Use cases

Same index. Very different questions.

K-Lake crawls, extracts and enriches your files once. What people then ask of it depends on who they are. These are the six starting points we see most often, with the capabilities each one leans on and the outcome it delivers.

For knowledge, operations and IT teams

Assistants that answer from company files, without a data export.

Most organisations want Microsoft Copilot, Claude, ChatGPT or an agent of their own to answer questions from policies, contracts, reports and correspondence. The obstacle is never the model. It is that the files live on shares and in libraries with decades of permissions attached, and nobody wants to copy them into a vendor's platform to make them reachable.

What K-Lake does

K-Lake indexes the files where they are and exposes them to any MCP assistant through its own MCP server. The assistant connects as the individual user, sees only what that user could already open, and receives a citation for every fact so people can open the original and check. Every query is recorded with who asked and what came back.

What changes

People stop hunting through folders and asking colleagues where a document went. Answers arrive with the file behind them, so they can be trusted and acted on. Security teams get an assistant rollout that does not widen access, and an audit trail they can query.

For legal, procurement and finance

Contract and counterparty intelligence from the archive you already hold.

A contract archive answers "which files mention this supplier?" only if someone knows to look. It does not answer which agreements renew this quarter, who signed them, which counterparties share the same adviser, or what else the same three people appear in together.

What K-Lake does

Entity enrichment reads every extracted agreement and records the people, organisations, locations and agreements it names, and the relationships the text asserts between them. The Explore page shows one counterparty at the centre of everything it touches; an intersection query lists the files that put a set of people in the same room; a reverse lookup says what any single document is about. Duplicate spellings of a name are merged, reversibly.

What changes

Renewals, obligations and concentrations of risk surface from documents that were previously only findable by name. Counsel can answer a diligence question with a list of files rather than a week of reading.

For data governance, security and storage teams

Know the estate before you classify, migrate or clean it.

Every migration, classification and cleanup programme begins with the same question: what do we actually have? Answering it usually means a one-off scan that produces a spreadsheet nobody trusts a month later, and that says nothing about what is inside the files.

What K-Lake does

Discover turns what K-Lake has already crawled into a live estate overview: totals, how much is searchable, and composition by type, language, size, age, owner, storage tier and source. Governance signals flag world-readable, stale, unowned and duplicated files. Click any bar and the files behind it are listed. Coverage is stated honestly wherever a facet depends on metadata only some sources captured.

What changes

Programmes start from a measured baseline and can be re-measured as they run. Storage spend on cold and duplicated data becomes visible. Exposure is a number with a list behind it, not a suspicion.

For defence, public sector, financial services and healthcare

Grounded AI where the data cannot leave.

Some estates cannot be sent to a hosted model or a SaaS index under any terms: classified material, patient records, regulated financial data, or simply a network with no route to the internet. Those organisations are told to wait, or to accept a diminished experience.

What K-Lake does

K-Lake runs entirely inside your cluster. Extraction, OCR and transcription run on engines you host. Licensing is validated offline with no call-home. In air-gapped mode a self-hosted model answers over MCP with the same per-file trimming and the same audit trail, so nothing crosses the boundary in either direction.

What changes

Teams in the most constrained environments get the same grounded, cited assistant as everyone else. Risk and compliance reviewers get a security overview written for them, a daily vulnerability report and signed images to verify.

For media, training, research and investigations teams

Make audio and video archives searchable by what was said.

Recorded meetings, training sessions, interviews, hearings and broadcast archives hold knowledge that no filename describes. Finding the moment a subject was discussed means scrubbing through hours of footage, so in practice nobody does.

What K-Lake does

The media engine transcribes spoken words and, for video, reads the text shown on screen: slides, captions and titles. Every segment is timecoded, so a search hit deep-links to the exact moment and the player opens there. Transcription runs on self-hosted models, so no media leaves your environment, and it scales out as a fleet.

What changes

A recording becomes as findable as a document. An assistant can answer from what was said in a meeting and cite the minute it was said.

For managed service providers and platform vendors

Offer governed AI over customer data, one deployment, many tenants.

Providers want to give each customer grounded AI over their own files without running a separate stack per customer, and without any possibility of one customer's data reaching another's assistant.

What K-Lake does

Every source, user, token and file belongs to exactly one tenant, and isolation is enforced by row-level security in the database rather than by application code. Each tenant gets its own MCP endpoint and sign-in issuer, can federate to its own identity provider and sign tokens with its own keys, and its administrators see only their own. Licences can carry a product name for white-label deployments, and Azure Marketplace entitlement follows the plan.

What changes

One platform, operated once, sold many times, with a tenant boundary you can explain to a customer's security team in a sentence.

Your use case

Not on the list? It probably still fits.

Tell us what you hold and what you want to ask of it. We will show you K-Lake on a representative estate and be honest about whether it is the right tool.

Request a demo
  • Evaluate on one VMThe full stack, in one command, with a 30-day licence.
  • Delivered by the buildersKDBL consultants run the proof of concept with you.