Turn scattered enterprise data
an ontology with sources attached
into
We connect the objects and relationships across ERP, groupware, spreadsheets, and documents into a knowledge layer where every answer can be traced back to its evidence. Deployment and operating scope are set after assessing your data environment and security policy.
without touching the source systems
and for declining to answer
environments and your own infrastructure
defined per project
Search works,
so why can you not trust the answer?
There is plenty of data but it is not connected, and the AI answers without showing why.
Fragmented data
Data spread across ERP, groupware, spreadsheets, and documents often has no shared semantic model. When that is the case, someone has to work out both what the question means and where the data lives, so an answer passes through several people.
The limits of RAG
RAG built on vector search alone suits similar-document retrieval, but a query that follows several relationships, such as “which contracts, orders, and owners are linked to this account?”, needs a separate structure and verification.
The build and operate barrier
An ontology has to keep reflecting changes in data and business rules after it is built. Without operational ownership and a review process, the knowledge does not stay current.
Connected data becomes
an organised knowledge layer
All you have to do is connect the data and review the schema once.
Connect
Databases connect read-only and documents connect by upload. The source systems are left as they are and the knowledge layer is composed on top. Korean HWP documents are in scope too.
Automated design
AI profiles the data and proposes an ontology schema. Your team reviews the reasoning and the samples on a read-only screen.
Build
Objects and relationships are extracted and verified against the agreed schema. Duplicates are merged non-destructively, and change history lets you inspect any earlier state.
Ongoing operation
The Alphaca Labs operations team detects and reviews changes in the data. Knowledge that does not rot over time: operations is part of the product.
Not a read-only knowledge graph,
an operating system you query and act on
Multi-hop queries
“Who owns the delayed orders linked to this account?” Questions that follow relationships are answered by traversing the graph. That is reasoning, not similar-document search.
Answers backed by evidence
Each answer is linked to the source document, the passage, and its review state. Queries that do not meet the configured bar can be declined or routed for review.
MCP standard integration
A built ontology can connect to any agent that supports MCP. Tools, permissions, and query scope are configured for the environment it runs in.
Action: beyond lookup
Actions such as state changes and notifications are defined together with validation, permissions, and audit logging. It is the layer where an agent can safely do work.
Dedicated instances and isolated deployment
A dedicated environment per client, your own VPC, or an on-premise or air-gapped deployment. Data movement and operational access are agreed against your security policy.
Built for Korean enterprise data
It handles the formats you actually meet in Korean enterprise data: HWP documents, normalising corporate name notations, and identifying the same company by business registration number.
Show me the evidence
is the whole of trust
Trace objects and relationships to their source
Every object and relationship in the ontology is linked to its source document and location. You can read the original passage an answer relied on.
Set the bar for declining to answer
When the evidence is thin or the review bar is not met, the policy can route the query to no answer or to review.
Operations is what keeps the trust
Changes in data and business rules are reviewed on a schedule, and any change to the schema or review criteria is agreed with your operations owner.
Who owns the delayed orders linked to Hangyeol Co.?
One order is delayed, #4821, and the sales owner is J. Kim, Sales Manager. The delay is recorded as a goods-receipt delay on 12 September.
The tighter your security requirements,
the better Ontokit fits
From cloud to air-gapped on-premise, the platform goes to where the data is.
Dedicated cloud
- An isolated dedicated stack per client
- Remote operating scope agreed
- Storage location and access rights configured
Your own cloud account
- Deployed directly inside your infrastructure
- Network and security policy reflected
- Data managed within your infrastructure policy
On-premise / air-gapped
- Isolated configuration limiting external connections
- On-premise model options reviewed
- On-site operating scope agreed separately
Frequently asked questions
The conditions people check most often before adopting.
What data sources can you connect?
Databases, business systems such as ERP and groupware, and spreadsheet or document files are all in scope. How each one connects is decided after checking the source system's API, permissions, and security policy, along with the document formats involved.
How long does building an ontology take?
It depends on the number of sources, schema complexity, how standardised the data already is, and the availability of your reviewers. We propose a schedule after the assessment stage settles which workflows to connect first and what the review criteria are.
Does data have to leave our cloud?
A dedicated cloud, your own VPC, and on-premise or air-gapped environments are all options. Data movement, calls to external models, and operational access rights are agreed against your security policy and technical environment.
How do you manage answer accuracy and trust?
Objects and relationships are linked to their sources, and the answer policy and review criteria are configured. Queries that do not meet the bar can be declined or routed for review, and errors found during operation are tracked as change history.
When is Ontokit not the right fit?
If there is no access to the source data, or the business terminology and its owners are not yet settled, building straight away is difficult. In that case it is better to start with an assessment that maps the data landscape and the priority workflows.
The moment you connect the data,
your company's knowledge becomes an asset
Tell us about your current data environment and we will propose an adoption scenario and a schedule.