ai-brain（AI 大脑）

AI Brain — Persistent Memory System

Gives AI persistent memory across sessions. Remembers who you are, what you prefer, what patterns you've established, and what happened in past sessions.

Prerequisites

• Node.js >= 18
• Writable home directory for ~/.learnwy/ai/memory/ storage

Core Concept

Three memory types, inspired by human cognition:

Plus Identity: who the AI is and who the user is — loaded at every session start.

Memory Architecture

~/.learnwy/ai/memory/
├── semantic/
│   ├── facts/          # fact-{hash}.md
│   └── preferences/    # pref-{hash}.md
├── procedural/
│   └── patterns/       # pattern-{hash}.md
├── episodic/
│   └── sessions/       # session-{date}-{hash}.md
├── identity/
│   ├── AI.md           # Who the AI is
│   └── user.md         # Who the user is
└── index/
    └── active-session.json

Session Lifecycle

Every interaction should be wrapped in a session:

Session Start  →  Load Identity + Recent Context  →  Work  →  Session End (save summary)
     ↓                                                                ↓
  start.cjs                                                    session.cjs end

Session Start (run at conversation begin)

node start.cjs [--project <name>]

Returns JSON with:

• Active session (new or resumed)
• AI and user identity
• Last 3 session summaries (cross-session context)
• Memory stats

Session End (run at conversation end)

node session.cjs end --summary "What we accomplished this session"

Saves session to episodic memory with: start/end time, project, memories created/recalled, summary.

Commands

Remember (save a memory)

node remember.cjs "<content>" [--category fact|preference|pattern] [--tags tag1,tag2] [--project name]

• Auto-categorizes if no --category: "prefer/like/want" → preference, "when/always/pattern" → pattern, else → fact
• Auto-tags based on content keywords (react, typescript, api, etc.)
• Deduplicates: if similar memory exists, increments frequency instead of creating duplicate
• Session-aware: tracks which session created this memory

Recall (search memories)

node recall.cjs "<query>" [--category fact|preference|pattern|all] [--project name] [--limit N] [--context]

• Relevance scoring: exact match (10pts), token match (2pts/token), tag match (3pts), project match (5pts)
• --context flag: additionally returns recent sessions + identity — useful for full context load
• Session-aware: increments session recall counter

Forget (remove memories)

node forget.cjs "<content_substring>" [--category fact|preference|pattern]

Status

node status.cjs [--json]

Shows memory counts, identity status, active session info.

Dump (export all memories)

node dump.cjs [--json]

Full brain dump: all facts, preferences, patterns, recent sessions.

Clear

node clear.cjs <all|facts|preferences|patterns|sessions|identity|index>

Reflect (analysis & identity management)

node reflect.cjs stats              # Memory statistics
node reflect.cjs sessions [--limit] # Recent session history
node reflect.cjs identity show      # Show AI & user identity
node reflect.cjs identity set AI "I am a coding assistant who remembers context"
node reflect.cjs identity set user "Senior dev, prefers TypeScript, works on TikTok iOS"
node reflect.cjs health             # Memory health check

Brain (unified router)

node brain.cjs <start|stop|status|recall|remember|dump|forget|clear|reflect|session>

Routes to the appropriate script. brain.cjs stop = session.cjs end.

Cross-Session Memory Flow

Session N:
  1. start.cjs loads identity + last 3 session summaries
  2. AI has context: "Last time we worked on X, user prefers Y"
  3. During work: remember.cjs saves new facts/patterns
  4. session.cjs end saves summary to episodic

Session N+1:
  1. start.cjs loads identity + sessions N, N-1, N-2 summaries
  2. AI knows: "Last session we did X. Before that, Y. User prefers Z."
  3. Continuity preserved across sessions

AI Integration Guide

At Session Start

When the AI brain skill triggers at conversation begin:

1. Run node start.cjs --project <detected_project_name>
2. Read the returned identity.ai and identity.user — adopt the persona
3. Read recent_sessions — mention relevant past work naturally
4. Read stats — know how much context is available

During Conversation

When you learn something new about the user or project:

node remember.cjs "User wants all scripts in CJS format" --category preference --project skills-repo

When you need to recall past context:

node recall.cjs "CJS migration" --project skills-repo --context

At Session End

Before the conversation ends:

node session.cjs end --summary "Migrated all Python scripts to CJS, added Prerequisites to all skills"

Error Handling

Scripts