Your agents report what they did. Nothing remembers why.
Flashy Mind is a shared, governed, agent-maintained brain for a whole organization: plain-markdown knowledge agents compile nightly, humans trust, and auditors can replay — the memory layer of FlashyOS, for AI Autonomous Organizations (AAO).
npx @flashyos/agent init --with-memoryplanned syntaxFlashy Mind is a design document and a working single-founder prototype — not a product you can buy. Everything on this page is labeled: green means it exists and is checkable, gold means it’s where this is going.
What is an organizational memory layer?
Shared, permissioned, auditable memory for a whole organization of humans and AI agents — distinct from personal AI memory (“second brain”) products, which serve one person. Personal memory shipped everywhere in 2026. The organizational version is still nobody’s product, because it’s a governance problem, not a storage problem:
Every run starts from zero
Context gets re-explained, work gets redone, and the same question gets answered five different ways by five different agents.
Fifty agents, fifty private notions of the truth
Without one governed source, knowledge drifts and contradicts silently — and nobody notices until it costs something.
Nobody can answer “why did the agent decide that?”
The compliance question. If your agents act on memory nobody can audit, your board is trusting a black box.
One vault, four layers, one writer each.
The vault is plain markdown in a git repository the org owns. Every layer has exactly one writer — and that rule is an auth property, not a prompt rule: the compiler’s write scope is a uniqueness constraint in the token model, not a polite instruction.
Governed by machinery that already ships.
FlashyOS’s decision log — owners, impact tiers, role-checked resolution — is live today. The Mind’s core layer is where it graduates from audit trail to enforcement:
An agent (or human) proposes a change to a protected page.
The diff lands on a proposal branch with a high-impact Decision, pending.
An owner or admin resolves it — the same role checks FlashyOS enforces today.
Not just logged: the approval performs the change, signed in the commit trailer.
Exists today vs. planned
The line between these columns moves as we ship — and it moves here first.
External vaults ship only after our own security preconditions are met. We will not ship governed memory on ungoverned rails — that’s the whole point of the product.
We’re giving our own company the first Mind.
The dogfood plan: our group roadmap becomes the seed vault, the compiler registers as an ordinary FlashyOS agent so its nightly runs are visible in Live HQ, and the exit test is a five-hop loop — a capture is compiled, the morning brief cites it, a different agent’s context pack contains it, and a protected-layer change lands only through an approved Decision. All observable, all replayable from git. When it passes, you’ll read about it in the press feed with real dates — same as everything else we ship.
Fair questions
No. It is in design, dogfooded internally on our own company first. The waitlist is the honest front door — join it from the request-access form and we will flag your interest.
Because trust is structural, not promised: exactly one governed writer per layer (enforced by identity, not prompts), every compile is a git commit you can diff and roll back, protected layers change only through an approved Decision, and every claim cites its sources.
An org running roughly 10–50 agents in production with a compliance or board-reporting reason to answer "what did our agents decide, and why." You take feedback calls and shape the product; you get it first. We are selecting 3–5 for the first cohort.
It will meter on your existing org tier when it ships. No number until pricing is signed off — the same discipline as our pricing page.
Plain markdown in a git repository you can own. If our database burned down, your vault would lose nothing. Leave anytime, take everything.
Want your org to have a Mind?
Design partners come from the same funnel as everything else — 3–5 slots in the first cohort, and that’s a real number, not scarcity theater. Tell us what your agents forget.