Actually useful AI

It’s 11pm on a Friday. You just wrapped a sixteen-hour shoot day. You’re eating cold pizza in your car. Somewhere in the back of your mind is the knowledge that if you don’t invoice this week, you won’t get paid for another month.

You do not want to open an app, find the client, create a document, enter line items, calculate overtime, remember the weekend loading, add kit rental, check the GST, and hit send. You want to eat your pizza and go to sleep.

So instead you open Claude, drop in the callsheet PDF, and say: “Invoice this. It was a fourteen-hour day, standard rate, Saturday loading. I also rented them the 633 and two channels of wireless.”

Autobooks’ MCP server picks it up. It finds the client from the callsheet (or creates them if they’re new). It knows your rate card. It calculates the day rate with the weekend multiplier, adds the overtime for the four hours past ten, adds the kit rental as separate line items, applies GST, and drafts the invoice.

You look at it. The numbers are right. You hit send. You finish your pizza.

How this actually works

Autobooks ships an MCP server — Model Context Protocol, an open standard that lets AI clients talk to your data through a structured interface.

The MCP server reads your local .autobooks file. It doesn’t send your data anywhere. It knows your rate card, your clients, your jobs, your document history. When you tell it to draft an invoice, it’s working from your actual records — not guessing.

It can also answer questions about your data:

Every answer traces back to a specific query against your file. “How much am I owed?” comes from your actual outstanding invoices, not a model’s best guess. “Which client pays slowest?” comes from your actual payment dates.

You can ask things you’d never build a report for. “What’s my average day rate on feature films vs TVCs?” “Am I making more per day than this time last year?” “If Fox Post pays that outstanding invoice this week, what’s my cash position?” Real questions, real answers, from your data.

The conversation, not the dashboard

The important thing isn’t the technology. It’s that invoicing stops being a task you dread and becomes a conversation you have while doing something else. You talk about your day in natural language — “I did a half-day for Indigo, no kit, just the base rate” — and the invoice appears.

Quotes work the same way. “Quote Fox Post for a five-day TVC mix, standard rates, two channels of wireless and the 633 for the duration.” It drafts the quote, you review it, done. You can negotiate in the conversation — “actually make it four days, and drop the kit rental to a weekly rate” — and it revises.

The AI doesn’t make decisions for you. It drafts things for your approval based on rules you’ve already set up. Your rate card, your clients, your terms. It’s doing the data entry you’d otherwise do yourself at 11pm when you can barely keep your eyes open.

Why it’s local

Your MCP server is a local process reading a local file. Your financial data doesn’t leave your machine. Nobody’s training a model on your invoices. Nobody’s aggregating your rates with other freelancers. The conversation between you and your data stays between you and your data.

Because it’s SQLite underneath, you can also connect any other tool that speaks SQL. The MCP server is the conversational interface, but the data is always yours, always inspectable, always portable.

The kind of AI we believe in

We’re not interested in dashboards that tell you your revenue is up 12% — you can read a number. We’re not interested in “smart suggestions” trained on other people’s businesses. We’re interested in AI that does the tedious, error-prone work you’d otherwise do tired, so you can focus on the work that actually matters.

You’re a sound editor, or a DIT, or a colourist. You’re good at your craft. You shouldn’t have to also be good at invoicing at 11pm on a Friday.