Invented figures
Chat tools can return a confident number that no query ever produced. In finance, a plausible wrong number is worse than no number at all.
Sovereign AI for finance & operations
Ask your business anything. Get numbers you can prove.
Royai answers questions about margin, cash, inventory and customers from your ERP and commerce data — on your own servers. Every figure comes from executed SQL against your ledger, never from the language model, and every answer and approval can be reproduced and audited.
- Runs on your hardware
- Figures from SQL, not the model
- Row-level security & PII masking
- Tamper-evident audit trail
What was our net revenue in Q2?
total_net_revenue = 91825.49
Where these figures are monetary, this tenant's ledger is denominated in EUR.
SELECT sum(net_amount) AS total_net_revenue FROM fct_sales JOIN dim_date USING (date_key) WHERE quarter = 'Q2'; -- read-only, under your RLS
Executed SQLnever model-written
Tenant scoperow-level security
PII maskedby the database
Loggedreproducible
Built for the systems you already run
- NetSuite
- SAP Business One
- Shopify
- CSV & Excel
- SQL databases
- REST & OData APIs
0figures the language model is allowed to invent. Numbers come from executed SQL; untraceable figures are removed.
100%of your data, models and inference stay on infrastructure you control.
SHA-256hash-chained audit trail for every approval, permission and identity change.
5business outcomes every agent is mapped to: margin, cash, revenue, operations and growth.
Why generic AI fails finance
Fluent isn't the same as right.
General-purpose AI writes numbers the way it writes sentences — by predicting what sounds plausible. That's fine for an email. It's a liability in a forecast, a price change or a board pack.
Data leaves your perimeter
Cloud copilots send your prompts — including your actual figures — to a third-party model provider you don't control.
No evidence trail
When your CFO or auditor asks “where did this number come from?”, a chat transcript is not an answer.
How it works
From question to provable answer.
Royai puts a deterministic gate between the language model and your numbers. The model helps with language; the database does the math.
- 01
Ask
Ask in plain language in Royai's chat — or schedule an agent to report to you every week.
- 02
Route
A classifier decides whether the question needs your data, your documents, or both.
- 03
Generate & guard
The model drafts SQL. A guard rejects anything that isn't a read-only query on approved, masked views; the database validates it before it runs.
- 04
Execute under your rules
The query runs read-only, under row-level security, with personal data masked by the database — per user and per column.
- 05
Answer with evidence
You get the figure, the SQL behind it and a log entry — SQL, data build and model fingerprint — that reproduces it later.
The language model writes SQL and sentences. It never writes the number.
The platform
One platform, from raw data to approved action.
Everything a finance and operations team needs to trust AI with real decisions — in one deployment you own.
Deterministic analytics
Natural-language questions become validated, read-only SQL. Every figure is the result of a query you can inspect.
Your data, modeled for decisions
A bronze-silver-gold warehouse on a single PostgreSQL: P&L, sales, margin, receivables, purchasing, inventory and a customer 360 — with data tests gating every refresh.
Private document intelligence
Ask about policies, SOPs and contracts. Answers cite their sources, respect module permissions — and say so when your documents don't cover the question.
Agents organized by outcome
Pricing, cash, inventory, customer and marketing agents — each mapped to the business outcome it serves.
Human approval, built in
Agents propose; people decide. Approve, edit, send back or reject — with a coded reason recorded for every rejection.
Measured value
Every approved proposal is tracked from predicted to realized result, measured on your own data — so you can see what AI actually delivered.
Agents
Agents that work on your P&L — under your approval.
Each agent declares the outcome it serves. Deterministic agents compute in SQL; advisory agents reason over your data, and any figure they can't trace to a query result is removed before you see it.
Dynamic pricing
SKU-level price elasticity with hard margin floors enforced in SQL; every change capped by your pricing policy.
Margin-leak detection on historical orders.
CFO advisor & collections
Working capital, receivables and cash questions answered from your ledger.
Dunning drafts for overdue invoices — refused unless sender and currency are verified.
Commercial & customer 360
RFM segments, cohort retention, churn risk and lifetime value — computed from your own orders.
Inventory & replenishment
Intermittent-demand forecasting (Croston, MAD) with reorder points and coverage, exported to Excel.
Approved replenishment can become a draft purchase order in SAP Business One.
Marketing advisors
Acquisition, conversion and strategy advisors with 48 built-in marketing skills — and no invented numbers.
CEO & process advisors
Weekly optimization brief
Signable AI data-readiness certificate
Scheduled reports in PDF & Excel
Questions finance teams ask Royai
Ask it the way you'd ask your controller.
Which SKUs lost margin last quarter, and why?
How is DSO trending by customer segment?
Which customers are at risk of churning?
What should we reorder this week to avoid stockouts?
Compare Shopify and wholesale gross margin year over year.
Which invoices are more than 60 days overdue?
What does our refund policy say about damaged goods?
Which vendors deliver late most often?
How Royai compares
Scored #1 of 7 for finance & operations decisions.
In our October 2026 assessment, seven platforms a company would shortlist for private, self-hosted AI were scored 0–5 and weighted for what finance needs most: trustworthy numbers, access control and evidence.
84%#1 of 7 as a finance & operations decision system
64%next best (n8n) on the same weighting
| Dimension | Royailocal-first P&L | OnyxMIT · enterprise search | DifyLLM app platform | n8nfair-code · automation | AnythingLLMMIT · doc chat | Open WebUIchat frontend | Mistral / Aleph AlphaEU sovereign models |
|---|---|---|---|---|---|---|---|
| Numerical determinismcan a figure be invented | 5 | 1 | 2 | 2 | 2 | 1 | 2 |
| Structured analyticsmodelled warehouse | 5 | 1 | 2 | 2 | 2 | 0 | 2 |
| Tenancy & access controlisolation, column-level control | 5 | 4 | 3 | 3 | 3 | 3 | 3 |
| Audit & evidenceprovable after the fact | 5 | 3 | 3 | 3 | 2 | 3 | 3 |
| Document retrievalchat over internal documents | 5▲1 | 4 | 4 | 2 | 3 | 4 | 3 |
| Agentic executionactions across real systems | 4▲1 | 3 | 4 | 5 | 3 | 3 | 4 |
| Outcome measurementproves delivered value | 4 | 2 | 2 | 3 | 1 | 2 | 2 |
| Local inferencemodel runs inside the perimeter | 5 | 3 | 3 | 3 | 4 | 4 | 4 |
| Jurisdictional sovereigntywho can be compelled | 5 | 4 | 3 | 4 | 4 | 4 | 4 |
| Connectorsreaching your systems | 3 | 4 | 3 | 5 | 3 | 3 | 4 |
| Time to first valuedecision to useful answer | 2 | 4 | 3 | 3 | 5 | 5 | 3 |
| Finance & ops scoreweighted total from the full scorecard | 84% | 54% | 58% | 64% | 55% | 49% | 59% |
4–5 strong3 adequate0–2 weak or absent▲improved since 26 Sep
Royai’s own assessment: Royai re-scored 2 October 2026 from its v1.10 codebase; other platforms from their official documentation, changelogs and licence pages current to 26 September 2026. Scores run 0–5. The total is the scorecard’s finance & operations score, which weights numerical determinism, structured analytics, access control and evidence most heavily and also counts two vendor-side dimensions not shown in this table. Weightings are a judgement, not a standard. Methodology available on request.
Security & sovereignty
Sovereign by architecture, not by policy.
Your data never has to be trusted to us. Royai is software you run — on your server or in your own cloud project.
Your hardware, your cloud project
Royai runs on a server you own or a GPU VM in your own Google Cloud project. We never host your data.
No third-party AI APIs
Models run locally. An undeclared remote inference endpoint is refused at startup — enforced in code, not in a policy document.
Isolation in the database
Separate schemas per company and PostgreSQL row-level security. A user without a membership gets nothing — never a default.
PII masked by the database
Column-level masking per group, enforced in SQL views — never left to the AI.
Secrets in a vault
Credentials live in OpenBao; per-company encryption keys never leave your server.
No telemetry
Traces carry metadata only — never prompts or answers. In our measurement of a full data refresh, the only outbound traffic was time sync and the admin VPN.
Royai (vendor side)Product updates, evaluation baselines and support. Never receives your questions, answers or data.
Your perimeter
Your server or Google Cloud project
ERPNetSuite · SAP B1
CommerceShopify
Files & APIsCSV · SQL · REST
PostgreSQL data spinebronze → silver → gold · tests gate every refresh
Router & SQL guardclassify · validate · RLS
Local language modelllama.cpp · open weights
Agentsby outcome
Approvalshuman in the loop
Audit chainSHA-256
On-premises
One Linux GPU server in your building. We provide the exact specification during discovery.
Your Google Cloud
A GPU VM inside your own project — inference stays in your tenancy, under your existing Google agreements.
Reproducible answers
Each answer is logged with its SQL, data build and model fingerprint — who was told which number, and when.
Tamper-evident audit
Identity, permission and approval events are SHA-256 hash-chained and verifiable end to end.
Daily self-checks
A built-in health check verifies the security invariants: row-level security, grants, audit chain, telemetry and egress.
Backups you can restore
Nightly backups with restore drills, so recovery is something you have tested — not something you hope for.
How to start
A fixed-scope path from demo to production.
Start with one decision that matters — margin, cash or inventory — and prove it on your own data before you scale.
- Step 130 min
Demo
See Royai answer real questions on a demo company, and map your systems to its data model. No data needed.
- Step 2Discovery
Data readiness & P&L priorities
We profile your sources, choose the first decision loop and give your CFO a signable AI data-readiness certificate.
- Step 3Fixed fee
Implementation
Your gold P&L models, approval workflows and agents — fixed scope, fixed fee, with a go/no-go decision at go-live.
- Step 4Ongoing
Run & measure
Annual licence with support and upgrades, while the value ledger tracks what each approved action delivered.
Founding customer program
A small number of companies get preferred first-year terms in exchange for a reference and a joint case study.
Consultancies & integrators
Co-sell and co-deliver Royai, keep 100% of your services and earn a recurring licence commission.
FAQ
Questions CFOs and IT teams ask us.
How is Royai different from ChatGPT or Copilot connected to our data?
Those tools generate answers — numbers included — with a language model. Royai uses the model only to draft SQL and sentences: every figure comes from a read-only query executed against your data, under your permissions, and every answer is logged with the SQL that produced it.
Can Royai still get an answer wrong?
Yes. A valid query can answer a slightly different question than the one you meant — gross instead of net margin, for example. That's why Royai shows the SQL, logs every answer so it can be reproduced, and keeps people in the loop for anything that changes your business. What it won't do is present a figure that no query produced.
Does any of our data leave our premises?
No. Your data, the models, inference and encryption keys run on your own server or in your own Google Cloud project. Royai sends no telemetry and does not use third-party AI APIs.
Which systems can Royai connect to?
NetSuite, SAP Business One, Shopify, CSV and Excel files, SQL databases and REST or OData APIs. New sources are added through a connector framework with conformance tests — tell us what you run.
What hardware do we need?
One Linux server with GPUs, or a GPU virtual machine in your own Google Cloud project. We provide the exact specification during discovery.
Can Royai's agents change data in our systems?
Only after a person approves. Agents write proposals; an approver approves, edits, sends back or rejects them, and every decision is recorded in a tamper-evident audit chain. Write-backs carry a recorded undo.
How long does it take to get started?
The demo takes 30 minutes and needs none of your data. Discovery and implementation are then scoped as a fixed project once we've seen your sources.
What happens to our deployment if we stop working with Royai?
Royai runs on your infrastructure, on open-source foundations such as PostgreSQL, dbt and llama.cpp, with no phone-home dependency. Your data and models stay where they are.
Book a demo
See your questions answered in 30 minutes.
Pick a time that suits you — the invite lands straight in your calendar with a video link.
- A live walkthrough on a demo company: questions, agents, approvals and the audit trail.
- Your systems mapped to Royai's data model — ERP, commerce, files.
- A first decision loop and what a fixed-scope pilot would look like.
- No data required. No obligation.
We use your details only to arrange and prepare the demo. See our privacy policy.
- 1About you
- 2Pick a time
Tell us a little about you
Takes under a minute. We'll tailor the demo to your systems and priorities.
Pick a time that works for you
Times are shown in your time zone. You'll get a calendar invite with a video link.
Loading available times…
Prefer email? Write to hello@royai.ai ·
You're booked.
A calendar invite with the video link is on its way to your inbox. If you'd like to share anything before the call, just reply to it.