FAQ

Questions people actually ask

Short answers, including the ones that end in “no”.

Is AETHER-33 a new moat?

No. CAIN's architecture is seven moats, and that is fixed. AETHER-33 is a product that extends the existing governed-memory work. Naming it a moat would be exactly the kind of inflation the project's own claim registry exists to prevent.

Can memory make an agent trusted?

No. Trust is computed from evidence by the CAIN runtime. Every AETHER-33 response says grants_authority: false and there is no endpoint that changes that.

Does it call a model?

No. Retrieval is BM25, screening is a normaliser plus pattern families, contradiction scoring is TF cosine. No token spend, no network, no drift when a model version changes.

Is it a vector database?

No. If you need one, put it in front. AETHER-33 is the governed record: what was written, by which source, whether it was altered, and whether it may be read.

Why 33?

Thirty-three invariants, each a check that runs against the real engine. The number is the size of the contract, not a model parameter. It is visible at GET /invariants.

Is it audited?

No. There is no third-party security audit, no SOC 2, and no certification. The 33 invariants are this project's own tests. They are evidence of what the code refuses to do, not an external assurance opinion.

Where did “Aether-33” come from?

The owner named the product. It did not previously exist in this repository — no file, branch or commit — so it was built rather than renamed. That fact is recorded here so nobody later mistakes a new product for a rediscovered one.