We're building one memory your whole team of agents can trust.
One AI that remembers is the easy part. The hard, unsolved problem is letting many agents — a lead agent, its subagents, and the tools they use — share one memory without stepping on each other, contradicting each other, or getting fed bad information. That's the problem we exist to solve.
- parent
- subagent
- peer
- tool
Shared memory is the missing piece.
Single-agent memory is the simple case. We started with the hard one.
AI is moving from one assistant to teams of agents working together. The moment you have more than one, an old problem gets ten times harder: what does the team remember, and can it trust what it remembers? We believe shared memory — governed, fast, and safe from poisoning — is the missing piece that makes agent teams actually work. So that's all we build.
The timing isn't opportunistic.
Three real shifts that make a shared, governed memory necessary now — not someday.
Agents went multi.
Orchestrators spawning subagents is now the default pattern — and none of them share what they learn.
Memory became an attack surface.
Planting a false 'fact' once, recalled forever, is a documented attack. Most memory tools have no defense.
Learn more →The benchmarks are contested.
Vendor self-benchmarks disagree; nobody owns honest, independent numbers. We want to.
Learn more →
Our values, in plain words.
Character without slogans — the principles every decision is measured against.
Lead with the hard problem.
Many agents sharing memory safely — not the trivial single-agent case.
Earn trust, don't assume it.
Every memory has a source, a reason, and an audit trail you can see.
Fast or it doesn't count.
Sub-200ms recall is a feature, not a stretch goal.
Open and honest.
Real numbers, plain language, no lock-in. Works with any model.
The people behind the mission.
Names, photos and bios are placeholders until launch — we don't fabricate people.
[Founder name]
Founder [One-line bio placeholder — replace before launch.]
[Team placeholder]
[Role] [One-line bio placeholder.]
[Team placeholder]
[Role] [One-line bio placeholder.]
[Founder placeholder — one short paragraph on why this problem, in their own voice. Keep it real and human; replace before launch.]
We're not vaporware.
No fake logos and no fake metrics. While we line up backers, advisors and design partners, the work is public.
Help us give every agent a memory it can trust.
Or just say hi →