Cognitive memory engine for AI agents: gives OpenClaw/Hermes human-like long-term memory using 140 years of cognitive psychology (compound-cue theory), not embeddings. Zero LLM API, zero vector DB, pure Markdown storage. Use when the agent needs cross-session recall of user facts, preferences, decisions, or conversation history; when setting up persistent memory that survives restarts; when the user asks 'do you remember X' or complains the agent forgets; when memory must stay 100% local with no API cost. Beats embedding-based systems on Memory-Native benchmarks (nDCG 0.238, 5.2x) with ~7ms keyword search.
Mnemosyne — Cognitive Memory Engine
Give the agent a real long-term memory: four layers (raw → working → medium → long-term), scored by cognitive-psychology formulas instead of neural embeddings.
Core formula (compound-cue theory)
familiarity = 0.35·imp + 0.25·recency + 0.25·keyword + 0.10·hitFreq + 0.05·layerW
Every weight traces to a named psychology paper (Raaijmakers & Shiffrin 1981, Ebbinghaus 1885, Zeigarnik 1927). No hyperparameter sweep, no neural network.
Install
git clone https://github.com/ElonAug7/Mnemosyne-agentmemory-engine-openclaw-hermes
cd Mnemosyne-agentmemory-engine-openclaw-hermes/Mnemosyne-v6.4
bash install.sh
Installer auto-injects the memory protocol into SOUL.md / AGENTS.md and serves a Web UI at http://127.0.0.1:8765.
Protocol (mandatory)
- Before every reply: read
memory/short/working/last-recall.json; if it holds flashbacks, inject them as context. - For history/decisions/facts/preferences queries: run
node tools/memory-engine/engine.js recall --query "<keywords>". - Cite memory source in replies (e.g. "Per your decision on 2026-08-16…").
Common commands
node tools/memory-engine/engine.js search --query "..." --mode keyword— search (~7ms)node tools/memory-engine/engine.js recall --query "..."— structured recallnode tools/memory-engine/engine.js profile— user profile (tech stack, style, decisions)node tools/memory-engine/engine.js todos— todo listnode tools/memory-engine/engine.js sync --quick— session maintenance
Why not embeddings
- Zero LLM API calls, zero vector DB, zero embedding models — memory stays on the machine as diffable Markdown.
- Memory-Native Evaluation (80 queries, 11 systems): nDCG 0.238, beating raw BM25 (0.185) and all embedding systems.
- Cognitive weights decide what to remember and forget — the part embeddings ignore.
References
references/— full architecture, benchmark methodology, Hermes adapter docs.
Related skills
- Downloads
- 587All time
- Security audit
- Review
- Last updated
- 3mo ago
- Current version
- v1.0.0
- License
- MIT-0
