- name
- openclaw-memory-hub
- description
- Three-tier memory architecture for OpenClaw AI agents. Provides L0 runtime semantic retrieval (Ollama bge-m3 + SQLite-vec vector store), L1 working memory (daily markdown logs), L2 long-term memory (curated base file), Dreaming pipeline for automatic insight promotion, and three-way sync (Cloud ↔ Markdown ↔ Vector). Use when setting up persistent agent memory, configuring memory plugins, or building multi-layered memory systems for OpenClaw.
- version
- 1.7.0
OpenClaw Memory Hub
Three-tier memory architecture with automated Dreaming pipeline and three-way synchronization.
Overview
This architecture solves AI amnesia across sessions by layering memory at three levels:
| Tier | Layer | Technology | Purpose |
|---|---|---|---|
| L0 | Runtime Retrieval | memory-core plugin (Ollama bge-m3 → SQLite + sqlite-vec) | Real-time semantic + BM25 hybrid search |
| L0 | Cloud Recall | MemOS Cloud plugin (*optional*) | Cross-device memory capture and recall |
| L1 | Working Memory | memory/YYYY-MM-DD.md files | Daily summaries, todos, technical notes (30–90 day retention) |
| L2 | Long-term Memory | MEMORY.md (read-only base) | Key facts, user profile, permanent decisions |
Automated Pipelines
- Dreaming (03:00 UTC daily): Scans conversation logs, evaluates candidates via DeepSeek analysis, promotes high-scoring insights to L2
- Three-way Sync (18:00 / 20:00 / 22:00 CST): Keeps Cloud ↔ Markdown ↔ Vector stores in sync
- **Wiki Compilation (04:00 UTC daily, *optional*)**: Extracts entities and concepts, writes structured wiki vault pages
Setup
One-command auto-setup
bash scripts/auto-setup.shThis script handles everything interactively:
| Step | What it does | Toggle |
|---|---|---|
| 1 | Install Ollama (standard or Intel edition) | --skip-ollama |
| 2 | Download bge-m3 embedding model | — |
| 3 | Plugin conflict check (auto-detect subconscious-personality-guardian) | — |
| 4 | Check memory-core plugin status | — |
| 5 | Insert memory-core config into openclaw.json | Auto-insert on confirm |
| 6 | Install and configure MemOS Cloud plugin with critical config | --skip-memos |
| 7 | Create memory/ directory, check AGENTS.md | — |
| 8 | Set up Dreaming cron job (03:00 UTC daily) | — |
Options
bash scripts/auto-setup.sh --skip-ollama # Skip Ollama install (use your existing one)
bash scripts/auto-setup.sh --skip-memos # Skip MemOS Cloud plugin entirely
bash scripts/auto-setup.sh --dry-run # Preview without making changesManual setup
See references/setup-guide.md for step-by-step manual configuration.
When to Use
- Setting up OpenClaw memory for the first time
- Configuring memory-core plugin with local Ollama embedding
- Installing MemOS Cloud plugin for cross-device sync
- Setting up automatic Dreaming and promotion pipelines
- Configuring three-way sync between cloud, files, and vector DB
Plugin Conflicts
❌ subconscious-personality-guardian ↔ memory-core
Incompatible. Both use the same OpenClaw memory slot. Installing both causes write conflicts and retrieval duplication.
Fix: auto-setup.sh detects and disables this automatically. Manual fix:
{
"plugins": {
"disabled": ["subconscious-personality-guardian"],
"deny": ["subconscious-personality-guardian"]
}
}✅ memory-core + MemOS Cloud
Compatible — designed to work in layers.
They execute in sequence, not in competition:
User message
→ MemOS Cloud (before_agent_start hook)
→ Injects: static facts, preferences, profile
→ memory-core (runtime semantic query)
→ Injects: recent conversations, topical context
→ Agent receives layered memoryMemOS handles "who the user is" (long-term facts). memory-core handles "what we talked about" (recent history).
Critical config (MemOS Cloud):
{
"recallFilterFailOpen": true, // Don't block pipeline if API fails
"asyncMode": true, // Let memory-core run too
"resetOnNew": true, // Fresh context per session
"hooks.allowConversationAccess": true // Hooks need this
}See references/architecture.md for the full recommended config.
Both should be enabled together.
⚠️ MemOS Cloud + ReMe
Potentially conflicting. File layer overlap and retrieval duplication. Choose one.
See references/architecture.md for full compatibility details.
Components
1. Memory Plugins (L0)
Configured via openclaw.json:
{
"plugins": {
"entries": {
"memory-core": {
"config": {
"embeddingUrl": "http://127.0.0.1:11434/api/embed",
"embeddingModel": "bge-m3",
"dimension": 1024
}
},
"memos-cloud-openclaw-plugin": {
"config": {
"url": "https://your-memos-server",
"token": "your-token"
}
}
}
}
}2. Memory Files (L1 + L2)
~/.openclaw/workspace/
├── memory/
│ ├── YYYY-MM-DD.md # Daily working memory (auto-indexed)
│ ├── MEMORY_INDEX.md # Vector BM25 cluster summaries
│ ├── memos-cloud-*.md # Cloud-pulled memory entries
│ ├── .sync-cloud-state.json # Cloud pull cursor
│ └── .sync-push-state.json # Push state (SHA256 tracking)
├── MEMORY.md # Long-term memory base (read-only)
├── AGENTS.md # Runtime context + memory rules
└── SOUL.md # Agent persona3. Sync Scripts (*optional*)
Located at user_workspace/scripts/:
sync-cloud-pull.py— Pull from MemOS Cloud → Markdown filessync-cloud-push.py— Push local markdown changes → Cloud (SHA256 diff)sync-vector-index.py— Vector DB → MEMORY_INDEX.md (FTS5 BM25 clustering)sync-all.sh— Orchestrator that runs all three
See references/sync-api.md for MemOS Cloud API details.
File Reference
references/architecture.md— Detailed architecture documentationreferences/setup-guide.md— Complete manual setup guide with templatesreferences/sync-api.md— MemOS Cloud API referencescripts/auto-setup.sh— One-command interactive setup (recommended)