Est Law · concept & progress

A family lawyer
for every individual.

Est Law is an AI legal advisor that reads the actual law — every act, every provision — and answers ordinary people's questions with exact references. Estonia is the proof of concept.

Why

The idea

Wealthy people and companies have lawyers who know which rule applies, which authority decides, and where it is written. Most individuals don't. Est Law aims to give everyone that "family lawyer": someone who knows the whole legal system and can point to the exact paragraph.

Estonia is a great place to prove it: a small, fully digital state whose complete legislation is published in machine-readable form in Riigi Teataja, together with the EU law it implements. If it works for one whole country, it can be repeated for the next.

Foundation

The data

~66 GB

raw downloads of Estonian law (Riigi Teataja API + version-history dumps) and related EU law

~11 GB

processed, structured legal text

~9.9 GB

SQLite database with versions, act status and EU transposition links

Exact

token count stored for every provision and work unit, so each step knows precisely how much text it is handling

Downloads are resumable and checksum-verified; the database is built to update daily with compressed backups before each run.

How it thinks

Architecture: navigate, don't guess

No vector-search guessing. Like a good lawyer, the system walks down a map of the law, opening detail only where it is needed.

Your question"Can my landlord keep my deposit?"
1 · Categoriesa taxonomy of all acts by area of life and law
2 · Act cardsshort cards describing what each act covers — decides whether to open it
3 · Section mapssignposts without details, e.g. "this section lists tax rates in percent"
4 · Provision summariesextensive summaries of the actual content, with exact § numbers and authorities
5 · Full legal textopened only when the answer needs the exact wording
Answer with referenceswhat applies, who decides, and where it is written

Summaries that scale

Long texts are split at natural breaks (subsections), summarised in parts and recursively condensed. Length caps scale with the source: 1,200 chars up to 1,500 tokens, 2,000 above that, 3,000 above 3,000 tokens and 4,000 above 6,000.

Precision rules

Summaries keep exact § numbers and act / EU regulation numbers, and name which authority does what. Each item gets search words scaled to its size (10–60). Repealed heading-only provisions get a fixed empty summary.

Validated, unit by unit

An automatic validator checks keys, order, language, length and search words of every unit before it is marked done.

Built to stay current

A daily update pipeline fetches changes, re-imports changed acts and queues only affected summaries for rework.

Where it stands

Progress

Current production batch: b01-seadus — all Estonian laws (seadused), processed by 4 AI workers in parallel.

Work units summarised: 1,243 / 1,653 (75%)

Provisions/items summarised: 24,155 / 35,651 (68%)

Numbers as of 5 November 2026, 19:00 Tallinn time (EET), read live from the project's status tool.

Done so far

Lesson learned

Running summaries through general-purpose AI agent loops processed roughly 190× more tokens than the law text itself. The next pipelines call models directly to cut that cost dramatically.

What's next

Roadmap

Data & database ✓

Estonian + EU law collected, structured and kept up to date.

Provision summaries — in progress

Finish all laws, then regulations and other act types.

Act cards & section maps

The upper levels of the navigation map.

Q&A advisor

The conversational advisor that navigates the map and answers with references.

Next jurisdictions

Repeat the method for further countries, Finland / Åland first.