If the tools change, the knowledge should survive.
Every index, connector, and AI client around your knowledge is designed to be replaceable. The thing meant to outlast all of them is the durable, human-readable record itself.
Human-readable by design
Durable knowledge is stored as plain, human-readable Markdown — not locked inside a proprietary database format only one piece of software can open. You, or any tool, can read it directly.
Replaceable clients and indexes
Search indexes, embeddings, and connector metadata are derived from the canonical knowledge. If one breaks, goes stale, or needs replacing, it's meant to be rebuilt from the source — not restored from its own backup as though it were irreplaceable.
Durable knowledge → backup → fresh environment → rebuild → reconnect → resume
Step through the conceptual recovery path below.
This describes the recovery model conceptually. It does not describe or reveal any specific backup destination, credential, or infrastructure detail.
History and provenance
Durable items carry enough history to show what changed and, where useful, why — a superseded decision doesn't just vanish into a newly overwritten value.
Recovery validation
Recovery is meant to be checked, not assumed — reconstructing an environment and confirming retrieval actually works against it, rather than declaring success once files exist on disk.
Want the developer-level contract for this?
For Developers covers the portability and recovery contract in more concrete terms.