Systems we builtLive
Crypto Watcher
A 24/7 AI investment desk for my Kraken portfolio — not a trading bot. A team of specialists that know my strategy and report to Gertrude.
Imagine hiring a small desk of analysts who never sleep, never panic-click, and actually remember that you're a long-term holder — not a degen chasing green candles. That's Crypto Watcher. A custom desktop system built to watch my Kraken investments, run a roster of specialist AI agents around the clock, and turn market noise into plain-English advice aligned with **my** goals. It is deliberately **not** wired into my exchange account. It does not auto-trade. It is nothing like the usual bot that YOLOs your stack because a line crossed a line. And when I want the full picture, I ask Gertrude — she reads every agent's reports and tells me what matters.
Yeah — it actually does that
This is Liam's live investment monitor — built with AI tools, running on his hardware, tuned to his risk appetite. Not a fintech product. Not financial advice on a plate. A system you can replicate for crypto, stocks, or anything you refuse to babysit manually.
A desk that never clocks off
Crypto Watcher runs a team of specialist AI agents 24/7 — each with a narrow job: macro context, sector moves, portfolio drift, headline risk, whatever Liam configured. They don't get tired, don't refresh Twitter at 2am, and don't forget the plan because a chart went red for an hour.
It knows the strategy — not just the price
These agents aren't generic "number go up" bots. They're prompted around Liam's long-term goals, time horizon, and rules — hold discipline, what counts as noise vs signal, what would actually make him change course. Recommendations arrive in plain English: *here's what moved, here's why it might matter to you, here's what I'd watch next* — not a wall of tickers.
Not a trading bot — on purpose
Crypto Watcher is **not** linked to Kraken by API choice. No auto-executes. No hidden keys trading while you sleep. Liam has seen what retail trading bots do to people's nerve and net worth. This system **watches, analyses, and advises** — you stay the one who clicks buy or sell. That separation is a feature, not a missing feature.
Specialists, not one confused chatbot
One model trying to be analyst, risk officer, and therapist at once is how you get slop. Crypto Watcher spawns focused agents — each with a role, boundaries, and output format — and collects their work as files and reports on disk. Looks like a boutique research team. Runs on a PC in Sheffield.
Gertrude is the boss
The killer integration: Liam doesn't live inside twelve dashboards. He talks to Gertrude. She can pull up Crypto Watcher anytime — read the latest agent reports, cross-check against his strategy, and answer *"should I care about this?"* in one conversation. One personality. One interface. A whole monitoring stack behind it.
The bit that should make you pause
A system this capable around money is serious kit. Wrong prompts, wrong guardrails, or treating AI output as gospel could cost you. We don't teach pump-and-dump nonsense or automated recklessness. We teach you to build **your** monitor, **your** rules, **your** human final say. Educational purposes only. Not financial advice — a pattern for adults who still want to think for themselves.
Right — here's the trick
No Bloomberg terminal. No hedge-fund budget. A tattoo artist wired a desktop app, a Python data lane, and a folder full of agent reports — then plugged Gertrude in as the front desk.
Desktop shell + data lane
Tauri wraps a React UI you actually want to open. A Python sidecar handles feeds, storage, scheduling, and the boring plumbing. Same pattern as other Liam desktop builds — native feel, scriptable guts.
Agents are scheduled specialists
Each "analyst" is a loop: pull context (market data, portfolio notes, news), run through a focused system prompt, write a timestamped report to disk. 24/7 means cron and always-on hardware — not magic, just persistence.
No exchange API — by design
Data in, analysis out. Liam feeds context manually or through read-only sources he controls. Deliberately no write access to Kraken. The system cannot trade even if a model hallucinates hard — because there's no pipe to do it.
Gertrude reads the filing cabinet
Agent outputs land as files. Gertrude's operator stack can read that directory, summarise across agents, and answer in chat. One integration point instead of a dozen tabs. Watch something → analyse it → ask your assistant when it matters — the curriculum pattern in one real project.
If you thought AI finance meant handing your keys to a bot — this is the other path. Build a monitoring desk that works **for** your strategy, plug it into an assistant you already trust, and keep the final decision human. Membership teaches the desktop shell and the agent patterns — so you can aim this at crypto, a shop's sales data, or your inbox. Full repo and follow-along coming in the Toolbox. Join the waitlist if you want to build your own Watcher, not rent someone else's black box.
Stack & tools
Built with VS Code, Cline, Cursor — plain English in, working software out.
In the curriculum
You'll walk through this pattern in Build Your Jarvis — desktop shell · agent modules — same approach, your project.
Try it
Overview only — no public demo yet. Full build ships in membership.
Member access
Full repo and customisation guide land in membership. Join the waitlist for first access when Toolbox drops.
Related systems
Playable demos
Try before you read more
Chat, tools, and infrastructure demos — interaction first.
Like what you saw? join the waitlist.
Limited initial membership opens soon — full curriculum, toolbox, and community in plain English. Join the list for first access; one email the day the doors open.