← Back to projects

Serum

A lightweight CRM and integrated sales playbook I built for the team at Tito to stay on top of our pipeline.

Interactive demo with made-up data. Open in full screen

What it does

Two cards from prospect pages: one won after Lead, Talking, Trying it and Proposal sent, and one switched after Talking and Trying it

Built-in playbooks for our two types of prospect

The sales funnel looks different for a brand new lead versus an existing customer switching from an older version of Tito. So the playbook steps change depending on where the prospect came from.

The Lead playbook, where Reply and offer a call shows Ticked when Vicky emailed them, and the chase waits on their reply

Updates itself based on team activity

People who submit an enquiry via the Tito website are added to Serum automatically, and it connects to our email, calendar and third-party support tool, so it knows when we've emailed a prospect and when they've replied. It ticks off the playbook step for us, so nobody has to remember to.

Serum's morning email: 2 things to do today, one of them overdue, with a button to open To do

A consolidated daily task list

Any outstanding playbook steps due today show in one list alongside to-dos the team adds manually. Serum sends an optional email at 7am to each teammate with their day's tasks, on days there's something due.

A conversation where Claude moves Harbour Lights Festival to Talking and adds a note in Serum, then offers to run /prospect-demo

Designed to work with Claude

We're already using Claude throughout the day, so now we can ask it to create or update a prospect, or even use skills we've set up, like /prospect-demo, and capture the details in Serum.

Report cards for Won, Switched, Not a fit and Lost

Only the metrics that matter to us

What did we win, are they paying yet, where do our best customers come from, and why did we lose the ones we lost?

How I built it

Serum is one Go binary and one SQLite database running on CARLOS, a system for building better web apps with AI agents. I built it with Claude Code in under a week, in a language I'd never written before.

I wrote up the whole story, including the decisions I made (and unmade) along the way, on my blog.

Read the full story

Go SQLite CARLOS Claude Code MCP