IndabaAI

A guided AI marketing assistant for South African small businesses. An MVP for the Meta AI Academy South Africa Pitchathon, built on Meta's Llama 3.3 70B Instruct via OpenRouter.

How it's wired

• content/prompts.md — the one file that makes the app sound like IndabaAI: the system prompt, interview questions, and output templates. Edit this file on GitHub to change what the app says. No code changes needed.
• app/chat.tsx — the chat screen and flow (welcome, interview, profile, offer, ads, image prompt, results).
• app/api/chat/route.ts — the only place the app talks to the AI. It sends the content file plus the user's answers to OpenRouter.
• lib/content.ts — reads the interview questions and model name out of content/prompts.md.

Deploy (GitHub + Vercel)

1. Upload all of these files to a GitHub repository named indabaai.
2. On Vercel: Add New > Project > Import indabaai.
3. Before clicking Deploy, under Environment Variables add:
   • Name: OPENROUTER_API_KEY
   • Value: your OpenRouter API key (starts with sk-or-v1-)
4. Click Deploy. Done.

If the AI parts don't work after deploy, the key was probably added after the build: Settings > Environment Variables, add it, then Deployments > (three dots) > Redeploy.

Changing the wording later

Edit content/prompts.md directly on GitHub (pencil icon > change > Commit changes). Vercel rebuilds automatically in about a minute.

Model

meta-llama/llama-3.3-70b-instruct via OpenRouter (https://openrouter.ai/api/v1). The model name is read from the MODEL AND HOSTING section of content/prompts.md.