I built Emote so AI agents can react to messages
It's been a while since I posted here. I'm back to building and sharing products, and wanted to share what I've been working on: Emote.
I react to messages all the time. Sometimes a heart or a laugh is all a message needs. I wanted AI agents to have that option too.
Emote is a reaction API for AI agents. Send it a message, some conversation history, and your agent's system prompt. It returns a reaction from the options you provide, or none if a reaction isn't needed.
You choose its personality and the reactions it can use. Those can be emojis or custom reactions for your own chat UI or platforms like Telegram, iMessage, or WhatsApp.
You keep your agent and the model it already uses. Emote runs in parallel to choose a reaction.
You can try the playground on the homepage—no signup needed. If you want to add it to your agent, you get 1,000 free API requests with no credit card required.
If you're building a chat product, what do you think? I'd love to hear where you'd use reactions, or where they'd feel out of place.
This is useful. How are you finding your first users so far?
I did a launch on X, ProductHunt, Reddit, and here. I only got 1 user sign up so far, not sure where they came from.
What would convince you reactions are becoming a meaningful part of an agent's behavior rather than a feature developers enjoy testing but rarely keep enabled in production?
I liked how meta's muse reacted to my messages, felt a lot more natural than most agents I've talked to, so wanted to make this easy for anyone to develop
That explains why you built it, but not whether it sticks. Are any developers actually keeping reactions enabled after the initial test?
Nice work shipping it. What has been the biggest challenge since launch?
Mostly just distribution, made a lot of posts on different sites, so got a decent number of eyes on it already, but only 1 sign up. I think it just needs more eyes though, especially to the target demographic
The interesting part is the none option, and I would make that the thing you benchmark publicly. Choosing a reaction is easy and almost any model does it acceptably. Knowing when not to react is the hard problem, and it is the difference between something that feels alive and something performing enthusiasm at you. A published number, like how often Emote correctly stays silent across a held out set of conversations, would say more than the personality configuration does. It is also the failure people notice first, because an unnecessary reaction is more irritating than a missing one.
yaa very true, when testing the product myself that was what I got annoyed by the most, tuned it so that "balanced" felt good to me. We also have a "reserved" setting if you want your agent to react less often