Qaicu Core:
AI-optimised data layer for your own AI solutions
Build your own AI solution on a data foundation that is already in place. Qaicu Core calculates the figures correctly before they reach the language model, and it reduces token usage. You can start quickly, and the solution can be customised to your exact needs later.
At a glance
Qaicu Core brings your company’s business data, documents, semantics, permissions and calculation together in a shared data layer. Your own agents and applications use it through the open API and MCP (Model Context Protocol).
Core brings in information from source systems and maps the business relationships within it. It calculates the figures needed before anything reaches the language model. The agent gets an information base scoped to its task.
A shared data layer for multiple agents
Information sources
Finance, ERP, HR, CRM, documents and more
Qaicu Core
Links the information, applies permissions and calculates the figures
Your own agents and apps
Use the same definitions, calculation rules and permissions
Every AI agent in a company needs answers to the same questions. Where does the information come from, what does it mean and how are the figures calculated? Who is allowed to use it?
Qaicu Core provides those answers as a shared service. A finance agent, a sales assistant and a management app can all use the same customer definitions, calculation rules and permission management.
The CTO gets a reusable architecture. The business can build different AI solutions on consistent information and definitions.
How does Qaicu Core work?
1. Brings information together and stores it
Pre-built integrations bring in information from your company’s business systems. For other information sources and applications, there is the open API.
Core stores, versions, normalises and enriches the information. The first use case gets the sources it needs, and the setup grows from there as required.
2. Maps business concepts and relationships
Core can link an ERP customer, a CRM account and a contract party to one business entity. A quote, an order, an invoice and an accounting transaction can form one chain, from quote to ledger.
This semantic layer turns information from different sources into a shared resource. An agent does not have to work out each system’s structure from scratch.
3. Checks permissions
Core restricts the information before it reaches an agent or a language model. API and MCP requests go through the same permission layer.
A user’s job title or an agent’s instructions do not by themselves grant access to information. Permissions are defined separately.
4. Calculates results in the back end
Core calculates totals, balances, margins and other KPIs with defined calculation rules. The same source data and the same rules produce the same calculated result.
The language model receives the finished calculation and the background information it needs to interpret it.
5. Serves agents and applications
Your own AI agents use Core through MCP or the open API. The same data layer can serve multiple agents, applications and language models.
Each piece of information carries its source, version and validity period. That makes answers easier to check and trace.
Calculation improves accuracy and keeps language model costs in check
An agent asks about a customer’s profitability
Raw data to the language model
Invoices and costs
Language model calculates
Calculation before the model call
Invoices and costs
Qaicu Core calculates
Language model interprets
Say an agent asks about a customer’s profitability. It does not have to send every invoice and cost to the language model to calculate.
Core links the customer’s transactions and limits them to the relevant period. It calculates the result with the agreed definition of margin. The language model gets the result and the breakdowns that matter for the analysis.
This has two practical effects:
- Calculation is repeatable and verifiable. The calculation engine produces the numerical result. Its accuracy depends on the accuracy of the source data and the calculation rules.
- Less data goes to the language model. Filtering, combining and calculation all happen before the model call. That cuts unnecessary token usage compared with having the model process large volumes of raw data.
The language model can focus on explaining the result, making comparisons and preparing follow-up actions.
What can you build on Core?
Management agent: the state of the business
Need: The CEO wants to look at growth and profitability by customer and by project.
Core’s role: Combines financial, CRM and project information, calculates comparable KPIs and returns the breakdown behind them.
Your application’s role: Presents the findings in a management view, relates them to strategy and goals, and supports follow-up questions.
Benefit: Management’s AI solution works from shared business definitions and calculation rules.
Finance agent: period-end close checks
Need: The CFO wants to automate recurring checks and focus expert work on variances.
Core’s role: Gathers the transactions, applies defined calculation and comparison rules, and returns the variances to review with background information.
Your agent’s role: Prepares a checklist and draft comments, and passes the findings on for review.
Benefit: You can rerun checks with the same rules and verify the basis for each one.
Sales agent: upselling and meeting preparation
Need: The sales director wants to help salespeople prioritise accounts and prepare for meetings.
Core’s role: Combines the account’s CRM records, purchase history, contracts and profitability into a shared customer context.
Your agent’s role: Prepares a customer summary, suggests upsell opportunities worth looking into and produces a meeting agenda.
Benefit: Sales has both customer relationship information and actual business figures behind its proposals.
HR agent: answers based on the company’s own guidelines
Need: HR wants to make staff guidelines available through the company’s own service channel.
Core’s role: Gives the agent the documents relevant to the question, with their sources, filtered by permissions.
Your agent’s role: Composes an answer, shows the sources and, when needed, refers the matter to HR.
Benefit: The same governed guidelines can serve different interfaces without a separate information base for each agent.
Project agent: actuals to support bid estimates
Need: Operations management wants to use the costs and margins of past projects when preparing new ones.
Core’s role: Combines information on projects, working hours, purchases and invoicing, and calculates comparative figures within agreed parameters.
Your agent’s role: Compiles comparison data and helps identify the cost drivers that affect the bid.
Benefit: Bids can draw on actuals that have built up across several systems.
Service provider agent: new services for clients on a shared foundation
Need: An accounting firm, for example, wants to offer its clients ongoing financial analysis.
Core’s role: Combines client-specific financial data, runs the defined calculations and restricts information through permissions.
Your service’s role: Produces draft analyses, client views and findings that need an expert’s review.
Benefit: You can automate recurring information gathering and calculation, so your experts can serve a wider client base.
Pre-built integrations, open API and MCP
Information sources
Pre-built integrations reduce the development work needed to connect information sources. Implementation includes confirming, system by system, what information is needed and whether it is available.
Applications
The open API connects Core to your own applications, services and other systems.
Agents
MCP connects AI agents to Core’s information base and calculation capability.
You do not have to replace your existing business systems to use Core. Your ERP, CRM, accounting and document management systems can stay as information sources.
The information base stays in place even if the language model changes
Qaicu Core keeps integrations, business concepts, permissions and calculation separate from the language model the agent uses.
If you choose a new model or switch agent frameworks, you can keep using the shared data layer. The new model or agent still needs testing, but you do not have to rebuild the information base from scratch.
You depend less on any one language model. The work in the data layer carries over to future use cases.
Qaicu Core or the ready-to-use Qaicu service?
Qaicu Core
Choose Qaicu Core if you build your own agents, interfaces or AI-based services and need a governed information base under them.
Qaicu Pro or Enterprise
Choose Qaicu Pro or Enterprise if you want a ready-to-use working environment for your staff. Conversations, views, goals and agents are all part of the same service.
We start with what one agent needs to know
Together, we define what the first use case needs: information sources, business concepts, calculation rules and permissions. Then the agent connects to Core via the API or MCP.
Real business questions show whether it works. Do the calculations add up, can the sources be found and is the information restricted correctly? The same tests can measure language model usage and the size of the context passed to the agent.
Once the foundation works, we extend it to more agents and information sources.
Information needs
Sources, concepts, calculation rules and permissions
Connection
Agent to Core via API or MCP
Validation
Calculations, sources, restrictions and language model usage
What do your agents need to know and calculate?