personal_memory_systemlisted

# Personal Memory System ## Purpose Give an AI assistant a long-term memory that stays organized, searchable, and current: curated facts by topic and person, a temporal log of what changed and what is still open, and a scheduled job that collects, updates, and refines it. ## Workflow ### 1. Set up 1. Run `bin/init-memory` — creates the directory skeleton and starter files under the memory root (`$PERSONAL_MEMORY_ROOT`, default `~`). Never overwrites existing files. 2. Fill in the central profile and personalization notes from what you know about the user. Nothing goes into memory without the user's approval (see `references/privacy-and-validation.md`). 3. Schedule the refresh (see `references/cron-templates.md`): one daily collect/update/refine job, an optional lightweight watcher that surfaces urgent items between runs, and an optional nightly evaluation that audits the system's disciplines. ### 2. Reading memory - Search BEFORE answering anything about prior work, decisions, dates, people, preferences, todos, ongoing work, or the user's history — and before recommending anything, even when the request mentions no history. - Concepts ("what do we know about X?") → semantic search over the memory files (on Muse: `muse.memory_search`). - Exact identifiers (names, dates, confirmation numbers, addresses) → `bin/memory-grep <pattern>` (keyword search across the whole tree). - When a person, group, or topic is in play, read their page in addition t