Use cases / Teams building chatbots and copilots
Ship the chatbot. Keep the IDs.
Models hosted in another country are often the best tools you have, and customer messages are full of identifiers. The gateway lets you use both, with one change to your base URL.
What changes
Today
With Onshore
Prompts
Raw customer messages go to the provider
Identifiers are tokenized first
Replies
No way to refer to the real value
Tokens are restored before your app sees them
Streaming
Filtering often means giving it up
Replies still stream
Oversight
No record of what was sent
Counts by data type and destination
What you'd use
Questions
Which models work?+
Any that speak the OpenAI-compatible API.
Will answers get worse?+
Tokens and placeholders keep their type, so the model still knows it's an ID number. Stand-in names read naturally. Test it on your own prompts.
Does it work in other languages?+
Formats such as IDs, IBANs and phone numbers are found in any language. Cue words and names have been tuned mostly on English, so test it on your own messages, and add cue words of your own.
Start with one leak.
Tell us which flow worries you most. We'll show you what it would take to keep it onshore.