Examples
Twelve workflows, in four groups. Each one is a single request; open it in the builder and change what you need.
AI under GDPR
Use any model while personal data stays out of the prompt, or goes only to a model in the EU.
Your support assistant runs on a US model, but names, emails and IBANs stay out. The answer comes back with the real values.
Comes back: The model's answer, with the real names back in place.
A prompt with personal data goes to Mistral in the EU, tokenized. A clean prompt goes to the model you prefer.
Comes back: An answer from the EU model whenever personal data was found.
Your code sets high_security for a customer whose contract requires EU processing. Their prompts go to a model in the EU, tokenized.
Comes back: Answers from an EU model only, for that customer.
Attacks and abuse
Stop what attacks your model or your users before it gets through: injected instructions, jailbreaks and abuse.
Hidden instructions and jailbreak attempts are blocked before your model reads them.
Comes back: A block with the reason, before your model reads a word.
Web pages, emails and documents can hide instructions for your agent. They are blocked before the text reaches the prompt.
Comes back: A block, before the page reaches your agent's prompt.
Hate is blocked. Abuse and threats get flagged for a moderator, in 23 languages.
Comes back: Three scores per comment. Hate blocks; abuse and threats get flagged for a moderator.
Data hygiene
Keep personal data out of logs and datasets, and turn free text into records your systems can store.
Log lines and tickets reach your warehouse or monitoring tool without names, emails or keys.
Comes back: The same log line, minus names, emails and keys.
Analysts get the records with names and emails swapped for tokens. Only you hold the map to reverse it.
Comes back: The records with tokens in place, and a map only you hold.
An order or complaint email becomes an order number, a problem and a date, ready for your CRM.
Comes back: The order number, the problem and the date, as JSON.
Language and content
Know what a text is: its language and topics, with a title and a description for search and feeds.
Tickets reach the right team by language and topic, without keyword rules, and without personal data in storage.
Comes back: The language and the topics, ready to send to the right team.
Index contracts and emails without personal data in the text, each with a title and a description for search.
Comes back: The redacted text, plus a title and a description read from the original.
Each post or notification gets a title, its topics and an emoji that fits.
Comes back: A title, the topics and an emoji for each post.
Pick an example, change what you need, send one request, and read the result before you finish this page.
