LexielBot

Who we are

Lexiel is an AI legal assistant for lawyers, built by Codelabs Studio, S.L. (Alhaurín de la Torre, Málaga, Spain). We use what LexielBot reads so that lawyers can check that a statute or a judgment exists and says what is claimed, instead of trusting an invented citation. When we visit a portal we identify ourselves with this string:

User-Agent

What it reads

Only official, publicly accessible sources:

• BOE (Spanish Official State Gazette): consolidated legislation and published provisions, through its open-data API.
• CENDOJ (Judicial Documentation Centre of the Spanish judiciary): public court decisions (Supreme Court, National Court, high courts of justice and provincial courts), and the one-off check of a citation that a lawyer asks us to verify.
• DGT (Directorate-General for Taxes): published binding tax rulings.
• Constitutional Court: published judgments, orders and decisions.
• Other official sources: EUR-Lex and official portals of Latin American countries, only to check public references.

We do not log in anywhere, bypass paywalls or identification, solve captchas, or collect personal data of those portals' users. What we read is legal text the authority already publishes.

How we read

There are four different ways and in all of them your server sees an identifier saying who we are. We never pretend to be a browser or another robot.

• Automatic crawling: the scheduled reading of new material. It identifies itself as LexielBot, checks robots.txt before asking for anything and respects the pauses and limits below.
• One-off lookup requested by a lawyer: when a lawyer asks us to check one specific citation (a statute, a judgment, a legal thesis). It identifies itself as LexielBot, makes one or two requests about that reference and does not request anything robots.txt forbids. If the portal rejects the lookup (for example with a 403 error) we do not insist or disguise who we are: we tell the lawyer that the citation could not be checked at that moment.
• A page a user pastes: when someone pastes a web address for us to import, we request only that page, once, at their request and without following links. That request identifies itself as Lexiel-AI/1.0 (legal assistant; +https://lexiel.ai) and is not crawling.
• Automated browser (an exception we are removing): a few readings that only work with a browser, such as checking a bar registration number in the legal profession's census or some manual loads from Latin America. They identify themselves as LexielBot, but they do not yet check robots.txt.

How fast

• CENDOJ, the most delicate portal: at most 40 requests a day in total across all our systems, of which at most 20 are document downloads (below the daily per-IP threshold the portal itself applies). Of that total, the automatic reading of new decisions may only use 28 requests and 15 downloads; the rest is reserved for one-off checks that a lawyer asks for.
• Between two automatic requests to CENDOJ we leave at least 6 seconds, above the 5-second Crawl-delay declared in its robots.txt. We do it once a day, at night.
• If a portal answers with a 403 or 429 error, or redirects us to a captcha, we stop that same day and do not retry. We wait at least until the next day and, if it happens again, we double the wait (2, 4 and up to 7 days). We never change IP address or use other machines to get around a limit.
• On the other portals (BOE, DGT, Constitutional Court and so on) we set a ceiling of between 1 and 4 requests per second, depending on the portal. It is a maximum we do not exceed, not our usual pace: in practice we are far below it, because almost everything is read once and stored.

What we respect

• Each portal's robots.txt file. In automatic reading we download it before asking for anything (and refresh it every 24 hours), we do not request any path it forbids, we wait for the pause (Crawl-delay) it declares when that is longer than ours, and if the file cannot be read we do not crawl that portal until it answers again. CENDOJ has its own regime described above, and there we apply our own rules, stricter than its pause.
• Automatic reading versus a one-off lookup: automatic reading (the scheduled sweep for new material) does all of the above. When a lawyer asks us to check one specific citation we make a one-off lookup on their request, which is not crawling: we still do not request anything robots.txt forbids, but we do not wait out long pauses or stop checking because the file does not answer.
• Each portal's usage limits, including daily per-IP thresholds.
• Any warning from the portal: when blocked, we stop.

How to ask us to stop

If you run one of these portals and want us to slow down, change our schedule or stop reading it, write to contacto@lexiel.ai with the domain or IP address you have seen. We commit to replying and applying the change as soon as possible, without you having to block our address to be heard.

Controller

Codelabs Studio, S.L. (tax ID B88708797), Pasaje Federico García Lorca, 6, 3º E, 29130 Alhaurín de la Torre (Málaga), Spain.