The work you are sick of doing? We build software that does it.
Custom software, automations, pipelines and AI features that remove bottlenecks in small companies. Proven on your real work first, then handed over explained, documented and trained. Never a black box. You own everything. AI models only where they help.
Approvals, handoffs and copy-paste relays turned into flows that run themselves and tell you when they need a human.
AI assistants on your docs
Answers from your own manuals, standards and contracts, with the page cited so each answer can be checked against the source.
System integrations
CRM, spreadsheets, accounting and email finally talking to each other, so nobody re-types a number twice.
Websites that do a job
Quote, book, capture, triage. A site that finishes a task instead of pointing at a phone number.
Dashboards and reporting
The numbers you argue about every Monday, pulled automatically, on one page, with the source of each one.
Documents to data
Scans, PDFs and paper forms turned into clean records, checked by a person only where the model is unsure.
How an engagement runs
Four steps, no retainer
Step 1 · 60 seconds
Tell us the chore
One paragraph about the task you are sick of. An AI model writes a first read of it, and we reply in writing with whether it is worth automating.
Step 2 · one afternoon
Find the real bottleneck
We follow one piece of work end to end and write down every place it waits. Usually the fix is not where you expected.
Step 3 · one to three weeks
Prove it on your work
A working version runs on your actual files and cases. You see it hold before you commit to the full build.
Step 4 · handover
Own it
Code, docs, training, credentials. We explain what every part does and stay reachable, but you do not need us to keep it running.
What we have built
Real systems. No invented clients, no invented numbers.
Publishing pipeline In daily use
A set of agents that gathers material, drafts, checks claims against sources and queues each piece for review before anything is published. We designed it and operate it every day. What works: the pipeline runs unattended and the review step catches what it should. Limits: a human still reads every piece before it goes out, and the gathering stage is only as good as its source list.
Our role
Designed, built and operate it
Editorial review portal
A single review queue for people who were previously checking six inboxes, with approve, return and comment in one place and a record of who decided what. Built to sit on top of the publishing pipeline. What works: one place to look, and the decision log. Limits: it is an internal tool shaped around one workflow, not a product.
Our role
Designed and built it
Trading research dashboard Streaming live
A research dashboard that streams market data and shows signals, positions and notes on one page so a decision can be checked against its reasoning later. What works: the live stream and the one-page view. Limits: it is a research tool for our own use. It is not advice, and we make no claims about returns.
Our role
Built and run it
Document brain over industrial standards
An assistant that answers questions over thousands of pages of industrial standards and cites the page it drew from, so an engineer can check the answer rather than trust it. What works: finding the right clause fast. Limits: the citation is the product; the answer still needs an engineer to read the cited page, and the corpus is only as current as its last import.
Our role
Designed and built it
Browser writing assistant
A browser tool that edits text in place against a set of personal style notes, rather than rewriting from scratch. What works: it holds a voice better than a general chat tool because the notes are explicit. Limits: it is tuned to one writer's rules, and it suggests rather than decides.
Our role
Built it for our own writing
This site's own lead intake
When you send the form below, your note is stored, one AI model writes a two-sentence first read of it (the likely root problem and a first automation to try), and a summary is posted to our Slack. What works: we see every lead within minutes, with the first read attached. Limits: the AI model writes the first read and never replies. A person reads each note and replies.
Our role
Built and run it
On this page
The form below sends your note to us
Three workflows, illustrated
Concept films. Nothing here is a recording of working software.
These three short films show the kind of workflow the studio builds, as a picture of the idea. They were generated with AI video tools from written descriptions. They are not recordings of delivered systems, and no client is shown. Press play to load one; nothing downloads until you do.
Forecast and reorderAI-generated illustration of a proposed workflow. Not a recording of working software.
The idea shown: sales history feeds a simple forecast, and when stock is predicted to run short the system drafts a purchase order for a person to approve.
MP4, about 13 MB, loads only when played
Paper to inventoryAI-generated illustration of a proposed workflow. Not a recording of working software.
The idea shown: a delivery note is photographed on arrival, the lines are read into structured records, and the stock count updates with a person checking only the lines the software was unsure about.
MP4, about 12 MB, loads only when played
Website to ordersAI-generated illustration of a proposed workflow. Not a recording of working software.
The idea shown: a small business website takes an order or a booking and posts it straight into the back office, so nobody copies it out of an email.
MP4, about 18 MB, loads only when played
Start the 60-second intake
Find your bottleneck.
Describe the work you are sick of doing, in your own words. No call to book. You get a plain written reply about whether it is worth automating and roughly what it would take.