Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
v0.6.0: Phase 1/2/3 pipeline redesign — history-agnostic extraction (prompt in .md file), embedding-based deduplication (cosine similarity replaces LLM hints), Phase 3 relation sweep background job (auto-builds knowledge graph edges). Secret facts excluded from all LLM operations.
What This Skill Does
Memento adds persistent long-term memory to OpenClaw agents. It captures conversations, extracts structured facts via a configurable LLM, and injects relevant context before each agent turn. All data is stored locally in SQLite, and nothing is uploaded unless you opt into a cloud LLM for extraction.
Unlike cloud memory services, it keeps all data on-device in SQLite and supports fully local extraction via Ollama, so conversations never have to leave the machine.
When to use it
- Remembering user preferences across separate sessions
- Tracking decisions and action items from past conversations
- Sharing knowledge between multiple agents on the same machine
- Running fully offline memory using a local Ollama model
- Migrating existing markdown memory files into a structured knowledge base
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: 'What did we decide about the database schema last week?'
- 1Recall layer runs FTS5 keyword search against local SQLite facts
- 2Optional semantic embeddings surface related entries
- 3Matching facts are ranked by recency, frequency, and category weight
- 4Top facts are injected into the prompt via the before_prompt_build hook
- 5Agent responds using the retrieved context
Agent accurately references the past schema decision without the user restating it
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.