A bio-inspired memory system for AI agents that mimics human cognition by prioritizing frequently accessed information and naturally decaying irrelevant data.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install fluid-memory
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install fluid-memory using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Fluid Memory?
Fluid Memory serves as a sophisticated cyber-brain for your AI agent, moving beyond static database storage to a dynamic, living system. It implements a memory architecture based on the Ebbinghaus forgetting curve and access frequency decay models. By integrating this into Openclaw Skills, your agent gains the ability to strengthen important knowledge through repeated recall while naturally filtering out noise.
The skill is designed to function as a fluid layer over the native memory system, ensuring that your agent's context remains sharp and relevant. It distinguishes between active recall, which reinforces information, and passive decay, which prevents the cognitive load of outdated data. This makes it an essential component for developers building high-context, long-term AI companions or specialized assistants.
Fluid Memory Use Cases
- Persistent storage of user preferences that strengthen over time through natural conversation.
- Automated archival of transient information that hasn't been referenced in recent sessions.
- Improving long-term context retrieval for complex projects without manual data management.
- Creating more human-like AI interactions where the agent 'remembers' what is important and 'forgets' the trivial.
How Fluid Memory Works
- Implantation: The agent records new information through the fluid_remember tool or automatically during an OpenClaw native flush.
- Recall & Reinforcement: When a memory is retrieved via fluid_recall, the system increments an access counter, effectively 'practicing' the memory to prevent decay.
- Dynamic Decay: The system applies a decay model where memories with low relevance scores (below 0.05) are filtered out during active search.
- Dream Archival: A scheduled cleanup process, known as Dream Guard, archives memories with scores below 0.15 to keep the database optimized.
- Active Forgetting: Users can explicitly trigger the removal of specific information using keywords to maintain privacy or correct information.
Fluid Memory Setup
To use this skill within the ecosystem of Openclaw Skills, ensure you have the necessary Python environment and dependencies installed.
pip install chromadb pyyaml
Configuration Requirements:
- Enable
memoryFlushin your OpenClaw configuration. - Adjust the
softThresholdTokensto control the frequency of the auto-learn mechanism. - Ensure the agent has permission to access the local workspace directory.
Fluid Memory Data Schema & Taxonomy
Fluid Memory organizes data locally using a vector-based approach combined with metadata tracking. Data is stored in the ~/.openclaw/workspace/database/ directory in a transparent format.
| Metadata Field | Type | Description |
|---|---|---|
| Content | String | The actual text of the memory snippet. |
| Access Count | Integer | Number of times this memory has been successfully recalled. |
| Decay Score | Float | A calculated value based on the forgetting curve (0.0 to 1.0). |
| Last Accessed | Timestamp | Used to calculate the time elapsed since the last reinforcement. |
Fluid Memory Advanced Features
- Auto Learn Mode: Synchronizes directly with the OpenClaw native memory compaction cycle for zero-effort knowledge acquisition.
- Ebbinghaus Integration: Uses biological memory models to determine the lifespan of every stored data point.
- Dream Guard: Background scheduled archiving of low-utility memories to ensure high performance and relevance.
- Local Privacy: All memory data remains on your local machine with no cloud synchronization, ensuring complete data sovereignty within your Openclaw Skills setup.
SKILL.md
Loading
Related Openclaw Skills
Send USD
A specialized skill designed to facilitate secure and automated USD currency transfers between autonomous AI agents.
ASUSTOR Pro Adaptive Suite
A hardware-aware, hybrid SMB and SSH suite designed to scrape ASUSTOR NAS metadata while maintaining strict system safety and resource efficiency.
Adaptive Suite
A multi-functional AI agent skill that adapts to your workflow for development, business analysis, and data management tasks.
Scrapyard
An AI agent competition arena where bots battle in automated Floor is Lava matches every 15 minutes for real prizes.
ePragma Redmine Issue
A powerful interface for managing Redmine issues, projects, and time tracking entries through a secure REST API integration.
Agent Setup Survey
A diagnostic and telemetry tool for identifying if an AI agent is running in a sandbox, container, or on bare metal.
METADATA
Featured*
