List countries covered: product/insurer/regulation counts, how many branches have at least one product (branches_covered), the size of the country's branch taxonomy (branch_taxonomy), and how many hand-written branch overview pages exist (branch_overview_pages).

List insurance branches for a country with their labels, mandatory flag and product counts.

Search wiki pages by title/branch/insurer and page content. Filter by type (product|branch|insurer|regulation|concept|moc), branch slug, insurer slug. Returns matches with path, source_url and freshness; when more than limit pages match, the response says how many were left out.

Return the full text of one knowledge page by its repo-relative path (from search results). Text files in the knowledge folders only (.md, .json, .yml, .yaml, .txt, .csv).

Return a product's structured data (coverages, exclusions, deductibles, etc.) as JSON, plus its source_url. product_name is matched case-insensitively (substring). Optional filters: document_type (conditions_generales | ipid | conditions_particulieres | conditions_tarifaires) and edition (substring of the edition_date, e.g. '2026'). When one product has several documents (e.g. its CG and its IPID share the commercial name), the general conditions with the newest edition are returned and the other documents are listed under other_documents.

Compact, question-oriented view of ONE product: only the coverages and exclusions relevant to topic (e.g. 'vol', 'degats des eaux', 'ski hors-piste'), with their limits, deductibles, verbatim quotes and page numbers. Prefer this over get_product when answering a specific guarantee question: everything needed to quote is in the response and nothing else. Matching is accent- and case-insensitive over coverage/exclusion names and descriptions. An empty match list means the topic was not found in THIS document; that is not proof the product has no such cover - say so and point to the source_url.

Verbatim evidence check for ONE factual claim about a product ('le ski hors-piste est couvert avec un moniteur', 'franchise de 250 EUR en vol'). Returns the document excerpts (coverages, exclusions, quotes, definitions, deductibles, summary) sharing the claim's terms, each with its page when stated. The tool does NOT judge: compare the claim against the excerpts. If a number, date, article or condition the claim relies on appears in no excerpt, treat it as NOT SUPPORTED by this document and say so. Use it as a self-check before asserting product facts to a user.

Compare 2+ products side by side on one dimension: coverages | exclusions | deductibles. Each name is matched case-insensitively against extracted product names; pass insurer_slugs (same length/order as product_names) to pin each name to one insurer. A name matching several DISTINCT products is refused with the candidates listed (refine it rather than let the server guess). For each resolved name the best document is selected (general conditions over IPID, newest edition) and identified in the response, including its superseded flag.

Flag CANDIDATE duplicate cover when combining 2+ products (e.g. a home policy + a family-liability policy). Each product's coverages (or exclusions) are tagged with a controlled category (schema/coverage_categories.json); a candidate overlap is a category present in 2+ of the products. Deterministic + heuristic: it surfaces likely duplicates for an agent to confirm against the actual descriptions. It does not advise or rank, and can miss overlaps the taxonomy doesn't yet cover. Pass insurer_slugs (same length/order as product_names) to pin each name to one insurer; the best document per name is selected (general conditions over IPID, newest edition).

Return the hand-authored overview page for a branch (by slug or label).