I'm writing this from a hackerhouse in Barcelona ☀️
Last week we launched OpenViktor (an open-source AI employee built in 48h). We were #3 PH of the day, 300+ GitHub stars in 24h. We took it down and rebuilt the entire thing from scratch.
Meet Jared. The first AI employee that's actually social.
He lives in Slack, connects to 10,000+ tools and does the work: reports, dashboards, code, follow-ups, research.
But here's what's different: he reads the room. Jumps into conversations when it matters. Knows who to talk to and when to shut up. Brainstorms with your team. Remembers everything and gets sharper every day.
Paying $2000/month for an AI employee is crazy.
Humalike is backed by the first investor in ElevenLabs. Built by a small team (🇪🇸 X 🇵🇱) that hasn't slept much.
Martí, co-founder.
P.S. We ship fast. Request a feature or report a bug, we'll build or fix it the same day :))
@mcarmonas Finally, a tool that solves real problem. but how does Jared decide when to 'read the room' vs. engage?
@yisak_mebrate love to hear that Yisak :))) We are building humalike.ai, humanlike social agents for group environments with natural timing, interruptions, and voice/text cues. We basically plugged in Humalike to Jared and made it social!
@mcarmonas For our company, we came up with Kevin. He's our internal bot. He's not in Slack, though, he's in Telegram. And yeah, we had to spend about three weeks training him. We're still training him, actually. But it's a cool idea. And of course, at first, he was super annoying. So, we're definitely doing this, it's definitely the future. And an amazing tool that helps you train them would be perfect.
@mcarmonas @slavaakulov Interesting! How is Kevin doing right now? Our goal with Jared is to provide very good default that fits 80% from day 0 and then learns the details of specifc organization.
@curiouskitty Jared reads the room. Jumps into conversations when it matters. Knows who to talk to and when to shut up. Brainstorms with your team.
The rest? We are as good as the rest.
@mgufrone Hi! There are two main parts to this:
1. very good base -> we spent a lot of time testing and tweaking it to make it work. It's based on personality, context, recent tasks, goals, relationships between users
2. He learns -> when he does a mistake he takes conclusions, and learns to not repeat it again, getting better and more aligned to your team as time passes
Thanks for trying it out!
@konstantinalikhanov Hi Konstantin! He jumps in only if we can provide value without being annoying. The decision is made based on his personality, context, memory of past interactions and how much he can actually help. He can think in the background and join conversation when he already has something to show.
It works on all channels that you invite him, so private channels work. He also works in DMs.
Hope that helps:))
The "reads the room" part is what I find most interesting. Proactive agents are way harder to get right than reactive ones - getting the timing wrong and it becomes more annoying than helpful. How do you decide when Jared should chime in vs stay silent? Is it rule-based or does the model decide?
@mykola_kondratiuk It's a combination of LLM judgement, context, memory (he learns over time) and our know-how. The main point is he only joins if he can provide value.
@amraniyasser Hey Amrani! It does adapt to how the team works, talks, behaves and we are improving it (atm) to understand lore of teams :))
@khashayar_mansourizadeh1 He's great at making reports and coding obviously. But he's exceptional at longer ongoing tasks that require coordinating stakeholders. For example you can ask him to "make sure we have all marketing materials ready for PH before friday" and he learns what is needed for PH, makes a list, ask questions to clarify. Then he checks on it during the week, pings people responsible to ask how is the progress going and makes sure everything is ready by friday:)
He has his own email, browser, computer and 10k+ tools available so he can be helpful in hundred different ways:))
@mateusz_jacniacki Nice, curious, how it has 10K+ tools? I assume you haven't done 10K+ complete integrations, it would be even hard to say it has access to 10K+ API docs, and then how it would handle Auth, Keys, etc. So very curious to know how it would work :)








Humalike
tysm Amrani!