From a Windows widget to a voice agent

My wife had been day trading for a while. She wanted something simple: a checklist that told her when her setup matched enough of her entry criteria to take the trade — and stayed quiet when it didn't. She'd been building it in Excel for weeks. It wasn't working.

I'd just wrapped a carpentry job and had time between projects. "Let me finish it for you." I opened Gemini for the first time. A few days later, a working Rainmeter widget sat on her desktop.

Quick context: five years in Australia doing carpentry. It's my second craft — I'd stepped away from coding long before. I didn't plan on coming back. This AI wave pulled me in.

After Jill's widget I didn't stop. I built myself a NYC-session trading clock — Asia, London, and New York markers, plus economic-release alerts. Nothing dramatic. But something had shifted.

The coding world I left when I became a carpenter looked nothing like this. Then I found OpenClaw — a multi-agent assistant. The idea of agents that could actually do things hit me in the chest.

I was already trading for myself. A half-finished thought surfaced: what if I built my own trading agent? One I could talk to. Tell her to backtest a strategy. Eventually hand her my live account. Later on she got a name: Moltfi.

I started with the frontend — trader-pro, my first real prototype. A few days in I realised the obvious: without a real broker connection and historical data, there was nothing for the UI to show. The frontend had to wait. Time to go build the backend.

(trader-pro didn't go to waste. It's now the foundation of Moltfi's frontend — the backtesting page and the copy-trade management UI both live there.)

The plan was simple. The execution wasn't. To build Moltfi I needed: a broker connection (Rithmic), historical market data, and a way to actually speak to her. Voice input wasn't optional — typing is slower than talking, and I wanted a real conversation with her.

Meanwhile Jill, watching me build, decided to start her own project. Bookkeeping. Her pain. She called it Booky. Under the same roof, the two of us working in parallel — each chasing our own interest, each scratching our own itch. Some days progress was one commit in one repo. Other days it was spread across four.

With voice input breathing, a broker connected, and historical data flowing in, I could finally build the body itself. At its core, everydays.agents is Moltfi — that's why I had to build my own. None of the existing agents could remember enough about my trading context across sessions, and that's the one thing trading actually needs. So I built Moltfi around two pillars: sharp low-latency responses, and deep long-term memory. Once those were in place, the rest of the family — Meander, Booky, Mate — came along for the ride. Same backbone, different personalities.

Around the same time I realised all these services needed a shared brain for login and billing. That became everydays.tools — the portal you're reading this on.

Two more machines came online in March. A production server for when Moltfi went live. And — after three months of patched-together hardware — a proper dev workstation. Carpentry jobs pushed away by now, savings burning. Bought the rig anyway.

April was the month the pieces snapped into place.

Cross-broker copy trading went live — one trade, replicated across multiple prop-firm accounts and different brokers in real time. Market-wide there's essentially one other service that pulls this off. It wasn't in the original plan. It came from needing to scale my own trades across several prop-firm challenges without babysitting each one. Eventually this becomes a paid product; right now it's not stable enough to charge for.

Meander wired up to everydays.agents. Voice input on one side, the agents on the other — now a single pipeline: I speak, and whoever I asked answers.

Then one day I sat at my desk and said, out loud: "Run a backtest on this strategy for the past year." Moltfi heard me. Started a run. Came back with results. Three months earlier I was watching Jill stare at a spreadsheet that wouldn't do what she needed. Three months later I was talking to Moltfi while she ran my own trading backtests.

The backtest engine still needs polish — I'm stuck on some edge cases. But the shape of the thing is there. The decision now: ship Meander first, let paying users fund the next round of work. Being a maker in Australia is expensive. Right now I'm burning through savings; the carpentry jobs are paused. Meander going live is what turns the light back on.