On-premise AI: your documents never leave the building

Kodetank Lokal KI: a finished product, not a project

Lokal KI is one service on one machine at your site. PDFs and scans become searchable text with the original alongside, and a language model running on the same machine answers questions about the documents. No cloud account and no telemetry – and you can check that yourself with a firewall.

• Search with Norwegian inflection and by topic, with hits per page and highlighting
• Ask the documents: answers cite document and page – and “not found” is a valid answer when it isn't there
• Number check: amounts, dates, organisation numbers, account numbers and payment references are read a second time, and mismatches are flagged
• Structured extraction from invoices and contracts to CSV or JSON
• Access per folder, an audit log and encrypted backup with a key only you hold
• API, webhook and MCP server, so your own systems and AI assistants on your own network can use the archive

Try the demo (in Norwegian) ↗

The toolbox: PDF work without the cloud

What you would otherwise do in Acrobat or an online service is done on your own machine. The tools come with Lokal KI, and the originals are never changed.

• Redaction: finds national identity numbers, account numbers, names and other details – the text under the redaction is removed for good
• PDF tool: merge files, move and rotate pages, sign, compress and save with searchable text
• Stamps and page numbers: “DRAFT”, “CONFIDENTIAL” or “Received” with the date
• Exhibit builder: one bundle with cover page, table of contents, numbered exhibits and bookmarks
• Compare versions: see what the other party changed between two drafts
• Clean before sending: removes author, comments, tracked changes and other hidden content
• E-mail to PDF: the correspondence in a case becomes PDFs with the attachments appended
• Fill in forms: PDF forms from public authorities and the courts are completed and signed in the browser

See all features (in Norwegian) ↗

When on-premise AI is the right choice

On-premise AI fits where data cannot be shared with third parties, or where internet access is not a given.

• Law firms: ask the case file without anything leaving the office, with access per case
• Accounting and audit: searchable vouchers with number checks and retention under the Bookkeeping Act
• Clinics and healthcare: referrals and discharge summaries read locally, rule-based deletion and full traceability
• Property and housing co-operatives: contracts, minutes and condition reports in one searchable archive
• GDPR and privacy: no third-country transfers, no processors abroad
• Trade secrets: pricing, contracts, product development
• Unstable or missing connectivity: construction sites, ships, industry

One user or the whole office

The same program and the same data format in every variant, so you can start small and move later.

• Local Mac: a Mac mini or Mac Studio delivered ready to use, with a server disk that can be locked in the safe
• Local Windows: on a PC you already have, preferably with NVIDIA graphics
• External closed server: a GPU server for you alone, behind a VPN, operated by Kodetank
• Single-user or company licence – every feature in both, and 30 days free
• Clients who use Kodetank to bring AI into everyday work pay no licence fee for as long as they use our services – hardware and server rental are charged separately

See pricing (in Norwegian) ↗

What is on-premise AI?

On-premise AI means the language model – for example an open model from Meta, Mistral or Google – runs on a machine you control, instead of at OpenAI, Anthropic or Microsoft. The model is connected to your own documents and systems, and every question and answer stays within your four walls. Today's open models are good enough for most real tasks: summarising, searching your own documents, drafting text, classification and extracting structured information.

When the need is something other than documents

Lokal KI covers the document work. If on-premise AI is to be connected to line-of-business systems, used in your own workflows or run on hardware you already have, we set it up as a project:

• Choice of model based on your tasks – not the biggest, but the right one
• Hardware: what is needed, and what you already have that can be used
• Secure connection to your own documents and systems (RAG, integrations)
• Access control and logging – you decide who uses what
• Training for the people who will use and run the solution

On-premise or cloud – an honest comparison

Cloud models are still stronger on the very hardest tasks, and they require no hardware of your own. Local models give full control, predictable cost and no data leakage – but they require setup and a minimum of computing power. Many choose a combination: on-premise AI for everything sensitive, cloud for what is not. Lokal KI deliberately has no cloud, not even as an option. Where the line should go for everything else, we help you draw.

How we get started

Try the demo first – it shows the whole service with fictional documents. Then we have a conversation about which documents and tasks are involved, and set up Lokal KI on your own documents, with 30 days free. If you need more than the product covers, we start with a well-defined pilot, so you see the quality before investing further.