AI Memory Protocol

Versioned, graph-based persistent memory for coding agents using Sphinx-Needs with typed entries, relationship tracking, confidence scoring, and stale detection.

Stores conversation context as structured reStructuredText files using Sphinx-Needs directives, combining human-readable documentation with machine-queryable data. Supports typed memory entries (observations, decisions, facts, preferences, risks, goals, questions) with graph-based relationship tracking (relates, supports, depends, supersedes, contradicts), hierarchical tagging, confidence scoring, and automatic stale detection. Provides a CLI with 16+ subcommands and autonomous capture from Git commits, CI logs, and discussion transcripts.