Oscar Lauge Hoffmann · Founder of lovguiden.dk · Copenhagen
AI agents that take over recurring work, and show their sources.
The technology behind Lovguiden, the Danish legal service used by more than 580,000 people in the past year, built around a company's own data and workflows.
- 01OKCollectedCourts, parliament, 25 ministries and around 220 agencies
- 02AgentRepairedOne source changed its website. The agent patched and verified the fix.
- 03OKReadScanned PDFs turned into searchable text
- 04OKStructuredSections, paragraphs and cross-references linked
- 05LocalSummarisedClassified and summarised by local models
- 06OKPublishedTo the website, to AI tools via MCP and to the API
- 51,820
- Monthly visitors
- 400+
- Companies & organisations
- 2,000+
- Public sources, kept current
- 138K+
- Verdicts & decisions
I'm very impressed with Lovguiden. It's a tool that has been in demand for a long time. Limiting it to legislation, verdicts and decisions makes the AI search far more trustworthy than the chatbots we know.

Søren Engers
VAT Director, Baker Tilly
A product that creates overview and is intuitively simple to use is something I have missed for about 40 years. Lovguiden is the product that today comes closest to giving me that overview of tax legislation.

Ole Aagesen
Lecturer, CBS, and author
The database gives quick access to legislation, from bill to enactment, along with a large number of rulings and decisions. The chatbot quickly gives an overview of the relevant legislation.

Reza Ahmadian
Lawyer
Systems that run by themselves.
The same technology that runs Lovguiden, built around a company's own data and workflows. Every system is designed to run without supervision and to show where its answers come from.
Agentic automation
Recurring work handed to AI agents that run on a schedule: collecting data, checking it, writing and reporting. Every run is logged, and when a source changes, the system diagnoses the fault and repairs it.
- Scheduled agents and data pipelines
- Monitoring, alerts and self-repair
- Runs locally or in the cloud
AI assistants on company data
Search and an AI assistant across internal documents, rules, contracts or case files. Answers come from that material only, and every reference is checked before anyone sees it.
- Answers with verified sources
- Local models when data can't leave the building
- Search tuned to the domain
Company data inside AI tools
MCP servers and APIs that bring internal data into Claude, ChatGPT, Copilot and custom agents, so AI tools work with what the business actually knows. AI assistants such as ChatGPT, Copilot, Gemini and Claude already send around 600 visitors a month to Lovguiden.
- MCP servers with access control
- Documented REST APIs
- Usage-based billing for paid access
AI-run operations
An existing business rebuilt so the routine runs itself: website, bookings, invoicing and back office, plus agents that write content and campaigns every day.
- Website, back office and payments
- AI newsroom and campaigns
- Migration alongside the old system
Five products, built and operated end to end.
Fully automated and running locally.
All of Lovguiden's data is collected, processed and updated by a purpose-built system that runs locally. When a public website changes and a data source stops working, an AI agent repairs it and verifies the fix. The figures below are based on the system's own measurements.
Around 40 jobs run on a fixed schedule, day and night.
Connections, database and storage are checked before each run.
Jobs run in parallel within set limits and recover after interruptions.
Every run is logged, and jobs that stop returning data are flagged.
An AI agent diagnoses a broken data source and prepares a fix.
A fix is only applied after it has been tested against the live source.
From first conversation to a running system.
- 01
Conversation
The work that takes too long is described, followed by an honest assessment of whether AI and automation can take it over, and roughly how.
- 02
Prototype
A first version on real data, so the result can be seen working before anything larger is committed to.
- 03
Build
The full system, built in short iterations with regular demos, so progress can be followed all the way.
- 04
Operate
Operated and monitored as a service, or handed over with code and documentation to an in-house team.
Questions
What does it cost?
It depends on the scope. A concrete proposal follows the first conversation, and the first step is kept small.
Is it only for legal tech?
No. Lovguiden is legal, but the methods apply anywhere with many documents, many sources or repetitive work. opkurser.dk is a course business.
Can data stay in-house?
Yes. Lovguiden's data processing runs on local models, and the same setup can run on a company's own infrastructure, so sensitive data never has to leave it.
What happens when something breaks?
The systems are monitored and repair common faults on their own, such as a data source changing its format. Anything they can't fix is flagged.
Who owns what is built?
It is agreed up front. The full source code and documentation can be handed over.
Which languages?
Danish and English.

Founder and engineer, based in Copenhagen.
Oscar Lauge Hoffmann founded Lovguiden in November 2023 and has developed and run it since. It is now used by hundreds of thousands of people a year, and its data operations are fully automated.
The work covers the whole product: data collection, search and AI, the interface and billing. The focus is on systems that run reliably without supervision and can show where their information comes from. Alongside Lovguiden, he builds the same kind of systems for other businesses.
Automation
Recurring work is automated, and the systems are built to detect and repair their own faults.
Runs locally
Data processing runs locally, which keeps costs predictable and the data under full control.
Accuracy
AI output is grounded in source material, and every reference can be checked.
End-to-end ownership
Products are taken from the first data source to the finished interface and the invoice.
Latest note · Nov 2, 2026Hybrid search in Postgres without pg_search
Get in touch.
For recurring work that takes up too much time, or data nobody has the hours to go through. A short description of the task is enough to start the conversation.
Danish or English
- 01The taskThe recurring work that takes too long.
- 02How it is handled todayWho does it, and with which tools or sources.
- 03How oftenDaily, weekly, or every time a document arrives.
The email button opens a draft with these three prompts in place.




