| name | memos-memory |
| description | Remember and recall information using MemOS long-term memory. Use when you want to explicitly save important facts, user preferences, or project context, or when you need to search past memories. |
MemOS Memory
Agent skill for actively remembering and recalling information.
Works alongside the lifecycle plugin (auto-capture) for hybrid memory.
Quick Reference
Use the bundled script scripts/memos.sh for all operations:
./scripts/memos.sh remember "User prefers TypeScript over JavaScript"
./scripts/memos.sh recall "programming language preferences"
./scripts/memos.sh remember-pair "What framework do you use?" "User prefers Next.js + Supabase"
./scripts/memos.sh status
When to Use
remember
Actively save important information when you discover:
- User preferences: Tech choices, coding style, language settings
- Project info: Architecture decisions, tech stack, config values
- Problems & solutions: Bugs found, workarounds, debugging hints
- Decisions: Technical judgments and policies made by the user
recall
Search past memories when:
- You want to check user preferences before starting a task
- You want to see if a similar problem was solved before
- You need background context about the project
Architecture
| Component | Role |
|---|
| MemOS API (localhost:8000) | Memory management server |
| Neo4j (localhost:7687) | Graph DB (relationships) |
| Qdrant (localhost:6333) | Vector DB (semantic search) |
| Ollama (localhost:11434) | Embedding + LLM |
Tag Convention
| Tag | Meaning |
|---|
| openclaw, auto | Auto-captured by lifecycle plugin |
| openclaw, manual | Explicitly saved by this skill |
Notes
- Auto-capture (lifecycle) records full conversations as a safety net
- This skill is for selectively saving important information
- MemOS internally handles dedup, merge, and hierarchical organization — save freely without worrying about duplicates
- Search uses hybrid: vector similarity + BM25 + graph traversal
