Explanation¶

These pages explain why anagrapheus is shaped the way it is: what each layer is for, what it guarantees, and which decisions it deliberately leaves to your application. They are for reading away from the keyboard. For step-by-step instructions, see the How-To Guides.

Start here¶

How the Layers Fit Together: one authoritative log, every other component a projection over it, and a clear line between what the library decides and what your application decides.

The foundations¶

The Event Log and Projections: what an envelope, a fold, a projection and a position are, and why every read model can be thrown away and rebuilt.
Pinning and Replay: how model and tool calls are recorded under a key derived from what was asked, so a run replays offline or says exactly where it diverged.

The layers built on them¶

Beliefs: per-holder beliefs as a fold over the log, with belief ids minted from declared predicates so two agents can disagree measurably.
Identity: how entities are established or resolved, why merges and splits are events, and why re-keying beliefs after a merge is planned by the library but applied by you.
The Chronicle: consolidating old history into prose sections that cite their range, and why this is the one projection that is better carried than rebuilt.
Prompt Assembly: building a model's context only from slices its role was granted, with every part citing its sources and a budget that never drops what must hold.