memory-manager

What This Skill Does

The Memory Manager Skill is a sophisticated three-tier architecture designed to endow OpenClaw AI agents with persistent, context-aware intelligence. It replicates human-like memory patterns by segregating information into three distinct functional layers. The Short-Term Memory utilizes a sliding window mechanism (FIFO) to maintain immediate, coherent conversational flow without overwhelming the model's context window. The Mid-Term Memory acts as a compression layer, automatically generating summaries of past interactions once specific token thresholds are met, ensuring historical context is preserved in a lightweight format. Finally, the Long-Term Memory leverages ChromaDB for vector-based semantic retrieval (RAG), allowing the agent to recall specific facts, user preferences, or historical data across multiple, disconnected sessions, effectively creating a persistent knowledge base for the agent.

Installation

To integrate this skill into your environment, run the following command via the OpenClaw CLI:

clawhub install openclaw/skills/skills/forvendettaw/monica-memory-manager

After installation, navigate to the memory/ directory to configure the config.yaml file according to your specific storage requirements, such as adjusting the window_size for short-term recall or the summary_threshold for automated summarization triggers.

Use Cases

• Personalized Assistants: Remember user preferences, past habits, or specific instructions given weeks ago.
• Long-running Research: Maintain focus across multiple sessions by pulling in relevant information from historical data.
• Knowledge Management: Build a searchable database where the AI can retrieve technical details or notes using semantic search.
• Workflow Continuity: Resume complex tasks or coding sessions without needing to re-explain the project context from scratch.

Example Prompts

1. "Check your long-term memory and recall the specific project requirements I mentioned last Tuesday regarding the API architecture."
2. "Summarize the current conversation state and store any critical insights into the vector database for future reference."
3. "Search through all previous interactions to find out what my stated preferences are for theme colors and font sizes."

Tips & Limitations

• Resource Management: Keep the summary_threshold balanced. Too low creates excessive summary files, while too high may consume significant tokens in your prompt.
• Semantic Accuracy: The long-term memory retrieval relies on embedding models. Ensure your min_relevance threshold is tuned to avoid hallucinations or irrelevant noise during search.
• Storage: Note that this skill performs local file system operations. Ensure your host environment has sufficient disk space for the vector store index, especially for large knowledge bases. For privacy-conscious setups, remember that memory is stored locally in the memory/ directory, so back up this folder regularly to prevent data loss.