference detection via next-themes
• Model capability detection: vision and thinking support are auto-detected per Ollama model
• SQLite persistence: chat history and the selected model are stored in data/app.db through Drizzle ORM

• Ollama installed and running (ollama serve)
• At least one model pulled, e.g. ollama pull qwen3.5 or ollama run qwen3.5
• Node.js 20+ installed

git clone https://github.com/aleksa-codes/next-ollama-chat
cd next-ollama-chat
bun install
bun run dev

Open http://localhost:3000.

The SQLite database is created automatically at data/app.db on first run.

If you change the schema (add/remove fields or tables), generate and apply a new Drizzle migration:

bun run db:generate
bun run db:migrate

Note: The server runs bun run db:migrate automatically on startup, so you typically only need to run it manually when working on migrations.

• Next.js 16: App Router
• Vercel AI SDK: streaming chat with useChat
• ai-sdk-ollama: Ollama provider and streaming helpers
• SQLite + Drizzle ORM: local persistence for chats and preferences
• Tailwind CSS v4: utility-first styling
• shadcn/ui: component library
• React 19 with React Compiler
• marked + shiki: markdown and code rendering
• Web Speech API: STT and TTS for voice mode

MIT, see LICENSE for details.