Private AI and Automation
Automation you own and AI that knows your business.
Most small businesses already pay for automation. A monthly glue tool here, a form service there, a chatbot that knows nothing about your products. I replace that with workers you own, running on your own server or mine, plus AI assistants grounded in your real product rules, documents and live data. Nothing goes out to a public model unless you decide it should, and nothing stops working because a subscription lapsed.
Workflows that run themselves
Automation workers instead of subscriptions charged per task
Leads from an ad campaign filed into the sheet the sales team already uses, by language and campaign. Orders pushed into the shipping system. Nightly reports written and sent. Quotes turned into work orders without retyping. I build these as small, owned workers using whatever fits the job, from workflow tooling to code I write, always with retries, logging and an alert if one stops.
For a window and door retailer this replaced a stack of Zapier tasks with workers that run on their own server, so the monthly bill went away and the integrations stopped breaking when a third party changed a field. The workers are documented, tested and yours.
For a window and door retailer this replaced a stack of Zapier tasks with workers that run on their own server, so the monthly bill went away and the integrations stopped breaking when a third party changed a field. The workers are documented, tested and yours.
AI that knows your products and rules
A private knowledge base the model consults
A general model guesses. A grounded one looks things up first. I build a private knowledge base from your product catalogue, pricing rules, codes, manuals and past jobs, and connect it to the assistant so every answer is backed by your own material. For a window and door quoting platform, the assistant then knows which codes apply, which combinations are not allowed and what a complete order looks like, because it reads the same rules your staff do.
The same assistant can be connected read only to your database and other systems, so it stays current on stock, orders and jobs without anyone feeding it. Read only means the assistant can answer, draft and flag, but it cannot change your records.
The same assistant can be connected read only to your database and other systems, so it stays current on stock, orders and jobs without anyone feeding it. Read only means the assistant can answer, draft and flag, but it cannot change your records.
Documents in, answers out
Document analysis and voice entry
AI now reads invoices, inspection sheets, supplier price lists, contracts and email threads well. The useful part is what happens next. I build pipelines that pull the fields you care about out of documents, check them against your rules and file the result where it belongs, with a human review step wherever money or safety is involved.
Voice is the other door. For a quoting platform, staff describe the job out loud and the assistant fills the form, behind a private gateway that filters personal data and blocks prompt injection before anything reaches a model. Fewer clicks, fewer mistakes and a record of what was said.
Voice is the other door. For a quoting platform, staff describe the job out loud and the assistant fills the form, behind a private gateway that filters personal data and blocks prompt injection before anything reaches a model. Fewer clicks, fewer mistakes and a record of what was said.
Private by default
Your data stays where you decide
Where it matters, models run on hardware you or I control, in Canada, with no outside provider in the path. Where a hosted model is the right tool, it sits behind a gateway with separate keys for each app, fixed prompts, personal data filtering and logging, so you know exactly what left the building and why. I run language models on my own hardware for my own monitoring and daily briefings, so this is the setup I live with every day.
Each build comes with a short note on what the AI can see, what it can do and what it cannot, written so you can hand it to a customer or an auditor.
Each build comes with a short note on what the AI can see, what it can do and what it cannot, written so you can hand it to a customer or an auditor.
Where to start
Automation audit
A look at the subscriptions, spreadsheets and copying and pasting your business runs on, then a written list of what to automate first, what it would cost and what it would save.
Owned automation worker
One workflow built as an owned worker with retries, logging and alerts, documented and handed over. Hosted and watched by me if you want it.
Private AI pilot
A knowledge base from your own material plus an assistant your staff can try for a month, with a written review of what it got right and wrong before you commit further.
Document pipeline
Documents in, structured data out, checked against your rules with a review step. Priced on the document types involved.
Every engagement starts with a free written scope. Larger builds run through the custom apps process and pricing.
Jobs are quoted in writing after a free scope, so you know the number before any work starts.
What you own at the end
- Automation you own, with no bill per task and no vendor that can switch it off
- An assistant that answers from your own rules and data
- Read only connections to live systems, so answers stay current and records stay safe
- A gateway and logging around any hosted model, so you know what left the building
- A short note on what the AI can and cannot do
AI and automation questions
- No. Models hosted on your own server never send anything out. Where a hosted provider is used, it is one with a no training contract, behind a gateway that filters personal data first, and you are told which provider and why.
Running in production today
We called Viscous Bits because ransomware was in the news and we wanted to know how exposed we were. Henry checked everything, fixed our email, put the website behind proper protection and set up backups that cannot be wiped. Then he came to the office and sorted out the network and the servers too. Once that was done he sat with us, wrote down how a lead moves through our business and built our systems around it. The quoting tool is the big one. Our sales team opens it every morning, enters windows by voice or a short code, and the customer signs on a drawing right there at their kitchen table. Quotes that used to take an afternoon are done in a sitting and the mistakes are gone. When anything comes up, the printer, a suspicious email, a new product to add, we call the same number and it gets handled.
- Optima Windows & DoorsFrom ransomware protection to a quoting platformIt began as security consulting to protect a window and door business from ransomware. It became a digital transformation, with a custom quoting tool the sales team opens every morning.Read the case study
- Sellmore Windows & DoorsNo lead left behindYears of looking after a lead generation business: their server, every landing site, the pipeline behind the forms and the day paid leads were being thrown away without anyone knowing.Read the case study
Tell me what you keep doing by hand
Describe the task, the tools involved and roughly how often it happens. You get a written scope back, free.