About the Client
A solo plumber based in western Sydney, 12 years in trade. Leads come from word of mouth + a couple of property managers on retainer. He runs 15–20 jobs a week, each 1–3 hours on-site. His 'office' is his ute, his 'business system' is his head.
Their Challenge
- Quotes from memory: prices for tap swaps or copper pipe runs all live in 'what I charged the last guy', so new quotes mean digging through chat history
- Photo flood: thousands of site photos buried in his phone — finding 'that leak from last month' is basically impossible
- 1.5 hrs of nightly admin: handwriting daily logs, sorting customer messages, planning the next day's run
- Patchy coordination with casuals: when he calls in a mate to help, there's no clean way to share 'where, when, what' for the day
Why ManifoldX
He couldn't use Procore-style software, didn't want a $50+/mo SaaS like Tradify or ServiceM8, and refused to put customer data in any cloud. He needed something running in his own home, operated through Telegram.
The Solution
1. Self-hosted Mac mini — the home for everything
A $700 secondhand Mac mini at his place, running a Python process + Obsidian vault. All data stays local — nothing goes to any cloud.
2. Telegram Bot — single interface
His customer chats, helper crews, and personal threads are all on Telegram. The bot joins as just another member, understands mixed Chinese-English speech:
- Customer: 'how much for swapping 2 mixers' → bot pulls a $XXX draft quote from his Obsidian price book, asks him to confirm before sending
- Him: '8am Eastwood tomorrow' → bot writes to Google Calendar, shows other nearby jobs
- He drops a site photo: bot identifies location / job type / linked customer and files to the right client's 'site photos' folder
3. Obsidian knowledge base — the quoting brain
Markdown files organized by trade / part / price band — the bot reads these rules to draft quotes. Obsidian works as both database (human-readable) and AI 'knowledge' input. He edits prices by opening Obsidian and changing Markdown — no code touch needed.
Tech stack
Working with us
Week 1: spent 2 mornings on the road with him to see how he talks to customers, logs jobs, prices work. Week 2: bot v1 with text chat + quote drafting. Week 3: photo recognition + Google Calendar, deployed to his Mac mini. He didn't have to learn anything new — operation = sending Telegram messages.
I used to spend an hour at home each night sorting out the day. Now it's 5 minutes from the cab of my ute. Best part — the bot's quotes match what I'd calculate myself, because it learned my own rules. — Client (paraphrased)
Impact
- Quote drafting from 5 min to 30 sec (95% faster); new-customer reply speed jumped, conversion improved
- 1.5h nightly admin → zero: everything handled by the bot during the day, no more after-hours catch-up
- Daily job cap from 8 to 20: he didn't get faster — admin overhead stopped capping him
- No more missed follow-ups: bot auto-sends 7-day / 30-day check-ins, repeat business up
What's next
On a $200/month retainer covering issue fixes and small features (latest: voice-to-quote — he speaks one line in the ute and the bot drafts a price). This 'solo tradie edition' is being abstracted into a Tradies template, targeting electricians, carpenters, landscapers — anyone running solo on the tools.