AI systems, built around your business10+ years, 8 products shipped

Grow the business, not the workload.

We find the bottlenecks slowing you down, then build AI agents and AI workflows that take repetitive work off your team and connect the tools you already use.

Agents inside your business systems

Each agent reads the systems you already run, does one job and logs every step. A person approves anything that moves money or reaches a customer.

See AI inside your systems

Models and clouds we build with

  • Anthropic
  • OpenAI
  • Gemini
  • xAI Grok
  • LangChain
  • Hugging Face
  • AWS
  • GCP
  • Azure

AI services for running businesses. One team, chip to cloud.

Process automation with AI

We map a process, then automate the reading, matching and re-typing.

Invoice flow
3 of 4 on
  • Read supplier invoicesPDF and email
  • Match to purchase ordersPrice and quantity
  • Post matched invoicesInto Sage
  • Pay without approvalA person signs off

Embeds engineers who build agents

Forward Deployed Engineers learn your business and build the tools it needs.

First two weeks
2 of 5
  • Sit with the finance team
  • Sit with operations
  • Map the quoting process
  • Rank bottlenecks with managers
  • Ship the first tool

Builds AI into devices

Boards, firmware and edge models, linked to the cloud that runs the fleet.

  • Node 14 needs waterSoil moisture under 18% in zone BOpen
  • P3Pump 3 scheduled05:40 for 25 minutes
  • Firmware 2.4 rolled out48 of 48 nodes updated

From chip to cloud. Engineered for impact.

  • Circuit boardsDesigned for the field
  • FirmwareC, C++ and Rust
  • Edge AIModels on the device
  • CloudAWS, GCP, Azure, Cloudflare
  • AgentsInside your systems

How an engagement runs in 3 steps

1

We map your processes for automation

Engineers sit with your teams and read the documents, emails and systems behind each process.

2

We rank automation by hours saved

We time each step and rank the ones worth automating with your managers.

Invoice matching, per weekWeek 1
41 hacross 3 people
3

We build, test and run the agents

Every agent ships with an evaluation set, and we keep running it after launch.

agents/invoiceseval
~/agents/invoices$bin/eval --all
4 of 4 sets
po-matching120 cases41s
vat-lines80 cases22s
duplicates60 cases18s
hand-offs40 cases12s
Cases300 passed(300)
Ready to shipin 1m 33s
~/agents/invoices$

AI hardware and IoT devices

We design the circuit board, the firmware and the cloud together, then put the model where it works best: on the device or in the cloud.

  • Schematics and PCB layout in KiCad
  • Firmware in C, C++ and Rust
  • Wi-Fi, BLE, LoRaWAN and CAN bus

See Shaduf, our own AI farm sensors

A panel of small round green circuit boards, each with a white connector, ready to be separated

Questions. Anything else, ask us on the call.

Do the agents replace our systems?

No. We build on the ERP, CRM and apps you run today and connect AI to them.

Who checks what the agents do?

Your people. Anything that moves money or reaches a customer waits for approval, and every step is logged.

How long until automation is live?

Discovery takes one to two weeks. A first automation usually reaches production four to twelve weeks after that.

Can our developers run the agents?

Yes. Our Forward Deployed Engineers work inside your team and hand over everything they build, with documentation.

Do you build the devices too?

Yes. We design circuit boards and firmware, and connect devices to the cloud and to your agents.

Do you work with businesses abroad?

Yes. We are based in Somerset West, Cape Town, and work remotely with teams in other countries.

Products we have shipped

8 products in fintech, property, health, safety, retail and automotive. Client names stay private.

See all the work

Tools in these products
4 shown
  • ReactAutomated testing platformIn use
  • PythonMachine learning for test resultsIn use
  • FlutterCAN bus configuration toolIn use
  • C++Vehicle module core, over FFIIn use

AI products for founders, from idea to launch

Starting something new? We cut version one to what proves the idea and build its AI on real data.

  • A clickable prototype in weeks
  • An MVP with app, backend and AI

See how we build AI products

MVP scope, draft 2
1. What version one does
Farmers photograph a leaf and get a likely disease and a treatment on WhatsApp within a minute.
The first 50 users are invited by hand. Payments in the app are out of scope.
Every answer the model gives is logged and reviewed by an agronomist each week.