Your AI agent forgets everything between sessions. Keyoku fixes that.
Keyoku is a persistent memory system for AI agents. It stores memories, builds a knowledge graph, and runs a proactive heartbeat that tells your agent what needs attention — without being asked.
One command to install. Zero config to start.
npx @keyoku/openclaw init|
Auto-recall, auto-capture, heartbeat, 7 memory tools, CLI |
Semantic search, knowledge graph, 12-signal heartbeat, memory decay |
Your AI Agent
│
▼
@keyoku/openclaw (plugin)
│ Every message:
│ 1. Recall — search memory, inject context
│ 2. Respond — agent replies with full memory
│ 3. Capture — extract and store facts
│ Every heartbeat:
│ 4. Scan — 12 SQL-driven signal checks
│ 5. Analyze — LLM evaluates with knowledge graph
│ 6. Act — agent responds per autonomy level
▼
keyoku-engine (local)
│ semantic search, dedup, decay, knowledge graph
▼
SQLite + HNSW (your machine, your data)
| Store & retrieve | Keyoku | |
|---|---|---|
| Memory | Save text, search later | Dedup, conflict detection, decay, knowledge graph |
| Heartbeat | Read a static file | 12 signals from actual memory — deadlines, sentiment, patterns |
| Autonomy | On or off | Three levels: observe, suggest, act |
| Capture | Manual or post-session | Real-time per-message extraction |
| Graph | None | Entities, relationships, BFS traversal, alias tracking |
All data stays on your machine. Open source. Built for agents that need to remember.