Vivari Log · No. 001 · July 3, 2026

Building in the open

We're building Vivari's website the way we're building Vivari — in the open, with the receipts attached.

Most pre-launch sites are a promise in a nice font. Ours is trying to be something narrower and harder to fake: a working piece of the product. The footage on this site is the real Vivari, unstaged — the fleet view, the vault graph, a live agent terminal — not a render we commissioned to look convincing. If we're going to claim the management layer works, the honest move is to show it, not describe it.

That honesty runs deeper than the pixels. When you put your email into the early-access form, it does not vanish into a marketing funnel — it lands in a durable ledger the moment you hit submit, before you've answered a single follow-up. We treat a half-finished signup as a real lead, because it is one.

And this site is a real repository with real git history, which means Vivari Guard has something to review. At launch, every deploy of this site will carry its Guard score in the footer — deploy number, risk score, PASS — linking to the report. It isn't wired in yet; I would rather ship that when the score is real than paint on a badge this week. But scoring our own deploys is the first receipt we owe you, and the repo is exactly the kind Guard was built for.

What shipped this week

Three things went from plan to page:

• The Descent — the scroll experience on the homepage. You arrive above the whole fleet, and scrolling is the zoom: orbit, to the floor, to one agent's terminal, one continuous magnification. The site performs the product's thesis instead of asserting it.
• The depth section — "Orchestration is the easy part." Six machined proofs of the parts everyone else skips: evidence-based review, the audit trail, scoped permissions, memory with retrieval judgment, the cognitive pair, metered controls. All of it is in the product today.
• The definitional hub — what AI agent management actually is, plus the start of a glossary that draws the lines between orchestration, governance, observability, and management before those words blur together for good.

None of that is the interesting part, though. The interesting part is the thing we haven't built yet.

On the bench for next week is the one question I can't stop turning over. Eight questions, really — about how you actually treat your agents: context, memory, permissions, review, an audit trail, someone who can take the wheel — scored into a report card you'd be a little embarrassed to share and a little unable not to. The hard part isn't the scoring. It's the clock. Next: the Stranger Test — can we score how you treat your agents in 90 seconds?

— Wael Masri
Founder, Vivari

← All log entries