Platform
Architecture, observability, testing, and internals.
How the zero-memory platform is built and operated.
Retrieval is the hybrid RAG pipeline behind every recall: two search legs, rank fusion, chunk-level embeddings, and the signals that reorder the result. The dashboard is composed from one component system, described in UI stack. The e2e harness that exercises it, and the testing contract that goes with it, are in End-to-end testing.
A running instance reports on itself in three places. Observability covers logging, request tracing, and diagnostics. Usage events is the append-only record of metered work, and the audit log records who did what, in the database.
What external clients depend on is written down twice. Contracts and errors describes the schema-first contracts, the versioning, and the error taxonomy; Deprecations is the ledger of deprecated surfaces and breaking changes in the client contract.
The last three pages are about whether memory is actually helping. Value metering is the instrumentation that measures what memory contributes, judge precision is how automated judgments are kept trustworthy, and the memory benchmark measures what you would lose without it.