Panda HQ

The problem

My business lived in six places. Clients and income were in spreadsheets and Airtable, plans in Notion, thinking in Obsidian and half-finished documents in Google Docs. The things that actually mattered that day were in whichever notepad file I'd opened last.

Every morning started the same way. I'd sit down at the PC and go looking through five notepad entries, a doc I might have updated and forgotten about, and a Notion board and an Airtable base that were already out of date, because keeping them current was a job of its own and nobody was doing it. It took ten to fifteen minutes of searching before I knew what the day needed, and even then I was trusting that nothing had fallen through a gap between two apps.

None of the tools were bad, they just weren't built for me. Each one wanted me to work its way, and I was bending my week around six different designs to fit them.

So I built my own.

The approach

It started small, with one place for clients, projects and income. It got a lot bigger once I asked myself what else I kept losing, and the answer was thoughts. The call I had to make, the idea from a walk, the thing someone asked me to do while I was out. So Panda HQ ended up as two things at once, the ledger of the business and a second brain that never forgets anything.

There were three big decisions.

The first was that capture has to be rough. I think a second brain only works if putting something into it is faster than deciding where it goes, so I don't file anything. I write the way I think, twenty bullet points at once if that's what the day produced, with two projects and some life admin tangled together, and the system does the filing.

The second was that the AI has to know me rather than the internet. I wrote a detailed profile of myself covering the work, the clients, the people, how I talk about them and what matters, and every capture is read against it. So "call him back about the invoice" lands on the right client, and a rough note gets a sensible due date and priority instead of a generic one.

The third was that it must never be confidently wrong. When the system's sure where something belongs, it links it. When it isn't, it still links it but flags it as Unsure, and I confirm or fix it in one place. Money gets a stricter rule. If the desktop and the phone ever disagree about a payment, it shows me the conflict and never settles it with a guess.

What I built

Panda HQ, the desktop app. The home screen answers one question, which is what I need to know right now. It shows tasks due today, so the first thing I see is what to focus on. Behind that is the whole business, with clients, projects and when they were completed, retainers, salary and freelance income, invoices, and financial reports that show the business progressing month by month. Money is tracked in pounds, dollars and lira, and it's stored as whole minor units everywhere, never as a floating-point number, because a rounding error in money is the kind of bug you don't find out about until late.

Panda OS, the phone half. An Android launcher that replaces the home screen on my work phone, so the phone runs as my own system with Android underneath. Its home screen is a three-by-three grid. The centre shows the time, today's agenda and this month's income. To the left is the workspace of tasks, notes and ideas, to the right is the money side with clients and projects, top right has the finance reports and bottom right has the settings. Below the centre is the normal app list, which is still there for everything else. This is where most capture happens, in a taxi, between meetings, or the moment someone asks me for something.

A capture brain that reads my profile. A rough note gets sorted into tasks, ideas and notes, linked to the right client or project, and given a due date, priority and effort level based on what I actually wrote. It's not magic and it doesn't invent anything. If a note doesn't say what it's about, it can't know, so anything it isn't confident about gets linked but flagged for me to check. You can see this in the demo at the top of this page, where four rough lines go in and two tasks, an idea and a note come out, with the one link it wasn't sure of flagged.

Questions, not just storage. Because everything I've captured is structured, I can ask it things in plain language. What ideas have I had in the last three months that I didn't act on? How many times did calls get rescheduled this month? It answers from my own records, read against the same profile.

Sync that cannot lose a thought. Everything I capture goes into an append-only log, and each device rebuilds its own view of the workspace from that log. The two views are proven identical, down to a checksum. Reads always come from the device itself, so nothing waits on the network. When I'm online, writes reach the other device in under a second, and when I'm not, they queue up and land a second or two after I reconnect. So a task captured on the phone is on the desktop by the time I sit down, and it stays on the phone until it's done, which means I can't forget it on the way home.

Locked to one person, and tested that way. Every record in the cloud is fenced to my account at the database level, and that fence was tested with a second real account trying to get through it. It's holding the whole of my business's finances, so making sure only the app shows my data isn't enough. I need to know no one else can reach it.

The front door is this website. When someone sends a brief from the contact page, it lands in Panda HQ as an enquiry, and one click turns it into a client and a project with a follow-up task already set. So the form on this site is part of the system this page describes.

The outcome

What used to take ten to fifteen minutes at the PC now takes about thirty seconds. I open the home screen, see what's due and start. There's no morning search any more, because everything's in one place.

It replaced all six of them (spreadsheets, Airtable, Notion, Obsidian, Google Docs and the notepad files). Every client, project, payment and captured thought is in one system, on two devices, in sync.

To be honest, it's still young. At the time of writing it's been in daily use for about a month, I capture at least one set of tasks every day, and I'm still building it. Next on the list are voice capture from the phone and a Panda assistant to replace the stock one on it, so I can ask the same questions I ask at the desk from anywhere and get answers with the context of my own work rather than the internet's.

The way I usually put it is that I have an app I operate my entire business from. I designed it, I built both halves of it, and it works the way I think, because it was only ever built for me.