The task

Build a chat agent we can talk to. It has one tool: web search.

Make the best chat agent with web search you can: as good as the best ones on the internet today.

A few examples from the dataset we'll test your agent on:

• "On Monday we shortlisted 5 CRM tools for a 20-person sales team. Has any of them changed pricing or shipped something big since?"
• "What did I ask you on 3 November? Is your answer still correct today?"
• "Last Tuesday I asked you about Postgres hosting. Has any pricing changed since then?"

How we judge

The best architecture wins. Not the first submission, not the prettiest screen, not the most features.

We're hiring for how you think: how you design an agent that answers fast, gets things right and is good to talk to. We'll chat with your agent like a real user, read your code and read your DESIGN.md.

Rules

• JavaScript or TypeScript on Node.
• It must run with npm install, then npm run dev.
• List every API key it needs in .env.example.
• Use any framework (or none) and any model.
• AI coding tools are welcome. You must be able to explain every decision in your code.

What to send

1. A public GitHub repo.
2. DESIGN.md in the repo, one page: how it works, why you built it this way, what you considered and rejected, and what you'd do with another week.

Submit your repo link through the application form.