Law as structure, not text
We convert legislation into a form in which the system knows the hierarchy of provisions, the references between them and how they change over time.
knowledge graph · versioned wording
AI for Czech law
Lexiqon searches statutes and case law, answers in plain language, and attaches the exact provision, in the version in force, to every statement. When it finds no source, it gives no answer.
Question
What is the limitation period for a claim for damages under the Civil Code?
Answer
The limitation period is three years1 and runs from the day the injured party learns of the damage and of the person liable to compensate it.2 The claim becomes time-barred no later than ten years after the damage arose, or fifteen years if it was caused intentionally.3
§ 1How it works
Lexiqon takes over the routine of research and checking. Decisions and responsibility stay with you.
Type your question in plain Czech or English, or upload a document from the case file.
It finds the relevant provisions and decisions in a verified base and traces the links between them.
eSbírka · NALUSIt prepares an answer or a draft that follows your firm’s templates and conventions.
Firm templatesYou see every statement with its source, and proposed edits against the original text.
Redlining§ 2Comparison
General-purpose language models write convincingly, but without grounding in sources. They make up sections, confuse repealed and current wording, and cannot tell when they are unsure.
| General AI chatbot | Lexiqon | |
|---|---|---|
| Where the answer comes from | Whatever the model memorised in training | A verified base of legislation and case law, with a citation for every statement |
| Version in force | Often mixes old and current wording | Uses the wording in force on the date you specify |
| Structure of the law | Reads text as a stream of words | Knows the act – section – subsection hierarchy and links to case law |
| When unsure | Answers anyway | Says it found no support and gives no answer |
| Where your data lives | With a foreign provider, outside your control | In a separated environment, in the cloud or on your premises |
| Language and legal system | Mostly English and common law | Built for Czech and the continental legal system |
§ 3Technology
The language model is just one layer. Accuracy comes from what surrounds it.
We convert legislation into a form in which the system knows the hierarchy of provisions, the references between them and how they change over time.
knowledge graph · versioned wording
It finds provisions that are related in substance, not just the ones that repeat the same words.
hybrid RAG · vectors + graph
Before you see an answer, the system confirms that every statement has a source, matches the wording in force and reaches the required level of confidence.
guardrails · citation verification
The language model can be swapped. You are not tied to one vendor or one cloud.
model-agnostic architecture
§ 4Security
Your clients’ data stays in the environment you choose.
Each firm works in its own isolated environment, with role-based access control.
Case files and queries are never used to train models, ours or anyone else’s.
Use it as a service with no infrastructure investment, or as a fully isolated installation on your premises.
You can trace which sources an answer came from, in which version, and when.
§ 5FAQ
Not in the usual sense. It does not answer from a language model’s general knowledge, but from a verified base of legislation and case law. Every statement has a source you can open.
Official sources of Czech law — legislation from eSbírka, Constitutional Court decisions from the NALUS database. It also works with historical versions, so it can answer as of a given date.
It says so. If it finds no support in the sources for a statement, it gives no answer and shows what it did find.
No. It takes over research and checking, so you have more time for strategy and clients. Decisions and responsibility stay with you.
No. Documents and queries stay in your separated environment and are never used to train models.
A short online meeting where we take a question from your practice from input to a sourced answer. You do not need to share any client data.
§ 6Contact
Tell us what you work on and what you would like to try. We will reply by email and suggest a time.
Or write to us directly