Smart reply
Stay organized with collections
Save and categorize content based on your preferences.
With ML Kit's smart reply API, you can automatically generate relevant
replies to messages. Smart reply helps your users respond to messages quickly,
and makes it easier to reply to messages on devices with limited input
capabilities.
iOS
Android
Key capabilities
- The smart reply model generates reply suggestions based on the full context
of a conversation, not just a single message. This means the suggestions
are more helpful to your users.
- The on-device model generates replies quickly and doesn't require you to
send users' messages to a remote server.
Limitations
- Smart reply is intended for casual conversations in consumer apps. Reply
suggestions might not be appropriate for other contexts or audiences.
- Currently, only English is supported. The model automatically identifies the
language being used and only provides suggestions when it's English.
How the model works
- The model uses up to 10 of the most recent messages from a conversation
history to generate reply suggestions.
- It detects the language of the conversation and only attempts to provide
responses when the language is determined to be English.
- The model compares the messages against a list of sensitive topics and
won’t provide suggestions when it detects a sensitive topic.
- If the language is determined to be English and no sensitive topics are
detected, the model provides up to three suggested responses. The number of
responses depends on how many meet a sufficient level of confidence based on
the input to the model.
Provide feedback
Due to the complexity of natural language processing, the suggestions provided
by the model may not be appropriate for all contexts or audiences. If you
encounter inappropriate reply suggestions, reach out to
ML Kit support. Your feedback
helps to improve the model and the filters for sensitive topics.
Example results
| Timestamp |
User ID |
Local User? |
Message |
| Thu Feb 21 13:13:39 PST 2019 |
|
true |
are you on your way? |
| Thu Feb 21 13:15:03 PST 2019 |
FRIEND0 |
false |
Running late, sorry! |
Suggested replies
| Suggestion #1 |
Suggestion #2 |
Suggestion #3 |
| No worries |
😞 |
No problem! |
Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates.
Last updated 2025-08-29 UTC.
[null,null,["Last updated 2025-08-29 UTC."],[[["\u003cp\u003eML Kit's Smart Reply API automatically generates relevant replies to messages, aiding quick responses and assisting devices with limited input.\u003c/p\u003e\n"],["\u003cp\u003eThe on-device model processes conversation history locally to provide quick replies without sending data to a remote server, prioritizing user privacy.\u003c/p\u003e\n"],["\u003cp\u003eSmart Reply is designed for casual English conversations and may not be suitable for all contexts; feedback on inappropriate suggestions is encouraged for model improvement.\u003c/p\u003e\n"],["\u003cp\u003eThe model analyzes up to 10 recent messages, identifies English language, filters sensitive topics, and offers up to 3 suggestions based on confidence levels.\u003c/p\u003e\n"]]],[],null,["With ML Kit's smart reply API, you can automatically generate relevant\nreplies to messages. Smart reply helps your users respond to messages quickly,\nand makes it easier to reply to messages on devices with limited input\ncapabilities.\n\n[iOS](/ml-kit/language/smart-reply/ios)\n[Android](/ml-kit/language/smart-reply/android)\n\nKey capabilities\n\n- The smart reply model generates reply suggestions based on the full context of a conversation, not just a single message. This means the suggestions are more helpful to your users.\n- The on-device model generates replies quickly and doesn't require you to send users' messages to a remote server.\n\nLimitations\n\n- Smart reply is intended for casual conversations in consumer apps. Reply suggestions might not be appropriate for other contexts or audiences.\n- Currently, only English is supported. The model automatically identifies the language being used and only provides suggestions when it's English.\n\nHow the model works\n\n- The model uses up to 10 of the most recent messages from a conversation history to generate reply suggestions.\n- It detects the language of the conversation and only attempts to provide responses when the language is determined to be English.\n- The model compares the messages against a list of sensitive topics and won't provide suggestions when it detects a sensitive topic.\n- If the language is determined to be English and no sensitive topics are detected, the model provides up to three suggested responses. The number of responses depends on how many meet a sufficient level of confidence based on the input to the model.\n\nProvide feedback\n\nDue to the complexity of natural language processing, the suggestions provided\nby the model may not be appropriate for all contexts or audiences. If you\nencounter inappropriate reply suggestions, reach out to\n[ML Kit support](/ml-kit/community). Your feedback\nhelps to improve the model and the filters for sensitive topics.\n\nExample results\n\nInput\n\n| Timestamp | User ID | Local User? | Message |\n|------------------------------|---------|-------------|----------------------|\n| Thu Feb 21 13:13:39 PST 2019 | | true | are you on your way? |\n| Thu Feb 21 13:15:03 PST 2019 | FRIEND0 | false | Running late, sorry! |\n\nSuggested replies\n\n| Suggestion #1 | Suggestion #2 | Suggestion #3 |\n|---------------|---------------|---------------|\n| No worries | 😞 | No problem! |"]]
