Content-addressable storage, commonly abbreviated CAS, computes an identifier from the bytes of an object. Identical content receives the same identifier, while any content change produces a different address.
An agent runtime can store prompts, skills, tool definitions and messages this way to preserve exact versions. Content addresses support deduplication, diffs, caching and reproducible replay, provided the hash algorithm and referenced components are recorded consistently.
ELI5
Content-addressable storage names a file or object using a digital fingerprint calculated from its exact contents. If even one byte changes, the fingerprint and therefore the address changes too.
For example, an agent system can store the exact prompt and tool definitions used in a run under their fingerprints. Later, a reviewer can retrieve those precise versions instead of accidentally using newer files with the same filenames.
What does CAS mean in storage?
CAS means content-addressable storage, a system that identifies an object using a hash derived from the object's content.
Why store AI prompts by content hash?
A content hash identifies the exact prompt version and makes changes, deduplication, caching and replay easier to verify.
