kovacEvery organisation we talk to has tried putting a language model in front of its documents. The first demo usually goes well. The trouble starts when people begin to rely on it: an answer that sounds right but cites nothing, a confidential file surfacing for the wrong person, a policy that was replaced months ago still being quoted, and no way to tell whether last week’s change made things better or worse.
kovac is our answer to those problems. It is a platform, provided as a service by Inteligenta AIX, for building agents that answer from your own knowledge, show where every answer came from, and can be measured before and after you change them.
Knowledge that stays current
An agent is only as good as what it can find. kovac connects to the places your documents already live, including cloud storage, Google Drive, Confluence, SharePoint, SQL databases and websites, and keeps them in sync on a schedule. It reads office files, PDFs including scanned ones, spreadsheets, email and much more, keeping tables intact and describing charts and figures so their content can be searched.
Documents have a lifecycle. You can keep versions, set the dates a document is in force, and archive what no longer applies, so answers come from the version that is current.
Answers you can check
Retrieval in kovac combines keyword and semantic search, reranks the candidates, and works across languages, so a question asked in one language can find a document written in another. Every answer carries citations, and each citation opens the exact passage in the original document. When someone asks “where does it say that?”, the answer is one click away.
Access that follows your rules
People sign in through single sign-on. Roles decide who manages knowledge, who builds agents and who only asks questions. Collections and agents can be limited to particular departments, and when a connected source reports who may read a document, kovac honours it. These rules are enforced inside the database, not only in the interface, and every change to knowledge is written to an audit log you can export.
Agents as configuration
An agent in kovac is a set of instructions, the knowledge it may search, the skills it may use and the model it runs on. You build and change agents in the management console without deploying code, or package them as plugins and import them. Skills that run code do so in an isolated sandbox, separate from the service that holds your data. Agents remember useful context across conversations, and guardrails check both the questions they receive and the answers they give.
When an agent is ready for the public, you can put it on your own website with a single script tag. The widget carries a key that can start a chat and nothing else, and limits apply per visitor.
Measured, not assumed
The part we care about most is evaluation. kovac generates test questions from your own documents for your experts to review. The accepted set becomes a yardstick: how often the right passage is found, how that changes when you adjust a setting, and whether answers stay faithful to their sources. kovac can try several ways of splitting your documents into passages and tell you which one performs measurably better on your own questions, before you switch.
What comes next
We are working with a small number of organisations to bring kovac into everyday use. If your team has a body of documents that people keep asking about, and you need answers that hold up to scrutiny, we would like to hear from you. Request access and we will set up a workspace and walk through it with your team.