メニュー
Hệ thống bộ nhớ 7 lớp chạy local dành cho AI agent — SQLite + Qdrant + Fabric, zero cloud dependency. Agent nhớ project, decision, reasoning pattern xuyên session.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/yanacuti1121/Yana-AI --skill claudiodrews-memory-osコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
Sovereign-grade safety OS for AI coding agents. 62 hooks, 2,025 skills, L1 memory, circuit breakers, and cross-engine enforcement — blocks rm -rf, force push, pipe-to-shell, and 40+ attack vectors before they reach your repo.
Use when the user wants to generate or keep repository documentation up to date via OpenWiki (langchain-ai/openwiki) — an LLM-driven CLI that writes a wiki for a codebase (or a personal knowledge base from Notion/Gmail/Slack/X/web search) and keeps it fresh via a scheduled CI pull request. Examples: "set up OpenWiki for this repo", "keep the docs updated automatically", "generate an agent wiki".
Use when implementing the core AR pipeline (camera pose estimation, marker tracking, projection overlay) from first principles — not when just using ARKit/ARCore/Unity's AR framework as a black box. Triggers on: 'build augmented reality from scratch', 'marker-based AR tracking', 'camera pose estimation', 'implement fiducial marker detection', 'AR projection matrix math', 'markerless AR tracking'. Covers marker-based vs markerless tracking, pose estimation, and the projection math to overlay 3D content on a camera feed.
SOC 職業分類に基づく
| name | claudiodrews--memory-os |
| description | Hệ thống bộ nhớ 7 lớp chạy local dành cho AI agent — SQLite + Qdrant + Fabric, zero cloud dependency. Agent nhớ project, decision, reasoning pattern xuyên session. |
| allowed-tools | Bash, Read, Write |
| user-invocable | true |
Memory OS biến Hermes Agent (hoặc bất kỳ LLM agent nào) thành collaborator dài hạn: nhớ đủ thứ giữa các session mà không cần cloud, không vendor lock-in.
Layer 1 — Workspace Files
MEMORY.md, USER.md, CREATIVE.md inject vào system prompt mỗi turn
Layer 2 — Session History
SQLite + FTS5 full-text search — archive toàn bộ conversation
Layer 3 — Structured Facts
Entity resolution + trust scoring + auto feedback loop
Layer 4 — Fabric (Cross-Session Extraction)
LLM-powered summarization, 16 tools: fabric_recall, fabric_write, fabric_brief
Layer 5 — Vector Database
Qdrant 4096-dim cosine + BM25 sparse, fallback: hybrid→dense→lexical→SQLite
Weekly decay scan + semantic dedup (cosine > 0.92)
Layer 6 — Auto-Curated Wiki
Self-organizing knowledge vault: concepts, entities, comparisons → Qdrant
Layer 7 — Ground Truth Hierarchy
SOUL.md + rulebook.md — injected memory = authoritative
Ngăn agent re-discover context đã có → tiết kiệm token
# One-command setup — Docker orchestrate Qdrant + Redis + ARQ worker
curl -fsSL https://raw.githubusercontent.com/ClaudioDrews/memory-os/main/install.sh -o /tmp/memory-os-install.sh
# Inspect first: head -40 /tmp/memory-os-install.sh — then run if safe:
bash /tmp/memory-os-install.sh
fabric_recall(query, n=5) # semantic search memory
fabric_write(content, tags) # lưu fact mới
fabric_brief(session_id) # summarize session → store
# Compatible với bất kỳ LLM provider nào Hermes support
# OpenRouter, OpenAI, Anthropic, Ollama, local models
# Databases:
# state.db — session history
# memory_store.db — facts + HRR (hierarchical relation)
# Qdrant — semantic vectors
| Feature | Memory OS | mem0 | Zep | Letta |
|---|---|---|---|---|
| Fully local | ✅ | ❌ | ❌ | ❌ |
| Ground Truth layer | ✅ | ❌ | ❌ | ❌ |
| Trust-scored facts | ✅ | ❌ | ✅ | ❌ |
| Self-curating wiki | ✅ | ❌ | ❌ | ❌ |
| No subscription | ✅ | ❌ | ❌ | ❌ |
Nếu không có Ground Truth ranking: agent bỏ qua injected context, gọi API tìm lại thứ đã có trong prompt → waste token + slow. Layer 7 force agent treat injected memory là authoritative source.