Clawvault

What This Skill Does

ClawVault is a comprehensive memory management system for OpenClaw agents, designed to act as an external "brain" for persistent information storage. It enables agents to move beyond simple context windows by utilizing a memory graph, structured storage, and semantic search. The skill operates locally, reading and writing markdown files within a defined vault directory, effectively preventing context death and ensuring long-term memory retention. It includes advanced capabilities such as session repair, which automatically identifies and fixes broken session transcripts, and observational memory, which uses LLM-powered compression to summarize complex interactions into distilled knowledge. ClawVault also integrates into the agent lifecycle via hook-based events (e.g., heartbeat, startup, and memory flushing), allowing for autonomous maintenance of the agent's knowledge base.

Installation

To install ClawVault, ensure you have the OpenClaw environment ready. The canonical installation method is via the OpenClaw hub: clawhub install openclaw/skills/skills/g9pedro/clawvault. This process installs the necessary CLI tool and the required backend via qmd. You should verify that your environment variables, specifically CLAWVAULT_PATH, are configured if you want the vault to reside in a custom location, though the system will attempt auto-discovery by default. If you intend to use the advanced observation/compression features, ensure GEMINI_API_KEY is set in your environment configuration.

Use Cases

Use ClawVault for persistent persona development, complex project tracking across multiple sessions, or maintaining a knowledge base of research notes that your agent needs to reference over weeks or months. It is ideal for developers who want to prevent their agents from forgetting previous architectural decisions. It should not be used for general file I/O tasks or moving large binary datasets, as it is strictly optimized for text-based semantic memory management.

Example Prompts

1. "Store these findings about the API integration in my memory graph so I don't have to look them up next time."
2. "Search my vault for any mentions of project milestones from last month and summarize them for this current task."
3. "Run a session repair on my latest transcript to fix the context drift I encountered during the last iteration."

Tips & Limitations

• Security: ClawVault performs local file system writes. Always review the hooks/clawvault/handler.js file before enabling hooks if you are running in a restricted environment.
• Data Privacy: All data remains local. Only the observe command, when compressing transcripts, will communicate with external LLM APIs (Gemini).
• Best Practices: Use descriptive tags when saving memories to maximize the efficiency of the semantic search functionality. Avoid storing sensitive credentials directly in plain text vaults.