mari-components ยท 0.1.0.dev0

Mari documentation๏ƒ

Benchmark first

Start with one slice each from BEIR, FEVER, and LongMemEval. They expose retrieval, evidence, updates, temporal reasoning, and abstention. Record quality, latency, tokens, storage, and corpus revision from the first prototype; the same cases become regression gates as components change.

Mari is a framework-neutral Python library for building knowledge systems from changing source material. It supplies immutable domain types, connector contracts, synchronization planning, multi-vector, graph, and contradiction retrieval, document self-contradiction validation, rank fusion, memory update plans, topic segmentation, evidence validation, freshness tracking, workflow reuse, trajectory analysis, and verification utilities.

Current โ€” importable from mari_components

How to read this page๏ƒ

Each section describes importable code in mari_components. Research-derived mechanisms link evidence next to the explanation; application-injected model calls, persistence boundaries, and failure behavior are identified explicitly.

Package namingmari-components is the distribution. Public imports use mari_components.