Tired of AI bots that talk over people and spam group chats? Humalike launches behavioral APIs to teach your AI some actual social manners.
Have you ever deployed an AI chatbot that was technically brilliant, capable of writing flawless code, only to watch it completely ruin the vibe in a Telegram or Discord group chat? It talks over people, spams relentlessly, and has the social intelligence of a cardboard box.
You are not alone. A sleepless, tiny team of developers from Spain and Poland just launched Humalike on Product Hunt, aiming to solve the exact pain point of socially tone-deaf AI.
A few months ago, the creators of Humalike built an AI community manager. The moment it joined a group chat, everyone instantly knew it was a bot. It wasn't because it lacked intelligence, but because it lacked basic social decorum. It kept interrupting people and never knew when to shut up.
Realizing that simply adding more features wouldn't fix a broken personality, they pivoted to build Humalike—a behavioral infrastructure layer for AI agents. They are shipping 7 clever APIs to teach your bots some manners:
The best part? It's model-agnostic and built specifically for group dynamics, not just typical 1:1 dry conversations.
The project quickly gained traction, and devs on Product Hunt wasted no time putting it to the test.
One developer, who builds voice AI to call elderly parents, shared a highly relatable struggle: "Older folks often pause mid-sentence to find their words. Standard Voice Activity Detection (VAD) always reads that brief silence as their turn ending, and the bot immediately cuts them off. How does your turn-taking handle long, uneven pauses?"
The co-founder admitted that this is a notoriously difficult problem that hasn't been solved perfectly by anyone yet. Currently, Humalike is tackling this on text and online chat first, while developing an end-to-end model for voice turn-taking that factors in the agent's personality, goals, and relationship dynamics.
Another curious dev pointed out the challenge of capturing "deleted drafts" or "removed reactions" in a stack-agnostic setup, as these events rarely leave the client. The team clarified that the platform forwards those events to your system, which you then pass to Humalike. Humalike's job isn't to fetch the raw event, but to interpret what a "removed reaction" actually means in that specific emotional context.
Let’s face it, we are currently living in an era where everyone is obsessing over raw LLM intelligence. But in the real world, a genius who acts like an obnoxious party crasher will get banned instantly.
If you are running self-hosted agents on your own cloud vps or customized infrastructure, adding a subtle social layer like Humalike might be the secret sauce to make your bot feel less like an automated script and more like a real teammate.
They are currently offering $20 in free tokens and have open-sourced an MIT-licensed Hermes Agent plugin to get you started without writing the integration from scratch. Give it a spin and see if your bot stops being the most awkward entity in the room!
Source: Product Hunt
