Skip to content

AI tools your team will actually use.

We build AI assistants, automations and AI features for your own apps. Each one solves one specific problem, works on your real data, keeps a record of what it does, and sends anything important to a person first. We use whichever AI model fits the job, so you are never locked in to one provider.

Book an AI planning sessionHow we build it

Fixed price · quoted within 24 hours

  • Anthropic Claude
  • OpenAI
  • Google Gemini
  • Open models on your own servers
  • AWS · Azure · Google Cloud
A network of connected points, lit up
AI, your systems and your people

01 · What we build

Four kinds of AI tool.

01

AI assistants for everyday work

Assistants that sort support requests, read and file documents, check invoices, or answer staff questions. They work inside WhatsApp, email, Slack or your back-office system. Anything important goes to a person first.

  • Support requests
  • Paperwork
  • Staff helpdesk
02

Automation between your systems

Connect the tools you already use, like your CRM, accounting software, payments and spreadsheets, so information moves on its own. You decide which steps need someone's approval.

  • System connections
  • Approvals
  • Alerts
03

Search across your own documents

Ask a question in plain language and get an answer from your policies, contracts, manuals and old tickets, with a link to the source. It respects who is allowed to see what. Your documents are never used to train anyone's AI.

  • Sources shown
  • Permissions respected
  • Tested for accuracy
04

AI features in your own product

Smart search, summaries, recommendations or a built-in assistant inside the app you already sell to your customers. Built by the same team that builds your web and mobile apps.

  • Web
  • Mobile
  • Fast responses

02 · What is inside

Every AI tool we build has the same six parts.

The AI model is only one of them. The other five are what make it safe enough for a bank or a clinic to rely on.

  1. 1What comes inEmails, WhatsApp messages, forms, PDFs, voice notes.
  2. 2Your informationYour documents and records, with permissions respected.
  3. 3The AI modelChosen for the job. Its instructions and tests are saved and versioned like code.
  4. 4ActionsThe things it is allowed to do in your systems, with limits.
  5. 5A person approvesYou decide what needs a human. One tap to approve or correct.
  6. 6A recordEvery step is logged, like the example on the home page. You can always see what it did and why.

03 · Where it helps most

Start with the job your team complains about.

The best first AI tool is rarely the flashiest one. It is the repetitive, paperwork-heavy job that eats hours every week. These are the industries we have built for since 2017.

Banking and financeChecking ID documents, sorting disputes, finding payments that do not matchSaves: hours per case
HealthcarePatient intake, summarising notes, appointment reminders on WhatsAppSaves: waiting time
LogisticsReading delivery notes and invoices, flagging problems, telling customers where their order isSaves: days to get paid
EducationAnswering enquiries, supporting students, helping teachers mark (with a teacher checking)Saves: enquiry to enrolment
Retail and online shopsOrder questions, returns, writing product descriptions, supplier emailsSaves: cost per ticket
InsuranceTaking in claims and sorting them, answering brokers' policy questions, flagging suspicious claims for reviewSaves: time per claim

04 · How we build it

Step by step. You can stop after any phase.

Each phase ends with something you keep: a plan, a working test version, a finished system. The timings below are typical. We confirm yours at the end of phase one.

  1. Phase 1about 1 week
    Understand the problemWe map how the work happens today, what data you actually have, what the rules are, and what success looks like in numbers. You get a written plan with a fixed price for the test version.
  2. Phase 22 to 4 weeks
    Build a test versionA working version your team uses on real cases. We measure how often it gets things right. You get a clear yes or no, backed by numbers, not a demo.
  3. Phase 34 to 8 weeks
    Make it safe and connect itAccess controls, a record of every action, what happens when the AI is unsure, cost limits, a POPIA review, and connections to your real systems. You get a system in daily use.
  4. Phase 4ongoing
    Keep it runningWe watch it, update the AI model when better ones come out, re-run the tests, and review what your team corrected each month, so it keeps getting better.

05 · The guarantees

The questions your legal and IT people will ask. Answered.

  • POPIA handled from the startThe AI only sees what it needs. We set rules for how long data is kept and keep a record of what is processed.
  • Your data stays yoursWe use AI providers that do not keep your data, or run the model on your own servers. Nothing is used to train anyone's AI.
  • Not tied to one AI providerClaude, GPT, Gemini or open models, chosen on results and cost. Easy to switch later.
  • You own everythingThe code, the AI instructions, the tests and the accounts are all yours, with documentation.
  • People decide the important thingsYou set the rules for what needs sign-off. The AI drafts, a person approves.
  • Tested before every changeEach tool has a set of test cases from your real work. If accuracy drops, we do not release the change.

06 · Common questions

Straight answers.

How much does it cost to build an AI tool in South Africa?

It depends mostly on how many of your systems it needs to connect to and how serious the decisions it makes are, not on the AI model itself. That is why we start with a fixed-price AI planning session: you leave with a shortlist and a first estimate before committing to anything.

Which AI do you use?

Whichever gives the best results on your test cases at a reasonable cost. That could be Anthropic's Claude, OpenAI, Google's Gemini, or an open model running on your own servers when your data cannot leave your building. We build it so the model can be swapped later.

Is our data safe? Is this POPIA compliant?

We design for POPIA from the first meeting. The AI only sees what it needs, we agree how long data is kept, we keep a record of what is processed, and we use providers that do not store your data (or host the model ourselves). Your legal team reviews the design before it touches real data.

How long before something is live?

A test version your team can use usually lands within a few weeks of the first phase. A finished system, connected to everything, takes a few weeks more, depending on how many systems it talks to. Exact dates are agreed in phase one, and you can stop after any phase.

Do we need to replace our existing systems?

No. AI tools work best as a layer on top of what you already use: your accounting system, CRM, WhatsApp Business, email and spreadsheets. We connect to them. We only suggest replacing a system when the system itself is the problem.

Find out where AI would help your business, in 90 minutes.

The AI planning session: a ranked list of where AI would save you time or money, whether to build or buy, and a first cost estimate. Fixed price, quoted within 24 hours.

Book the sessionhello@liorra.io