openclaw-memories

What This Skill Does

The openclaw-memories skill provides an intelligent, multi-layered memory architecture for OpenClaw agents. It integrates three distinct components: ALMA, Observer, and Indexer. ALMA (Algorithm Learning via Meta-learning Agents) leverages evolutionary computation to refine and optimize memory structure design. The Observer component acts as an extraction engine, utilizing external LLM APIs (OpenAI, Anthropic, or Gemini) to parse unstructured conversation data into structured, priority-ranked facts. Finally, the Indexer performs robust local full-text search across your workspace Markdown files, ensuring that your agent can reference daily logs, entity summaries, and core belief systems without needing a database server.

Installation

To add this memory system to your environment, use the OpenClaw repository manager:

clawhub install openclaw/skills/skills/arosstale/openclaw-memory-2

Alternatively, you can install the underlying library directly via npm:

npm install @artale/openclaw-memory

Ensure you have configured your environment variables if you intend to use the Observer component. You will need to provide an OPENAI_API_KEY or ANTHROPIC_API_KEY. The ALMA and Indexer modules are purely local and require no additional credentials.

Use Cases

• Long-term Persona Maintenance: By indexing bank/entities/*.md, the agent can recall past interactions with specific users or entities.
• Adaptive Memory Evolution: Use ALMA to iteratively improve how the agent stores and retrieves knowledge over time, ensuring the system grows more efficient as your workspace matures.
• Daily Insight Synthesis: Automatically summarize daily logs found in memory/YYYY-MM-DD.md to provide the agent with a 'briefing' at the start of a session.
• Fact Retrieval: Query the memory system for specific opinions or facts stored in the bank/opinions.md file, filtered by confidence levels.

Example Prompts

1. "Analyze our recent conversations and index any new facts about project timelines into the system."
2. "Search through my memory bank for all opinions on the latest documentation update and list them by confidence score."
3. "Run the ALMA proposer to generate a new memory storage architecture and display the current leaderboard metrics."

Tips & Limitations

• Performance: While the Indexer is powerful, it currently uses an in-memory mock database. For very large workspaces with thousands of Markdown files, expect slightly slower performance compared to a native SQLite solution.
• Network Dependency: The Observer component is the only part of this skill that requires network access to call LLM APIs. If your privacy requirements strictly prohibit outbound data, you can disable the Observer while still utilizing ALMA and the Indexer.
• Data Structure: Keep your MEMORY.md updated as the primary source of truth. The Indexer relies heavily on consistent Markdown formatting to ensure accurate search results.