Open Chat

Open-source AI-chat-bot: an open-source version of ChatGPT that runs any local or hosted LLM with web search and custom modes to tailor responses.

Features

• Hybrid AI control in one UI: Switch between local and hosted models without leaving the chat flow.
• OpenRouter = model superhub: Tap into OpenAI, Anthropic, Google, Qwen, MiniMax, and dozens more through a single key and use their free tiers to prototype fast.
• Offline-first with LM Studio: Run everything locally by loading compatible models in LM Studio and keeping keys private.
• Persona-driven customization: Drop new prompt files to craft modes tailored to any workflow or teaching style.

Prerequisites

• LM Studio: Download the installer for your OS from the official Downloads page.
• Node.js + npm: Get the latest LTS release for your OS from the Downloads page.

Setup

1. Clone: git clone https://github.com/Dom-ML/open_chat.git && cd open_chat.
2. Install dependencies: npm install.
3. Create a .env with AI_GATEWAY_API_KEY, OPENROUTER_API_KEY, and LMSTUDIO_BASE_URL=http://localhost:1234/v1.
4. Start LM Studio with the models you want available (match the names listed under each provider).
5. Run the dev server: npm run dev.

LM Studio Default Model

• Default local model: mlx-community/LFM2-2.6B-4bit. In LM Studio, search for this model, download it, then hit Serve so it runs on http://localhost:1234/v1 (see LM Studio docs for details).
• To swap models later, update the first entry in modelsByProvider.lmstudio inside src/app/api/config/route.ts so its value matches the Serve ID you expose from LM Studio.

How It Works

• Providers live in src/app/api/config/route.ts; add models by extending the modelsByProvider list for the provider you use.
• Streaming chat logic is in src/app/api/chat/route.ts, which reads your provider choice and system prompt.
• Prompt modes come from markdown files in src/prompts; copy one, edit the front matter, choose any Lucide icon name, and reload to see it.
• UI helpers for fetching provider data sit in src/lib/ai-config.ts, and prompt loading is cached in src/lib/prompts-server.ts.

Tech Stack

• Next.js app ready for Vercel hosting or local dev.
• UI layer built with shadcn’s @ai-elements components.
• AI SDK suite: ai, @ai-sdk/react, @ai-sdk/openai-compatible, @openrouter/ai-sdk-provider, plus zod for validation.

How to add custom modes

• Copy one of the markdown files in src/prompts, change the id and name, and write the persona instructions under the front matter.
• Keep the instructions clear and scannable: define the tone, outline the response structure (bullets, summaries, steps), and give 1-2 concrete dos/don’ts so the model stays on brief.
• Pick an icon by matching any entry from the Lucide icon list (https://lucide.dev/icons) or by browsing node_modules/lucide-react, then drop the plain name into the icon field.
• Restart the dev server or reload the page so the cache in prompts-server picks up the new file.

Limitations/Future Improvements

• Persistent storage
• MCP support