Solutions

One problem, one page, and how it is solved

Each page explains a problem, the mechanism that solves it, the answer the API gives and how the work is split between the instant checks, the model and your rules. The examples on them are judged by the same engine the API runs.

• DSA compliance Every block comes back with the statement of reasons article 17 of the Digital Services Act asks for, in eight languages. Appeals decided by a person, and filing to the Commission's Transparency Database. →
• Text moderation Moderate comments, posts and messages with one call: fifteen categories instead of one toxicity score, allow, review or block, and the reason in a sentence with the words that triggered it. →
• Image moderation Check user images before they are published: nudity, gore and ID documents read by the model, the words in the picture run through the same checks as your comments, and a refused picture recognised when it comes back. →
• Conversation moderation Send the conversation and get a verdict on its last message, read in the light of who said what before it. Pile-ons, approaches to minors, scams told in instalments and lead types, with nothing about the thread stored. →
• Scam detection Recognise the shapes scams take on your platform, the flat that does not exist, the job that charges to start, the romance that ends in a money request, in eight languages, without filtering the people who report them. →
• Personal data redaction Find phone numbers, email addresses, bank accounts, card numbers and Spanish ID numbers in what your users write, and get the same text back with them masked, so the rest can still be published. →
• Spam detection Catch link spam, shortened links, chat invites, spaced-out domains, floods and the same message sent again and again. Free checks in about a millisecond, with the reason in a sentence. →
• Fake signups Check a registration's name, email address and bio in one call: throwaway providers, email-to-SMS gateways, Gmail aliases, domains with no mail server, placeholder names and bios made of links. Allow, review or block, with the reason. →
• Lead classification Every message and every conversation comes back with leads: eleven types, from a cold sales pitch to free work for equity, each scored from 0 to 1 and acted on only where your rules give it a line. →
• Prompt injection Check text before it reaches your model: families of attempt scored together, chat template markers, closed fences, disguised spellings and encoded blobs. Allow, review or block, with the reason in a sentence. →
• Unwanted topics Every answer measures how much a post is about betting, cryptocurrency or replicas, from 0 to 1. Nothing acts on it until your policy gives the subject a line, and somebody describing a gambling problem stays out of reach. →
• Multilingual moderation Insults, threats and scams caught in English, Spanish, Portuguese, French, Italian, German, Catalan and Dutch in about a millisecond, the model for every other language, and an honest answer about which one read your text. →
• Human review Allow, review or block. What lands in review waits in a queue for your moderators, through the API or the panel, and a signed webhook tells your site the moment somebody approves or rejects it. →
• Moderation policies Per-category thresholds, word lists, subjects and lead types in a versioned policy, or as rules in the call. Try a new policy in shadow mode on your own traffic before it decides anything. →