Releasing October 2026
The hard part of AI on business data isn’t the AI
It’s knowing which table holds revenue, which of four date fields matters, and who’s allowed to see the result. Sharperlight has spent two decades building that layer. Version 8 opens it to the AI your organisation already uses.
50+ datamodel connectors
Live, governed data access
Excel, web, dashboard delivery
Connected or fully air-gapped
The argument
An AI is only as good as the model of your business it is given
Point a general-purpose AI at a raw ERP database and it does what anyone would with no context: guesses. It finds six columns that look like revenue and picks one. It joins tables that shouldn’t be joined. The number comes back confident, plausible and wrong.
The fix isn’t a better model. It’s giving the model a description of the business it can rely on.
That’s what a Sharperlight Datamodel has always been – data structures, relationships and business rules encapsulated at the lowest level and presented upward as humanised names, defined joins and enforced security. It’s why Sharperlight users have never written SQL.
Version 8 exposes that layer to AI. The assistant doesn’t see your database. It sees your business.
The industry has a name for this now. Every AI analytics vendor is racing to build a semantic layer because without one, the model has nothing to anchor to. Sharperlight has been shipping that layer since before the term existed – in production, across 50+ connectors.
Version 8 isn’t us adding AI. It’s the hard part already being done.
The advantage is already in production
Everyone else is building the semantic layer.
We’ve been running one for twenty years.
Two ways to connect
Version 8 doesn’t put a chatbot in Sharperlight
It makes Sharperlight available to the AI you already use – two ways, because not every organisation is allowed to connect one.
What your assistant can do
In practice
One question, end to end
In your own AI assistant, with Sharperlight connected.
Governance
Where does the AI actually see my data?
It doesn’t see your database. It sees the datamodel – structure, not records – and uses it to build a query Sharperlight executes under the credentials of the person asking.
Read, build, or act – you choose which
Sharperlight’s assistant can do more than read. It can publish reports, build dashboards, trigger schedules and write back to source. So access is scoped per connection:
| Tier | The Assistant can | Typical Use |
|---|---|---|
| Read | Query and return results | The default posture |
| Author | Create and modify reports, dashboards, datamarts, schedules | Report developers and partners |
| Write | Update records through the datamodel's validation rules | Budgets, forecasts, corrections |
Authoring and writeback are opt-in per connection. Writeback through an AI obeys the same validation and constraint rules as writeback through Excel – different interface, identical enforcement.
Air-gapped in detail
If the AI can never touch your network, export the map instead
Some organisations won’t connect an AI to a production system. Defence, government, health, anyone mid-audit – the answer is no, and no amount of governance architecture changes it.
The export describes the whole datamodel: every table in its business folder, every field with its data type, aggregation, default filter and valid values, and every join with the field that relates them. Structure only.
It isn’t a schema dump. Before writing anything, the AI must produce an execution plan and verify every table, field, join and filter against the metadata. If a path can’t be confirmed, it stops rather than inventing a field name.
Export what you need. Take the whole model, or scope it to the folders the work needs. A smaller export leaves the assistant more room to reason. Connected over MCP, the server supplies schema on demand and holds large result sets server-side.
Proof
The engine underneath this isn’t new
The AI layer is new. The engine has been in production for years, on databases most reporting tools won’t touch.
Mining and resources, environmental organisations, aquaculture and agrifood, universities, government, healthcare, retail, aviation and professional services. Multi-database environments consolidated into one reporting layer after system upgrades left incumbent tools behind.
That is what Version 8’s AI inherits – not a demo dataset.
What ships
What ships, and what’s licensed
Already on Sharperlight? Your datamodels are the upgrade path – the work your team has done defining tables, relationships and rules is what makes the AI useful on day one.
Nothing needs procuring before you start. You bring the AI – Version 8 ships no model and no chat interface of its own.
For partners
A differentiator you can demo
Your competitors’ AI story starts with “first we need to model your data.” Ours starts with “your data is already modelled.”
Two conversations, two doors. MCP is a licensed component with configuration work attached – publishing models, scoping Service Codes and defining what each assistant may reach. The export carries no additional licence, opening accounts that would otherwise refuse the AI conversation outright.
Lead with whichever the customer’s security posture allows.
FAQ
Version 8 questions
The essentials for customers, security teams and delivery partners.
When is Version 8 available?
October 2026. Both pieces ship in the standard install.
Does Version 8 include an AI?
No. It makes your datamodels usable by an assistant you already have – connected, or running inside your own network.
What has to be licensed?
Only the MCP server, and only when enabled.
Which AI assistants work with it?
Connected: any assistant that supports MCP. Air-gapped: any AI at all – the export is a text file.
Is my data sent to an AI provider?
That is determined by the AI you choose. Sharperlight has no model and no provider relationship. With the export, no results are involved at all.
Can the AI change or delete our reports?
Only if you grant that tier. Connections default to read.
Can it write back into our ERP?
Only where explicitly granted, and only through the datamodel’s existing validation rules.
Do I need to know SQL?
No. Sharperlight has never required it.
Will my existing reports still work?
Yes. Version 8 adds to the platform rather than replacing it.
What does it cost?
Pricing will be confirmed before launch. Contact the Sharperlight team for current guidance, contact sales to begin your AI journey with Sharperlight.
Releasing October 2026
The semantic layer is built. Bring your AI.
Leave your email for release notes, demos and upgrade guidance – or book time now.