An employee that knows your numbers.
Not a chatbot. A working assistant wired to your live business data. It answers from your actual books, flags what needs attention before you ask, and prepares the work for you to approve.
Questions an owner actually asks. Answered from your books.
Every figure comes out of your live SQLite — never a guess. The model only explains the numbers and tells you what to do about them.
“What made the most profit this month?”
Returns the top products by profit (not just revenue), with margin %, computed from your live books — never guessed.
“What should I reorder, and how much?”
Uses each product's own sales velocity, lead time, and stock to suggest order quantities, sorted by days of cover left.
“Why did revenue change this week?”
Compares this week against the previous equal period and names the products that drove the gap, gainers and losers.
“Which customers have stopped buying?”
Segments your customers into VIP / loyal / at-risk / churned by recency, frequency, and spend — at-risk surfaces first.
“Which supplier is most reliable?”
Scores every supplier on on-time delivery %, fill rate, and total spend across your purchase orders.
“What stock is expiring soon?”
Lists batches expiring within N days and the value at risk — pharmacist-grade for chemists, useful everywhere else.
Prepares the work. You approve.
The assistant never mutates anything on its own. When it has work ready, it shows you exactly what will change — counts, totals, who it affects — and waits for one tap. Every action runs through Omnix’s permissions, just like a human user.
Every proposal and outcome is logged to an action ledger you can review in /settings/ai.
Draft a purchase order
Generated from reorder suggestions with the right supplier, items and quantities. Nothing is sent — you review and approve.
Categorise products
Tidy a messy catalogue: the assistant proposes a category for each product; you click Apply.
Set reorder levels
Suggests sensible reorder thresholds from velocity. Your approval applies the change.
Harmonise a messy import
Maps any supplier CSV / Excel into Omnix's fields — even with Swahili headers, weird casing, or missing columns.
Six kinds of business intelligence, built in.
Inventory intelligence
Dead stock by value, reorder suggestions with quantities, expiry risk, duplicate detection, margin issues, price anomalies.
Sales & financial insights
Profit leaders, revenue change explained, cashier performance, payment-method mix, Z-report summary in plain English.
Customer insights
VIPs, at-risk, churned, inactive — segmented by RFM. Know who needs a follow-up before they slip away.
Supplier intelligence
Score every supplier on on-time %, fill rate, and total spend. Find who actually delivers.
Imports & spreadsheet harmonisation
Drop in any supplier file. The assistant detects structure, maps columns, normalises units, flags duplicates, previews before you commit.
Anomaly narration
When something looks off — revenue dip, below-cost pricing, an unusual void — the assistant explains why it matters and what to do.
The AI is yours, not ours.
We don’t resell tokens. We don’t mark up inference. Pick a provider, paste your key, and your calls go direct from your machine to your provider — Omnix never sees the prompt or the response.
Free tiers from Groq and OpenRouter handle a busy day comfortably. Paid tiers (OpenAI, Claude) plug in identically. Or point it at a local Ollama model on the same PC and run AI fully offline.
Groq
Free tierFree, very fast
OpenRouter
Free tierAggregator with free models
DeepSeek
Pay per tokenCheap, very capable
OpenAI
Pay per tokenGPT-4o, GPT-4o-mini
Anthropic
Pay per tokenClaude Sonnet, Haiku
Google AI
Free tierGemini Flash, Pro
Custom / Ollama
Local or hostedAny OpenAI-compatible URL
Private by default
Every prompt is scrubbed for PII before it leaves the machine — phone numbers, KRA PINs, IDs, API keys. Keys are stored encrypted in your local SQLite database. Disable AI completely from settings if you prefer, and the rest of Omnix still works.
Online when it helps, offline when it must
Insights, reorder suggestions, dead stock, customer churn — these are all computed by SQL on your machine and work with zero internet. The model only steps in to explain in plain language; without a provider configured, the numbers still come through.
Shipping regularly.
Vision ingestion
Drop a photo of a supplier invoice; the assistant drafts the goods-received note for review.
Demand forecasting
Ingredient & medicine shortage prediction, fast-mover forecasts — explained, never opaque.
Local embeddings + RAG
Semantic search over your own data + the docs. Fully offline via a local Ollama model.
Per-trade playbooks
Pharmacist, retailer, restaurateur and hardware modes with proactive insights tailored to the trade.
Roadmap is honest, no calendar dates · all shipped features in /changelog