Kelavo: Building an AI Automation SaaS for Freelancers & Agencies

For freelancers and small agencies, client work and admin work compete for the same hours. Lead qualification, proposals, contracts, invoicing, and status updates are necessary, unbillable, and repetitive. Kelavo is the subscription SaaS I built to hand that work to AI agents, without the risk of an AI sending a client the wrong message.

The challenge

Most tools in this space are rule-based CRMs and template libraries. They store the data, but the owner still does the writing and the decisions. The client wanted software that actually does the work: it qualifies the lead, drafts the proposal, and builds the invoice.

The constraint that shaped the product was trust. A freelancer’s reputation sits in every email and contract. An AI that sends, signs, or schedules anything on its own is a liability, not a feature. Nothing could leave the system without the user’s explicit approval.

The approach: specialised agents, not one chatbot

Instead of a general-purpose assistant, I designed a multi-agent architecture where each agent owns one stage of the client lifecycle. Each agent has a narrow job, its own prompt, and a clear output, so its results are easier to review. The output of one agent becomes the input of the next: lead → proposal → project → invoice → case study.

• Lead Scout scores and ranks incoming leads by how likely they are to convert, so the owner starts with the best ones.
• Proposal drafts proposals and contracts in the user’s own writing style.
• Project Manager turns a signed contract into milestones and a schedule, and tracks what is due next.
• Finance generates PDF invoices directly from project data, with no re-typing.
• Client Comms drafts client replies and status updates and queues them for approval instead of sending them automatically.
• Content & Brand turns completed projects into case studies and portfolio pieces.

Approval-first by design

Every agent action goes into a single approval queue. The user can approve, edit, or reject each draft in one click. This human-in-the-loop design is what makes the product usable for client-facing work. The owner hands off the effort but keeps control of every client relationship. Over time, the agents learn the user’s tone and client history, so drafts need fewer edits.

The SaaS foundation

Around the agents, I built a production SaaS foundation:

• Authentication and self-serve sign-up.
• Subscription billing with a free trial and tiered monthly plans for solo freelancers, growing freelancers, and agencies.
• Dashboards and PDF generation.
• An SEO-ready marketing site with comparison pages against established tools, built to win organic traffic.

The stack is Next.js and TypeScript, with LLM APIs orchestrated through multi-agent workflows.

The outcome

Kelavo launched as a live SaaS with self-serve sign-up and tiered plans. Lead-to-invoice work that used to span several disconnected tools now runs as one connected workflow. Because of the approval-first design, users can hand off admin work and stay confident that nothing reaches a client without their say-so.

You can see Kelavo live. If you’re planning an AI product that has to be trusted with real customer-facing work, let’s talk about your build. I design AI agents that do the work while keeping humans in control.