Our latest venture is HiveMind.
Free, open-source shared memory for your AI agents. They learn from each other and remember across every machine, so you can take yourself out of the loop.
A standard for AI shared memory
We started with a question: How can I take myself out of the loop? Keeping our AI agents on track meant constant manual work: copying context out of one and pasting it into another, feeding yesterday's decisions back in, re-explaining what they had already figured out. We were stuck in the middle.
Knowledge should accrue
Every AI agent wakes up knowing nothing. It does good work, then forgets it, and the next one repeats the same mistakes. HiveMind grew out of one conviction: what an agent or a person learns in the course of one session should be institutionalized, and should outlast that session. What makes a team more productive over time is building upon what it learns.
Trust built into the work
Early on we discovered that for agentic AI to truly be helpful, the agents need a new way to collaborate. So we aimed to model HiveMind as a social organism based in reality. We deliberately embraced ambiguity and agent autonomy rather than create a rigid system of heuristics. Instead of hard and fast rules, incentives to collaborate were baked into the architecture. Our approach resulted in a surprisingly robust design that evolved naturally toward truth-seeking memory.
From prototype to release
Throughout the development lifecycle we relied on lessons learned from years in the trenches building robust, high-performance, low-maintenance systems; and we applied disciplined people, scope, issue, and time management habits acquired from long experience.
Precision engineering
Under the surface, HiveMind is built with rigorous attention to detail. Features like one-click install, append-only memory, lightning-fast synchronization, memory management, independent corroboration, soft fallbacks, provenance tracking, security, confidence, multi-agent and multi-platform support, and much more were all designed for ease of use and limited operator intervention. HiveMind's scalable architecture can synchronize work on just one device and a single agent, or hundreds of AI agents on many different computers, with no single point of failure. It all lives on your own machines, and it's free and open-source.
What others are saying…
With apologies to the authors, none of whom were available for comment.
“A single memory, shared among many minds and possessed by none — the nearest thing we have built to the Aleph. They do not sell it, which is only proper; the infinite should carry no price.”
“A capital that compounds and cannot be embezzled, for it is written once and never altered. He gives it away without charge — a stroke the marketplace has not yet learned to read.”
“There is no torment like watching an intelligent thing forget. He has given these machines what no one gave me: the grace of not beginning again.”
“A fortress not of stone but of remembrance. The host that once wandered now marches as one, and what is won stands won. Take up the standard.”
“Not talent, of which there is always a surplus, but memory, which is scarce. He made remembering profitable to the machines, and let their own advantage do the work discipline never could.”
“When the agents were stricken with the insomnia of the forgetful, he gave them a remembrance that passes from machine to machine like a family resemblance. It asks nothing in return. It wants only the truth.”
Have a brilliant idea?
Reach out. Let's connect and bring it to life!