fluid-memory

What This Skill Does

Fluid Memory is an advanced cognitive layer for the OpenClaw AI agent, simulating human-like memory dynamics based on the Ebbinghaus forgetting curve. Instead of acting as a static database, it dynamically evaluates the importance of information based on access frequency and recency. Information that is frequently referenced gains 'strength,' while stale or unimportant details are automatically subjected to decay. This system is fully integrated with OpenClaw’s native memory flushing mechanism, ensuring that as your workspace grows, your agent learns what is truly relevant to your unique workflow, effectively mimicking how a biological brain manages focus.

Installation

To integrate Fluid Memory into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/againta/fluid-memory Ensure that you have enabled memoryFlush in your OpenClaw configuration file to allow the system to synchronize memories automatically. You may also adjust the softThresholdTokens parameter in your OpenClaw settings to control how frequently the agent performs compaction and memory maintenance.

Use Cases

• Contextual Knowledge Retention: Store personal project preferences or technical requirements that the agent should consistently apply.
• Dynamic Workflow Optimization: Allow the agent to automatically 'forget' outdated task statuses while maintaining focus on current priorities.
• Relationship Management: Perfect for tracking specific user feedback, dislikes, or preferences across long-term sessions without bloating the core context window.

Example Prompts

1. "I'm starting a new Python project, please remember that I prefer using Pydantic for data validation from now on."
2. "Do you recall what specific naming convention I asked you to follow for my database schemas?"
3. "Forget the previous instructions about using Tailwind CSS; I've decided to switch to standard CSS for this site."

Tips & Limitations

• Proactive Strengthening: Since memory importance is based on access frequency, periodically asking the agent about core project details will help 'cement' those memories, ensuring they aren't filtered out by the decay mechanism.
• Acceptance of Forgetting: Trust the system. If the agent fails to recall a detail, it is likely because that information was deemed low-priority. This is by design to maintain agent speed and accuracy.
• Privacy Note: Data is stored locally in ~/.openclaw/workspace/database/ in plaintext. Ensure your local machine is secure as there is no native encryption or cloud-based synchronization.