Half your AI models may be doing the same job.
The Astrolabe proves which — so you stop paying twice.
The AI Astrolabe is the measurement engine inside every Backstaff Board. Point it at your fleet of hosted AI models and it measures how each one actually behaves, then groups them — showing how many are genuinely distinct and how many are redundant copies you’re paying to run for behavior you already have. The same signed measurement that flags the waste is the proof a regulator accepts, and only that summary ever leaves your network; the models and your data stay inside. From there, Backstaff turns it into an audit trail, Starboard gives operators the readiness action surface, and each Backstaff Board packages it as evidence.
See the BoardsWhere this instrument sits.
Astrolabe takes the reading, Backstaff keeps the record, and Starboard steers the fleet. Each instrument inherits the same proof object without moving the underlying cargo.
What the Astrolabe measures.
The 15th-century astrolabe was the era’s computer: a brass instrument that took the position of a star and resolved it into a fix you could act on. Planisphere’s Astrolabe keeps the function. It is the measurement kernel — the thing that takes the reading on a fielded AI fleet and resolves it into a number you can attest to.
Any hosted AI tool docks at the Port, the universal, content-agnostic boundary. Behind it, the Astrolabe probes each adapter and grades three things: distinctness (is this agent actually different from its neighbors, or a clone), drift (has it moved from the behavior it was certified at), and coherence (does it hold together under load). Those grades fold into a per-adapter signature, a fleet sha, and a single Merkle root. Only that sha-pinned, Merkle-rooted evidence crosses the boundary. The model weights, the prompts, the conversation — none of it leaves your enclave.
This is the load-bearing claim, and it is the opposite of a dashboard. A dashboard reflects telemetry someone else already trusts. The Astrolabe produces the trust — a measurement a third party can recompute byte-for-byte without ever seeing the cargo. The measurement is the deliverable. U.S. patent pending.
The visual of the Astrolabe.
You never look at the Astrolabe directly — you look at what it casts. Every surface Planisphere ships is the same reading rendered for a different reader. The kernel takes the position fix once; the visuals are how that fix becomes legible to a compliance officer, an operator, a court.
*-board are not separate products. They are the single Astrolabe reading, shown to whoever needs to act on it.
Under the Astrolabe.
The Astrolabe sits at the center of a small, named set of surfaces. The Port is where the fleet docks; the Schema is the form the reading takes; the Backstaff and the Starboard are the two faces of the finished reading; the verticals are the renders; and the methodology is the published proof that none of it is hand-waving.
Next step.
If your domain is Law, Education, Defense, or Medicine, open its board above. If you want to see the measurement before you commit, the evidence check recomputes a sample evidence record in your browser with nothing leaving your device. If you want a reading taken on your own fleet, open a procurement channel.