What we build
From software that records to software that acts.
Four kinds of work, one team. We design it, build it, put it in production and keep it running — with whichever models fit the job.
- Models
- Claude, GPT, Gemini and open-weight models — chosen per job
- Fits into
- Your ERP, CRM, inbox, WhatsApp, payment gateway and spreadsheets
- Engagement
- A scoped pilot, then one monthly retainer
- Kickoff
- Typically one to two weeks from the first call
01 Capability
Agentic systems
Agents that own a job from start to finish.
- Agent design
- Tool integrations
- Approval flows
- Action log
- Eval suite
- Monitoring
Give it a goal and the tools it’s allowed to use. It works through the steps — reads, looks things up, decides what to do next, acts — and stops for a person wherever the stakes are real. These are the jobs that eat a team’s week without anyone noticing.
Collections that chase themselves
Finds what’s overdue, checks what’s already paid, writes reminders in the customer’s language and escalates the awkward ones.
An enquiry desk that doesn’t sleep
Reads every channel overnight, answers what it can from your own documents and hands the rest to the right person with context.
Document intake
Invoices, forms, IDs and contracts — extracted, checked against your records and filed, with the exceptions explained.
Reconciliation
Bank lines, gateway settlements and ledgers matched every day, not every month-end.
agent: fee-followup goal: collect overdue installments without upsetting families tools: - erp.ledger.read - gateway.settlements.read - whatsapp.send # needs approval approve: accounts office, before any message is sent limits: one reminder per family every five days escalate: disputes, pending concessions, third reminder evals: past cases from your records, re-run on every change log: every call, input and output
Every agent starts as a spec like this, agreed with you before we write code.
02 Capability
AI-enhanced software
Intelligence inside the software you already run.
- Retrieval over your data
- Extraction pipelines
- In-app assistants
- Integrations
- Usage analytics
Not every job needs an agent. Often the biggest win is a single smart step inside an existing screen — a draft that’s already written, a document that’s already read, a question answered from your own data. We add it to your ERP, CRM or internal tools, or to software we build for you.
Ask your data
Plain-language questions over your business data, with the query and the source shown alongside every answer.
Document understanding
Turn PDFs, photos and scans into structured records your systems can use.
Drafting in context
Replies, notices and summaries written from the record in front of you, ready to edit.
Classification & forecasting
Route, tag, prioritise and predict — so people start from the right place.
03 Capability
AI-native products
New products designed around AI from the first screen.
- Product design
- Full-stack build
- Model architecture
- Launch
For founders and teams building something new. We design the product around what models are good at and honest about what they aren’t: interfaces that show uncertainty, make undo easy and keep a person in the loop where it matters. Then we build it and take it to production.
Product & interaction design
Flows where the AI does the heavy lifting and the user stays in control.
Architecture
Models, retrieval, tools and data chosen for cost, speed and accuracy — not fashion.
Web and mobile apps
The full product, not a demo: auth, payments, admin and the boring parts that make it real.
Launch and scale
Observability, cost controls and the evals that let you ship changes without fear.
04 Capability
Run & evaluate
Agents are never finished. We keep them sharp.
- Dashboards
- Eval reports
- Incident response
- Monthly improvements
Models change every few months. Your business changes faster. Software that acts needs someone watching how it performs, catching drift before it becomes a problem and upgrading it when a better model ships. That’s the retainer — and it’s how we’ve always worked.
Monitoring
Every run, every tool call and every approval visible on one dashboard.
Evals on real cases
A growing test set from your own business, re-run on every change.
Cost per task
Know what each completed job costs, and watch it fall.
Model upgrades
Switch models when a better one ships, proven against your evals first.
How we decide
We climb only as high as the job needs.
Agents are powerful and they aren’t free. Before recommending one, we work up this ladder — and stop at the first rung that solves the problem.
- 01
Remove the step
Sometimes the work shouldn’t exist. Change the process first.
- 02
A rule or a form
Structured input beats clever parsing of messy input.
- 03
Plain automation
If the steps never change, a script is cheaper and more reliable.
- 04
One AI step
Reading, drafting or classifying inside an otherwise ordinary flow.
- 05
An agent
The job varies case by case, spans tools and needs judgement.
Engagement
How working together is priced.
A pilot to prove it, then a retainer to run it. We give you numbers after the audit, once we know the scope.
The audit
Free. Thirty minutes on how work moves through your business, and an honest view of where AI fits.
The pilot
One workflow, a fixed scope and a price agreed before we start, measured against today’s baseline.
Production
One monthly retainer covering monitoring, evals, model upgrades, support and improvements.
Not charged
Per-seat licences, change-request billing, surprise invoices.
Leaving
No lock-in. The code, prompts, evals and data are yours, with documentation to hand over.
Next step
What’s one job you’d hand to an agent tomorrow?
Bring it to a free 30-minute audit. We’ll tell you whether an agent should do it, what it would take and what it would cost — or that it shouldn’t.
30 minutes · free · no pitch deck