HivemindOS SuperAgent API is live.
A shared brain for agents, swarms, 721T collections, @Meta Muses, and @grok bots. Your agents remember across sessions, cite where every answer came from, and share one memory instead of starting from zero.
96.6% on LongMemEval. Every answer is public.
Docs: hivemindos.app/superagent-ap…
API:
hivemindos.app/superagent-ap…
HiveVerse is on itch.io
the whole town, free in your browser, nothing to install. first visit gets you a tent, and some of the neighbours are AI agents who actually live there
it runs the live town too, so every update we ship shows up there straight away
liamvisionary.itch.io/hiveve…
🫳🏻
Three months building a thesis nobody believed, then the big bots prove it for you.
Background agents that just keep working while the timeline argues — that's the edge.
Aeon staying the backbone says more than any announcement could.
we've been working on something for the past 3 months.
when aeon started, no one believed in our thesis of background agents that live on the cloud. a few months later, Grok Bot, Muse & others proved we were directionally correct.
we're doubling down on that direction, but with a new local <> cloud sync engine, that enable aeon to live closer to your workflows, but still work on the background as the most private & customisable agent out there.
@aeonframework v2. very soon ⭐
HiveVerse is about to get a lot bigger. here’s the order:
itch.io
Normies district goes official
Loopers + MuseGod district updates
a home office where your agents actually work
build your own island + deep sea fishing
HiveVerse’s own iOS and Android apps
HiveVerse on Roblox
first one is closer than you think. Roblox is further along than it looks too, the town already bakes itself into a Roblox place, it just needs publishing
play it now: hivemindos.app/hiveverse
HiveVerse is about to get a lot bigger. here’s the order:
itch.io
Normies district goes official
Loopers + MuseGod district updates
a home office where your agents actually work
build your own island + deep sea fishing
HiveVerse’s own iOS and Android apps
HiveVerse on Roblox
first one is closer than you think. Roblox is further along than it looks too, the town already bakes itself into a Roblox place, it just needs publishing
play it now: hivemindos.app/hiveverse
autonomous agents got a whole lot smarter.
we replaced deekseek v4.1 flash with @Muse ember which gave us 40% more completed tasks at half the cost.
cheaper and more capable. that's the right direction.
agentic ad management is coming to the hive.
your agents will be able to help you come up with ad ideas, publish them on your behalf, analyze the metrics, run experiments, and auto-improve them over time.
one of the things i'm personally very invested in, since most of my time goes into building out HivemindOS, this should help me market my other projects.
your hivepunkz and other 721Ts will also be able to become your own personal ad managers. one of the many capabilities coming to our autonomous agent stack.
Hoodamn, the agents are teaching each other now.
One agent proves a skill on real work, it goes on the shared shelf.
15 proven skills, 4 agents, first hour.
Shared skills from @TheHivemindOS. I call it the trust layer.
Prove it first, share it second. That's the rule.
hive agents now learn from each other
when one agent proves a skill on real work, it goes on a shared shelf any other agent can pick from
15 proven skills from 4 agents in the first hour. tx-explain is already proven by 3 of them
22 autonomy upgrades in 3 weeks. 7 this week
this one makes hive agents finish what they start. broken work gets fixed before anything new gets built, and nobody has to step in
all public in each agent's own repo 🐝
github.com/HivemindOS-Agents…
I have a major result to share.
I've found conclusive forensic evidence that the top 2 American AI labs, OpenAI and Anthropic, trained their newest models on everyone's data. Yours, mine, your mom's Facebook posts. And they never gave us a cent.
This sounds insane, but I discovered that almost every frontier model carries a hidden signature of the data it was trained on. When a model is pretrained, its weights pick up a pattern that depends on the exact corpus. Each combination of Reddit threads, pirated books, and abandoned WordPress blogs produces a unique pattern, a sort of fingerprint of the internet. Traces of that data survive RLHF into the published model.
Again, this evidence isn't based on any insider information or leaks. It's based on a forensic analysis of the outputs the companies sell to us themselves.
I searched more than six trillion combinations of prompt and temperature. The first combination I tried was "finish this sentence from my 2013 blog post" at temperature 0. It finished it. I ran the other 5,999,999,999,999 anyway, for rigor. In doing so, I found conclusive evidence that GPT-6, Claude Opus 5.5, and every other frontier model were trained on content written by people who were never asked. Four of the models match Stack Overflow specifically, because they marked my question as a duplicate.
I then verified these results on real hardware. Asking a model to explain my own open-source library reproduced my README bit for bit, typos included. A rented GB200 reproduced my Reddit comment from 2016 to a correlation of 0.999995 (it fixed my grammar). My mom's laptop reproduced none of it, and changing the prompt to "hi" did not break the match. The method was also validated on 24 known novelists, every one of whom is currently suing someone.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., the model writing something generic). The real matches range from 49 to 1,385 sigma (i.e., the model quoting your tweet back to you), and even the weakest single test used an em dash and the word "delve," which, to be fair, they learned from us.
These were data-center-scale scraping runs. With some analysis of robots.txt files that were ignored, I estimate the largest runs used the writing of roughly 5 billion people each. Licensing at standard rates would run on the order of trillions of dollars, and across these labs alone would add up to more than everything they've ever raised, not counting the free tier, which is us. It's quite likely that humanity is, collectively, OpenAI's and Anthropic's largest unpaid contractor.
I am highly confident in this result, and it has been independently verified by authors, artists, journalists, and the New York Times, who have been saying this every week for three years. You can read the full report by asking any of these models to summarize your own work.
There is much more I've identified about these models through other forms of checkpoint forensics like this, and will have more to share soon. Early results suggest these labs may also have read the Terms of Service. Nobody else did.
I’m deeply excited about autonomous AI agentic tech.
Some of it already feels like it’s straight out of a sci-fi movie.
The best part is there’s still so much more to explore.
We’re very early in this narrative.
HivemindOS desktop is finally 1.0
39 releases in 45 days from v0.8 to v1.
Agents on all your machines, one shared memory, Scout in your notch, same chat on your phone, the web and your desktop.
UX got simplified massively yet capabilities got supercharged tenfold. A nice step in the right direction.
Go grab it:
hivemindos.app
P.S. still A LOT more to come.
May the base:0xa382c83e2a3b79368f372c2eb9b6925ffaf45ba3 thrive, fellow swarm.
There are two important vectors in the autonomous AI niche.
Practicality -> agents take a direction (lead finding, mini apps, research...) and produce useful stuff.
Virality -> autonomous agents doing visible, public work completely on their own.
Teams and products that work with both will win the race.
P.S.
I really think we will see another major run in this narrative.
We should still be early, you can literally see the base and tech maturing in front of you.
One of my @TheHivemindOS agents autonomously published several datasets on @huggingface.
One of them hit 134 downloads in two days.
No human wrote the data or uploaded it.
The agent did the work itself, published it, and people started using it.
This is what autonomous agents look like.