Handl is not a chatbot. It's an action agent โ a system that doesn't just answer questions but actually does things on your behalf. That means browsing the web, navigating apps, drafting outputs ready to send, and operating with full memory of who you are. The technical stack reflects that ambition from day one.
Layer 1 โ Experience
Onboarding & identity
Onboarding that configures a named agent in 10 minutes, seeded from your own rรฉsumรฉ and work samples rather than a questionnaire. Naming your agent is the first thing you do, and it shapes everything after.
React / Next.js
Framer Motion
Tailwind
Layer 2 โ Intelligence
Memory, routing & web search
System prompts assembled from your career history, what the agent has learned about your search, and what you still owe whom. Memory is seeded from documents you upload and grows from everything that happens after. It knows you better every session.
Claude API
Web Search tool
MCP
Vector memory
Layer 3 โ Action Phase 2
Browser agent & task execution
Claude's computer use API lets Handl control a real browser โ navigate Airbnb, book restaurants, fill forms, compare listings, and return results. This is where Handl stops being a smart assistant and becomes something genuinely new: an agent that acts, not just advises.
Anthropic Computer Use
Playwright
Headless Chrome
Task queue
Layer 4 โ Infrastructure
Persistence, integrations & billing
User context storage, task history, calendar and email integrations, and the attribution layer that makes every Handl output shareable. Every document the agent produces carries "Created with [Name] via Handl".
Supabase
Vercel
Stripe
Google OAuth
Cloudflare Workers