We use browser storage, strictly for the Functional category, only if you opt in — to let our AI Pilot chat widget remember your conversation. Nothing else runs until you choose. See our Cookie Policy.
Waves AI Pilot
Behind the build
Building Waves AI Pilot: an AI sales employee you can actually trust
Waves AI Pilot started from a frustration with AI chatbots: they're confident and wrong. For a business, a bot that invents a price or promises a refund isn't a feature — it's a liability. So the goal from day one was a different kind of system: an AI employee that only answers from what a business has approved, knows when to get out of the way, and never acts on its own. Here's how it's built.
01
Grounded, never guessing
Every business's knowledge — products, services, FAQs, uploaded documents — is broken into chunks, embedded, and retrieved per question, so the AI answers from real, approved information instead of its own imagination. If it doesn't know, it says so and routes to a human rather than making something up.
02
Strict tenant isolation
This is multi-tenant software, and one business's data must never leak into another's. Isolation is enforced at the database layer with row-level security, not just in application code — so a business's knowledge, conversations, and leads are provably its own. There is no shared knowledge base between businesses.
03
AI as a signal, humans in control
Lead scoring is deliberately split in two: a deterministic layer computes priority from real stored facts (lead score, pending appointments, time since last activity), and the AI only ever writes a brief on top — grounded in that prospect's actual conversation, forbidden from inventing a budget or a promise. The AI never sends a message or takes an action by itself; a person always triggers the real step.
04
An automation engine that won't double-fire
Follow-ups, tags, tasks, and notifications run through a workflow engine built to be idempotent and crash-safe — a retried job resumes exactly where it left off and never executes the same action twice. Boring reliability, which is exactly what you want running unattended against real customers.
05
Omnichannel, one brain
The same AI employee works across a website widget, WhatsApp (Meta Cloud API), and Instagram DMs, with every conversation landing in one inbox and clear attribution between AI, prospect, and staff.
The stack
Next.js on Vercel, Supabase (Postgres with row-level security) for data and tenant isolation, Clerk for authentication and role-based access, Claude and Gemini for language and embeddings, and Resend for email. Every AI feature respects a per-business usage quota, and sensitive actions are recorded in an audit log.
Where it is now
Waves AI Pilot is feature-complete and in private testing, being hardened against real-world use before launch. Early access is opening soon.
In private testing · Launching soon
Request early access