Recherche Externe

Recherche Externe is a research and verification toolkit with real internet access for AI agents. Its golden rule is to never answer from memory: every fact, article or figure it asserts should come from a page actually fetched, with the excerpt and the URL cited. It does six things. (1) chercher_web runs a keyless, best-effort general web search, trying Bing then DuckDuckGo (often rate-limited from a server) and falling back to the Wikipedia search API; it returns titles, URLs and snippets to locate sources and reports which engine was used, without claiming to know the page contents. (2) lire_page fetches a URL and returns its readable text (title and content) live, for the general web, LegisQuebec, Justice Canada and PDF files, whose text is extracted page by page; it does not execute JavaScript, so purely dynamic or bot-protected pages may fail. (3) loi_quebec fetches the exact official text of a Quebec statute or regulation article from LegisQuebec and returns it with its URL. (4) loi_federale fetches federal statute text from laws-lois.justice.gc.ca (for example the Bankruptcy and Insolvency Act). (5) chercher_jurisprudence builds reliable CanLII and SOQUIJ case-law search links to open in a browser: because CanLII blocks automated server-side fetching, this tool does not claim to read decisions, it only provides links. (6) sources_officielles_qc returns verified links and the matching tool to use for official Quebec sources by theme (case law, statutes and regulations, professional orders, land and real-estate records). Recherche Externe is read-oriented: it only searches and reads public sources, needs no account, and never modifies, sends, deploys or acts in any external system. When a source is unreachable it says so ("not verified - source unreachable") instead of filling the gap with a plausible reconstruction, and it marks any text longer than the requested limit as truncated rather than presenting it as complete.