Llama Coder

Llama Coder is an open-source project by Nutlope (Hassan El Mghari) that generates small apps from one prompt. The repository describes it as an open source Claude Artifacts, originally built on Llama 3.1 405B and served through Together AI. It is released under the MIT License, and a hosted version is available at llamacoder.io.

What It Is

Llama Coder is a prompt-to-app tool. A user describes an idea, optionally attaches a screenshot, and chooses a model from a dropdown on the hosted site. The app is generated and shown in a live preview. The homepage shows example prompts such as an expense tracker, team chat and beat maker.

How It Works

The tech stack documented in the README is:

• Llama 3.1 405B as the LLM, with Together AI handling inference.
• esbuild-wasm and esm.sh powering an in-browser preview renderer that runs in a sandboxed iframe.
• A Next.js app router frontend styled with Tailwind.
• Braintrust for optional observability and Plausible for site analytics.

Running It Yourself

To self-host, clone the repo and create a .env file. It needs a Together AI API key and a PostgreSQL connection string (Neon is suggested) used via Prisma. A Braintrust key is optional. Screenshot uploads need S3 credentials. Then run npm install and npm run dev. A contributing guide is included in the repository.

The homepage states that it is powered by Together AI and used by 1.1M+ users; this is a vendor claim.