A class above Cardinal. Every other LLM reasons within a turn. Supremi reasons across time — quadrupled, audited, recorded.
Every output passes through four verification passes and is recorded in six second-order registries before it reaches you. Daily briefings before you arrive. Drift detected before you notice it. Directives proposed from a substrate Supremi has watched grow for ninety days.
Supremi never returns a single-pass answer. Every output is examined four times by independent verifiers, and the result of each pass is recorded in a second-order node registry — the reasoning's reasoning, kept on disk, audit-grade.
The passes catch what they catch because Supremi is built on principles, not preferences. These are the four that do not bend.
Every claim is traceable to a node, a snapshot, and a date. If a provenance chain breaks mid-answer, Supremi stops the answer rather than smooth the seam.
The substrate is not a cache. Today's reasoning is read against three months of you, and changes to old nodes preserve the prior version — provenance is never overwritten.
Where two of your nodes disagree, Supremi shows both — labelled, dated, and side-by-side — and lets you decide. It will not quietly pick a winner to make the answer feel cleaner.
Supremi does not praise the question, validate the framing, or open with a compliment. The linguistic pass rewrites flattery out before it ever reaches you.
Cardinal reasons about your substrate. Supremi reasons about how your substrate is moving.
Treats the substrate as a sequence of snapshots taken over months, not a graph queried at one moment. Detects voice drift, theme migration, registry-mix imbalance.
Daily briefings, weekly syntheses, monthly trajectories. They materialise before you ask. The first thing you see in the morning is Supremi's read of yesterday.
Proposes typed directives from a whitelisted vocabulary that modulate how your substrate engine reasons. Every proposal needs your acceptance. Expires after 90 days.
Deep Research is three passes per sentence. Supremi is N passes per question, bounded by node budget. Each pass refines the prior and is allowed to contradict it.
Answers carry inline indicators showing where each claim pulls from chronologically — yesterday, last week, three months ago. Hover for the temporal lineage.
Pruning, merging, contradiction surfacing — performed continuously within a strict permission envelope. Pruned nodes recoverable for 30 days. Merges preserve provenance.
No per-turn ceiling. Supremi dispatches as many MCP tool calls as the reasoning requires. The chain budget is wall-clock time and node emission rate, not call count.
By the time a competitor has the architectural prerequisites — substrate registry, fifteen migrations of schema, protocol stack discipline — your substrate has eighteen months of accumulation theirs cannot replicate at any cost.
Odahl Legendaria has two doors and only two. The price exists so the work doesn't have to. The work exists so the price doesn't have to.
Active Maximalis-or-above subscription, at least 30,000
substrate nodes (roughly thirty days of accumulation), and
acceptance of the consent text. Endpoint:
POST /api/supremi/enable.
Hold a daily score of ≥ 70% across the seven substrate-health dimensions for 30 consecutive days and Legendaria activates for free — no card, no trial-to-paywall trap. Lapse and you have a thirty-day grace window to rebuild the streak before access reverts.
A class that runs autonomously between sessions and writes self-modifying directives is not appropriate for every user. The default Odahl experience is Cardinal-and-below — stateless, predictable, easy to audit. Supremi is for the user who has decided their substrate is a long-term cognitive partner, not a chat tool. The plugin format makes that decision explicit: you do not get Supremi by default, you install it deliberately, and you accept what four-pass autonomy means before it begins.