Skip to content

AI & Workflow Automation

AI tools and automations that take repeated admin off your team: reminders, follow-ups, reports, data entry and document drafting, with a person signing off.

  • A fixed price for the build
  • AI and provider costs shown separately
  • A person signs off what matters
An enquiry comes in

[01] When it fits

Sounds familiar?

  • 01

    Someone copies the same details between systems every day.

  • 02

    Leads wait hours for a reply because nobody saw them.

  • 03

    A report takes a morning to put together, every week.

What we do about it

If your team does the same task by hand every week, software can probably do most of it. We automate the repetition and keep people in charge of the decisions.

[02] What we build

What we build.

Plain rules where they are enough, AI where reading or writing saves real time. Each one runs on a schedule or a trigger, with a log.
  1. Form sent

    Website

    Rules

    New client?

    Done

    Task + email

    09:41:02 form · Lerato, new client

    09:41:02 task created for Sipho

    09:41:03 welcome email sent

    01

    Workflows that start themselves

    A form, an email or a new record kicks off the next steps, without anyone pressing go.

  2. Scheduled

    Today
    Visit reminder to Thandi08:00 · WhatsAppSent
    Deposit still due · Naidoo10:30 · EmailSent
    Feedback request · 3 guests16:00 · SMSQueued
    02

    Reminders and follow-ups

    The right message to the right person at the right time, on WhatsApp, email or SMS.

  3. Every Thursday, 09:00

    Your week: 58 bookings, 6 more than last week

    Written from your booking data. Friday is filling fastest.

    03

    Scheduled reports

    Your numbers pulled together and written up in plain language, every week or month.

  4. Roomwise: The AI draft
    04

    AI that reads and drafts

    AI that reads documents and photos, sorts what comes in and drafts the reply or the report.

    Screen from Roomwise
  5. Casepath: Every detail, next to its page
    05

    Review queues and alerts

    Anything unusual goes to a person to check, with a record of what ran and what failed.

    Screen from Casepath

[03] How it works

How it works.

We map how the task really runs today, agree a fixed price, and test on real examples before anything runs on its own.
  1. Start

    Step 1: A call

    Pick one task your team repeats every week. We ask where it comes in, who touches it and what goes wrong.

  2. Before any work

    Step 2: Brief and fixed price

    We choose plain rules, an existing tool, an AI step or a mix. Fixed price for the build; AI and provider costs estimated separately.

  3. Day 0

    Step 3: Agreement and deposit

    You sign and pay the deposit. We set it up in accounts you own.

  4. About a week later

    Step 4: First working run

    The automation running on test data, with the log so you can see every step it took.

  5. While we build

    Step 5: Tune it

    Send the odd cases by voice note or screenshot. We adjust the rules and prompts, often the same day.

  6. On your date

    Step 6: Switched on

    AI agents test it on real examples, including the strange ones. You choose the day it starts running.

Afterwards

Then, your choice.

  • A

    Handover

    It runs in your accounts. You get the docs and the log.

  • B

    Monthly plan

    We watch the runs, fix failures and adjust it as your process changes.

  • C

    Product partnership

    For a bigger system: we keep automating the next task with you each month.

[04] Seen in the work

Real projects. Real screens.

AI we run in our own products every day: inspection reports, marketing and voice typing.

All our work

Tools we work with

The tools behind it.

Every tool here is in a product we have built and shipped. We pick them once the problem is clear, and set them up in accounts you own.

See the full toolbench
  • OpenAIText & imagesGPT models for drafting, pulling data out of documents and generating images, in Moteef and client tools.
  • ClaudeTextAnthropic's model for careful reasoning and writing. Our Ask x Create assistant is built on it.
  • GeminiVision & imagesGoogle's model for reading photos and PDFs, and for on-brand images in Moteef.
  • ElevenLabsVoiceSpeech-to-text that copes with Afrikaans and South African accents, plus voiceovers.
  • DeepSeekTextWrites scheduled owner briefs from a day's sales, through Azure AI Foundry.
  • WhisperOn-deviceOpenAI's speech model, running offline inside Seshat on Mac and Windows.
  • ConvexReal-time backends for our own products and client platforms: data, jobs and sign-in in one place.
  • PostmarkReceipts, sign-in codes and reports that land in the inbox, not spam.

[05] The proof

Why work with us.

We run AI in our own products every day. Roomwise drafts property inspection reports from photos, Moteef plans and creates marketing in a brand's own voice, and Seshat turns speech into text on the Mac itself. AI agents also test and report on our builds. So we know where AI saves hours and where plain rules do the job better.

AI models and providers in work we have shippedFrom OpenAI and Claude to models that run on your own computer
12
From the first line of code to a live pilotRoomwise, our AI inspection reports, from our own repository
5 days
Languages Seshat types on the Mac itselfFrom English and Afrikaans to isiZulu and Sepedi
29

Our own counts come from a dated snapshot of our repositories and release feeds (11 October 2026).

Fair questions.

Next step

Start with a call.

Tell us what you need. You get a short brief, then a written proposal with a fixed price before any work starts.

Start a conversation

With one task your team repeats every week and hates. Something you can see and count: copying details between systems, chasing documents, building the same report. We map how it really works before we pick a tool.

No. Plain rules are cheaper, faster and easier to test, and they handle most admin. We use AI models from OpenAI, Anthropic and Google Gemini where reading, sorting, summarising or drafting text saves real time and a person can check the result.

Depending on the setup: hosting, email or message sending, automation tools, AI model usage and storage. We estimate these upfront and say what would push them up, such as volume.

You hear about it. Good automation shows errors, keeps the original record, retries safely and hands odd cases to a person. Nothing should fail quietly.