Persistent memory layer for Council Room — reads, writes, and searches agent memory across sessions. Use when saving important decisions/facts to long-term memory, recalling past sessions, updating MEMORY.md or daily notes, searching the Obsidian vault for context, or injecting memory into agent prompts so agents never forget who Lucy is or what DataFlow is building. Replaces the goldfish memory problem.
Gives Council Room agents persistent memory across sessions — like Kora's shared memory, but self-hosted and unlimited.
Short-term → memory/YYYY-MM-DD.md (daily raw notes)
Long-term → MEMORY.md (curated facts)
Agent facts → vault/00_System_Core/ (Obsidian vault)
Council KB → council_memory.json (shared agent facts)
Paths:
/home/openclaw/.openclaw/workspace//home/openclaw/livesync-bridge/vault//home/openclaw/council-room/data/council_memory.json/home/openclaw/.openclaw/workspace/memory/YYYY-MM-DD.mdpython3 ~/.openclaw/workspace/skills/council-memory/scripts/memory_op.py read
python3 ~/.openclaw/workspace/skills/council-memory/scripts/memory_op.py write \
--fact "Lucy decided to target bookkeepers with 5+ years experience" \
--category "decisions"
python3 ~/.openclaw/workspace/skills/council-memory/scripts/memory_op.py search \
--query "bookkeeper validation"
python3 ~/.openclaw/workspace/skills/council-memory/scripts/memory_op.py daily \
--note "Rafael standup: OCR accuracy at 87%, needs 8% improvement"
python3 ~/.openclaw/workspace/skills/council-memory/scripts/memory_op.py context \
--agent "rafael"
Returns structured context string to prepend to agent prompts.
| Situation | Where to Write |
|---|---|
| Important decision made | MEMORY.md + council_memory.json |
| Daily event/progress | memory/YYYY-MM-DD.md |
| Agent-specific lesson | vault/Lessons/<agent>.md |
| Lucy preference discovered | MEMORY.md under "How Lucy Works" |
| Metric/KPI update | council_memory.json categories.kpis |
| Technical decision | MEMORY.md under "Technical Decisions" |
When starting a new session or agent chat, call context op to get:
This prevents agents from giving generic answers — they always know:
See references/schema.md for full schema. Key categories:
facts — persistent truths about Lucy/DataFlowdecisions — logged decisions with date + rationalekpis — current metrics (OCR accuracy, bookkeepers found, etc.)lessons — what agents learned from past sessions