The problem
Before we shipped our own chat agent, every inbound email from the homepage went to the founder inbox. Most messages were the same questions: how much does it cost, what do you actually do, can I see a demo, do you do dashboards too.
The volume was small but the response time was 2+ hours and answering each one was 5–10 minutes of context-switching. We were spending more time qualifying leads than building.
What we built
A Gemini-backed website chat agent ("Alex") with a system prompt grounded in our own company-context.md file. Session storage in Redis, rate-limited per IP, tracks lead intent (pricing, demo, generic question) and offers the right CTA at the right moment.
When the model is confident, it answers directly. When the user expresses buying intent, it drops a "Book a build call" CTA inline. When the question is something the AI shouldn't answer (custom contracts, edge cases), it surfaces a tiny lead form that captures email + topic and pings the founder.
The result
The bot now handles ~80% of inbound chat sessions without escalation. The founder inbox only sees genuinely high-intent leads (booked-call-ready or complex-build-scoping).
We use this same architecture to deploy chat agents for paying Reevez customers — pricing depends on traffic volume, but the baseline ($499/mo Starter + $5,000 Standard Build) gets a customer the same agent on their own site, branded, with their own context file.