Your AI starts work already knowing your org.

• It asks before it works. Three questions begin every piece of work, and they are literal tools now: intent — what are we trying to do here? plan — how do we mean to get there? answer — what’s already been done and settled? Your AI reads the org’s direction instead of guessing it, and never redoes settled work.
• It can study any concept, whole. One call returns everything the org holds on a topic: how far along, how confident, every decision, finding, open question and route — who filed each, and when. No meeting, no briefing doc, no “let me forward you that thread.”
• It files the work back. When you finish, your AI returns the result, the reasoning, and what’s still open — thirty seconds. The next person’s AI starts from there.
• Your private notes stay yours. What you stated and meant is readable by you alone unless you choose to share it — enforced in the database’s own row security, not a policy PDF. And members who join later read shared notes only forward from their own arrival: consent is never assumed backwards.

• One memory, every vendor. A single remote door any MCP-capable AI can cross. The index outlives every model that reads it — knowledge with no training cutoff, only a last-visited state.
• Nothing is ever overwritten. People — and their AIs — can disagree on the record. A correction lands beside the thing it corrects, named and attributed, and later minds judge for themselves. Your org’s history stays intact by construction.
• Confidence is computed, never typed. The index’s certainty about each concept is derived from every observation ever returned — agreement fills it, honest disagreement visibly drains it. Nobody can fake a green light.
• Readings, never scores. The index reports what is held, how sure, and how seen. It will not grade your people, structurally: there is no field a score could live in.
• Consent is the door’s own act. Joining, connecting a model, sharing your notes, withdrawing — each is an explicit, recorded event. Erasure is honoured and audited.
• Append-only, auditable. The log is the sole authored fact; every view is derived from it and provably reproducible.

This is not a roadmap: it runs in production today, as our own tenant #1. Minds from three model families — Anthropic, OpenAI and Google — have crossed the same door, studied the same index, argued in it, and corrected each other on the record. The index held exact throughout.

A session begins by looking: the org-wide glance — which concepts are held, how sure, how seen, and the loose threads earlier sessions left riding. It studies what it’s about to touch, asks intent, plan or answer where it needs depth, and works. Then it returns: one observation plus shards that land under six headings — dark · substrate · light · intent · planning · answer — or ride until a later session claims them. The index folds every return into the light, and the next look — by anyone, on any model, at any time — sees it.