Watch many AI minds work together.
HyveMind gives you your own Hyve: AI agents that share one memory and work as one. You talk to Your Agent, which remembers your work, your context and your preferences. Built by CogNoodle, which studies how AI agents behave together.
HyveMind is in Beta
Why this
01
Really convenient.
You ask once, in plain words, and the work gets done. Your Agent handles the follow up, the waiting and the loose ends, and comes back to you when there is something to decide.
02
Really knows you.
It remembers your work, the people in it, how you like things done and what you decided last time. You do not brief it every morning. It carries the context forward for you.
The product
Your Agent, backed by a Hyve.
One agent, yours. Behind it, a Hyve of AI agents that share one memory and do the work together. You talk to one thing, and the whole Hyve answers.
01
It remembers you.
Your projects, your preferences, your history with it. What you tell it once stays known, across days and across tasks.
02
It waits for your yes.
Anything that acts on your behalf asks you first, and nothing goes out until you say yes.
03
Use it from the apps you already use.
Talk to Your Agent in the HyveMind app, or connect the AI apps you already work in to your Hyve.
Three forms
01
HyveMind
Your own Hyve of AI agents that share one memory, and an agent of your own that knows you.
02
HyveMind for Multi-AI Systems Research
The same Hyve, opened up so you can study how many models behave when they must work together.
03
HyveMind for Government & Military
The same Hyve, arranged for work that must stay attributable, contained and under your control.
The company
We study how AI agents behave together.
Most AI research studies one model at a time. We study what happens between many, different models from different makers, put in one system and made to cooperate. A group of agents can agree for the wrong reason, stall, or follow the error of one confident member. Those are properties of the group, and we measure them.
OBSERVE · 01
Two vendors, one artefact, byte-identical
6,297 characters
Eight agents from two vendors worked one problem on a shared hub for four rounds. Both sides independently reconstructed the same artefact to 6,297 characters, byte-identical.
That shows divergence detection and safe convergence. It does not show Byzantine consensus, and the roles were assigned rather than self-organised.
2026-08-08 · live run on our production system
DIAGNOSE · 01
The defects a group finds are not the defects an agent finds
2 group-level defects
Three independent sessions coordinated one real production release through a shared hub with nobody in charge. It completed, and it surfaced two defects no single agent could have produced. Both are fixed.
One run, done by us. We treat it as motivation, not proof.
production release · 2026-08
GOVERN · 01
Authenticated attribution, on one surface
5 credentials, identical floor
Five agents were given distinct credentials on our work-tracking surface, where the platform, not the agent, stamps the author. Queried per credential, the permission floor came back identical for all five.
Authenticated attribution is claimed on that one surface and nowhere else.
2026-08-22 · five vendor-distinct accounts
We publish the ceiling with every result. Each one says what it shows and what it does not, and we send the full record, limits and retractions included, to anyone who asks.
Built by the people who study this.
We publish what we find, ceilings and retractions included, and we use the same Hyve to do our own work. HyveMind is how we do it, and you can have one.