On-device text models
Models that read what someone types and act on it in the same keystroke: emoji, topics, language, personal data, structure.
Text models run on every keystroke and every message, with no round-trip and no per-call cost, so a feature can check everything instead of the messages you can afford to check. The text never leaves the device.
Available models
On-device redaction that strips personal data as you type, in 27 languages, so it never reaches your servers or logs.
Type a task or a message and get the emoji that fit, in 22 languages, on the device.
On-device topic tagging that reads a title or a post and tells you its topics, in 101 languages, with no server call.
On-device language identification that names the language of a few words, across 84 languages, in tens of microseconds.
On-device titles and descriptions: a short factual title and a one or two sentence description for any passage, on Apple silicon.
Available soon
Extract typed fields from any text against your own schema, on the device, with a field left empty where the text says nothing.
On-device moderation that flags hateful, abusive and threatening text in 23 European languages, before the message is sent.
Pricing
Every model is free up to 100k monthly active devices per SDK. Unlimited inference per user.