AI AGENTS

AI that knows your business — and works on WhatsApp.

We build AI agents trained on your product catalog, pricing, FAQs, and order data. Deployed where your customers already are: WhatsApp, web chat, your support inbox. Handles the same five questions you answer every day, escalates anything complex to a human, costs less than a part-time staff member.

₦1.5M – ₦12M2–10 WEEKSOPTIONAL ₦150K+/MO RETAINER
Discuss your agent
SHAPES

Three kinds of agent.

Most agents we ship fall into one of three shapes. The shape decides the price, the timeline, and what data we need from you before kickoff.

01CUSTOMER-FACING

Customer-Support Agent

₦1.5M – ₦3M2–4 WEEKS

Trained on your product catalog, FAQs, and order data. Deployed on WhatsApp. Speaks English and Pidgin. Handles "how much?", "is it in stock?", "where's my order?" 24/7 — and hands the conversation to a named human the moment it doesn't know.

  • WhatsApp Business API on your existing number
  • Catalog + price book + order data, indexed nightly
  • Pidgin and Nigerian English, tuned to your tone
  • Human-handover with full conversation transcript

Best forteams answering the same five questions 200+ times a week — most of them about price, stock, or order status.

02INTERNAL-FACING

Internal Operations Agent

₦3M – ₦7M4–7 WEEKS

Trained on your internal SOPs and data. Helps your team draft quotes, summarize meetings, generate reports, and route work. A senior assistant for everyone — sat behind a Slack or WhatsApp thread, not a new dashboard nobody opens.

  • Fine-tuned on your SOPs, templates, and historical work
  • Tool use against your CRM, ledger, and Drive
  • Per-role permissions and audit log per request
  • One-month bedding-in with weekly retraining

Best for15–80 person teams where the same senior person is the bottleneck on every quote, every report, every "quick question".

03VERTICAL

Industry-Specific Vertical Agents

₦4M – ₦12M6–10 WEEKS

Built for specific verticals — clinic patient intake, real estate lead qualification, school parent communication, law firm contract drafting. Trained on the workflows of your industry, not a generic chatbot bolted onto your homepage.

  • Domain fine-tuning on your sector's vocabulary
  • Workflow scripts written with a domain expert
  • Named accuracy targets with an evaluation harness
  • Production deployment, monitoring, and rollback

Best foroperators in clinics, schools, agencies, and chambers who've watched generic chatbots fail on their domain — and want one that knows the work.

LOCAL DETAIL

We build for how your customers actually talk.

Generic chatbots fail in Nigeria because they assume customers ask in textbook English. Ours are trained on Pidgin, on local product names, on how Nigerians actually negotiate, and on what to do when the network drops mid-conversation.

They escalate to a human when the customer is angry, when the question is high-stakes, or when they don't know — and they log every conversation for review.

Pidgin-capable
Tuned on conversational Pidgin, not the textbook kind. Handles "wetin be the price na?" the way a real customer would ask it.
Local product names
Knows that "Indomie" is a category, "ata rodo" is scotch bonnet, and your SKUs by their warehouse nickname, not just the formal name.
Network-aware
Resumes context after a dropped connection, deduplicates retries, and never charges a customer twice when the second "send" was a network artifact.
Pricing guardrails
Quotes only from your live price book. If a price isn't there, the agent says "let me check" and routes to a human — never invents a number.
Escalation rules
Anger detection, refund requests, and any question outside the trained scope hands off to a named human with the full transcript attached.
Conversation log
Every conversation written to an append-only log you can read, sort, and export. Used for weekly retraining and for sleeping at night.
START HERE

Most builds start with an audit.

Before we quote, we spend a week inside your operations — reading your WhatsApp threads, sitting with your support staff, looking at the actual questions customers ask. We come back with a ranked list of what an agent should handle, what it should escalate, and what it shouldn't touch at all.

₦750K – ₦1.5M1–2 WEEKSREPORT IS YOURS TO KEEP
Start with the audit
QUESTIONS

Things people ask before buying an agent.

Four we hear on every first call. If yours isn't here, the audit conversation is the place for it.

No — and we treat this as the single biggest risk. The agent never generates prices from its own memory. It reads from a live price book wired into the same system your sales team updates, and quotes only from there. If a SKU isn't in the book, the agent says "let me check" and hands off to a human. We test this with an evaluation suite of 200+ adversarial prompts before launch — invented discounts, made-up bundles, "but your competitor said…" — and the agent has to refuse every one before it ships.
Yes. Every customer-facing agent ships with a Pidgin layer trained on real Nigerian WhatsApp conversations — not the textbook version. It understands "abeg how much for am?", "wetin dey?", code-switching mid-sentence, and the dozen ways a customer might ask whether something is in stock. It also replies in whatever register the customer started in — formal English to formal English, Pidgin to Pidgin. Your tone, your way of speaking, stays consistent across the whole conversation.
It says so, and hands the conversation to a named human with the full transcript attached. We build three kinds of escalation: low-confidence (the agent isn't sure of the answer), out-of-scope (the question is something we agreed the agent shouldn't handle — refunds, complaints, anything legal), and sentiment (the customer is angry, frustrated, or has used a keyword you flagged). The customer never sees a dead-end — they see “let me get someone on this” and a human picks up the thread within minutes.
Off-the-shelf tools are decision trees with a chat interface on top — you draw the flow, the customer clicks through it. That works for the easy 20% and fails the rest. Ours are language models fine-tuned on your data: they understand intent, handle the messy 80% of conversations that don't fit a tree, and only fall back to a scripted flow when you want them to. The integration also stays yours — the model weights, the prompts, the price book pipeline, the conversation logs — not rented from a vendor who can change the terms next quarter.
DISCUSS

Tell us what your customers keep asking.

Send us the shape of your support — channels, volume, the five questions that come up every day. We'll come back inside two business days with a fixed quote, a build timeline, and the three questions that decide whether you're at the lower or upper end of the range.

Discuss your agent