AI Bookkeeper for African Businesses

Inspiration

African small businesses do not fail because they lack ambition. They fail because the work never stops, and the tools were built for offices they do not have.

A tailor in Aba, a pharmacy in Enugu, a fashion seller in Lagos; they run inventory, customers, bookkeeping, and follow-up from one phone. They do not have an operations manager. They have WhatsApp.

Most AI products assume a laptop, Slack, and a team. We kept meeting owners who had sat through an AI class, then gone back to a notebook. The gap was not knowledge. It was an operator that would stay: remember yesterday’s sales, chase the unpaid invoice, and answer the customer at 9pm.

We built Verxio so a shop that cannot hire an operations manager can still have one. An AI employee. Gemini is the brain. WhatsApp or Telegram is the office.

What it does

Verxio is an AI operator, not a chatbot. The owner messages it the way they message staff. The same Gemini-powered agent lives on WhatsApp and Telegram, continues in the web workspace, remembers the business, and does the work.

It can:

Live where the business already is: WhatsApp, Telegram, web, desktop, Slack, and Discord, with shared memory
Take real action: Record keeping, bookkeeping support, customer replies, scheduled follow-up, files, and 1,000+ connected apps
See what the owner sends: Text, voice notes, receipts, stock photos
Ship default workers: Customer support, SDR follow-up, and an AI Micro-Manager for shop-floor tasks
Keep every workspace isolated: One business never sees another’s memory, files, or keys
Get better with use: Skills, memory, and workflows the owner should not have to rebuild

Hosted workspaces run on the Gemini API. Owners never paste a model key. They text. Gemini decides and executes.

How we built it

The deployed app calls the Gemini API for hosted inference. That is the production default, not a side demo. Hosted workspaces use Gemini (gemini-flash-lite and the Gemini 2.5 / 3.x catalog).

Verxio is the product layer we built in this 90-day window:

verxio-api: auth, workspaces, hosted Gemini key injection, runtime orchestration
verxio-web / desktop: operator workspace, model picker, agents, messaging
Phone gateways: WhatsApp and Telegram, where African SMBs already work
Kubernetes runtimes: one isolated agent runtime per business
Workflow agents: support, SDR, Micro-Manager, and cron delivery back to the phone

We built on Hermes, Nous Research’s open-source self-improving agent runtime, the same way a serious company builds on Linux. The business, Gemini hosting, isolation, academy motion, and phone-native operator are Verxio.

The first Verxio web scaffold landed 7 June 2026, after the 19 May submission start. 549 commits through mid-August 2026.

Challenges we ran into

Phone-first is harder than a web chat. WhatsApp pairing, media, and a single bot token cannot be shared across leftover processes. Gemini keys have to survive image rolls without being written onto disk.

Isolation is non-negotiable when an agent can see a ledger photo. Every customer gets a private runtime. That is operationally expensive. It is the only way this is safe.

The other challenge is trust. Owners will not paste API keys. Hosted Gemini is how we removed that step. If the operator doesn't live on WhatsApp or Telegram, it doesn't live in the business.

Accomplishments that we're proud of

A live product at www.verxio.xyz and app.verxio.xyz
Gemini as the hosted production model
The same operator on WhatsApp, Telegram, and the web
Isolated runtimes and a scale path on Kubernetes
Default support, SDR, and Micro-Manager agents
Verxio AI Academy: 3,000+ Nigerians and 400 small businesses trained on practical everyday AI — operations, bookkeeping, data entry, and record keeping
549 product commits inside the XPRIZE window

What we learned

If the operator doesn't live on WhatsApp or Telegram, it doesn't live in the business.

If Gemini is not hosted, the owner never starts.

If workspaces are not isolated, you cannot touch real records.

Training without a product is a workshop. A product without training does not reach the shops that need it. The academy and the operator have to be one motion.

AI agents are only valuable if they are accessible. Powerful runtimes are not enough. Businesses need a phone channel, hosted intelligence, private workspaces, and an operator that still remembers Monday on Friday.

What's next for Verxio

More owners on WhatsApp and Telegram. Industry packs for retail, fashion, and pharmacies. Local-language voice. Usage-based billing with hosted Gemini credits.

The longer bet is cluster operators; academy graduates who run Verxio for a street of shops, the way a bookkeeper once ran a stack of ledgers. Not replacing the owner.

Giving the owner a staff member they could never afford, and creating a new kind of work around it.