🧠 Memoria — Multi-Layer Persistent Memory for OpenClaw

The most complete memory system for OpenClaw. 20 layers of memory that work together, powered by YOUR choice of LLM.

Why Memoria?

🏗️ 20 Memory Layers (not just a fact store)

Facts — Durable knowledge extracted from every conversation
Procedures — HOW to do things, improves with repetition, learns from failures
Knowledge Graph — Entities + relations connecting your facts
Topics & Expertise — Tracks what you talk about most, specializes over time
Observations — Short-term working memory for active context
Error Detection 🔥 — Touch fire once, remember forever. Dangers captured on first occurrence
Lifecycle — Fresh → Settled → Dormant. Nothing deleted, priority shifts naturally

🔌 Bring Your Own LLM

Configure each layer independently. Mix and match:

Ollama — Run gemma3, qwen3.5, llama, or any model locally (recommended)
LM Studio — Use any GGUF model from your local server
Remote APIs — OpenAI, Anthropic, OpenRouter as primary or fallback
Fallback chains — Ollama → LM Studio → API. If one fails, the next takes over automatically

🏠 100% Local-First

SQLite + FTS5 — No external database needed
Local embeddings — nomic-embed-text via Ollama (zero API cost)
Zero cloud dependency — Works offline, your data stays on your machine
Fallback chain — Degrades gracefully if a provider goes down

🧬 What Makes Memoria Different

Feature	Memoria	Basic memory plugins
Memory layers	20 specialized layers	Single fact store
LLM choice	Any local or remote model	Usually hardcoded
Per-layer LLM config	✅ Different model per layer	❌
Procedural learning	✅ Learns HOW, not just WHAT	❌
Error detection	✅ Auto-captures dangers	❌
Knowledge graph	✅ Entities + relations	❌
Lifecycle management	✅ Smart aging, never forgets	❌ or simple TTL
Cost	$0 with local models	Varies

Installation

As Plugin (recommended — one command)

openclaw plugins install clawhub:memoria-plugin

This installs Memoria from the ClawHub registry. No manual steps needed.

From source (for contributors / advanced users)

If you prefer to inspect the code first:

Browse the repository: github.com/Primo-Studio/openclaw-memoria
Review the source code, especially index.ts (main entrypoint) and openclaw.plugin.json (config schema)
Clone and install:

cd ~/.openclaw/extensions
git clone https://github.com/Primo-Studio/openclaw-memoria.git memoria
cd memoria && npm install

Then add to your openclaw.json under plugins.entries:

{
  "memoria": { "enabled": true },
  "memory-convex": { "enabled": false }
}

Configuration

Minimal (works out of the box with Ollama)

Just install and restart. Defaults: Ollama + gemma3:4b for extraction, nomic for embeddings.

Custom LLM per layer

"memoria": {
  "enabled": true,
  "config": {
    "llm": {
      "default": { "provider": "ollama", "model": "qwen3.5:4b" },
      "procedural": { "provider": "lmstudio", "model": "your-model" },
      "graph": { "provider": "openai", "model": "gpt-4o-mini" }
    }
  }
}

Source Code

The full source is available on GitHub: Primo-Studio/openclaw-memoria

Key files:

index.ts — Main plugin entrypoint (hooks, extraction, recall pipeline)
procedural.ts — Procedural memory (how-to learning)
lifecycle.ts — Lifecycle management (fresh/settled/dormant)
scoring.ts — Temporal scoring and relevance ranking
selective.ts — Dedup, contradiction detection, fact quality
openclaw.plugin.json — Configuration schema

Feedback & Community

We'd love your feedback! Tell us how Memoria works for you:

🐦 Tweet us @Nitix_ — share your setup, results, or ideas
⭐ Star the repo: github.com/Primo-Studio/openclaw-memoria
🐛 Issues: GitHub Issues

Built with ❤️ by Primo Studio 🇬🇫 — AI tooling from French Guiana.

Memoria for OpenClaw

Safety Notice

Copy this and send it to your AI assistant to learn