lily-memory

What This Skill Does

Lily Memory acts as a persistent, high-performance long-term memory layer for OpenClaw agents. Unlike ephemeral context windows that clear upon session restarts, Lily Memory leverages a hybrid approach combining SQLite FTS5 for rapid keyword retrieval and Ollama-powered vector embeddings for nuanced semantic understanding. The system features autonomous lifecycle management, including auto-capture for identifying critical facts within chat logs, deduplication through periodic memory consolidation, and stuck-detection to prevent the agent from looping on repetitive topics. It operates as a zero-dependency plugin, ensuring minimal overhead while providing robust recall capabilities.

Installation

Installation is straightforward using the OpenClaw plugin manager. First, ensure your environment meets the requirements: Node.js 18+ and a standard SQLite installation with FTS5 support. Run the command: clawhub install openclaw/skills/skills/kevinodell/lily-memory-plugin. Once installed, navigate to your openclaw.json configuration file and add lily-memory to your plugin slots. Define your database path and initial allowed entities within the configuration object, then restart your gateway using openclaw gateway restart. The plugin will automatically handle initial schema creation and data consolidation upon its first run.

Use Cases

Lily Memory is ideal for agents that act as long-term personal assistants or project managers. Use it to maintain context across multi-day coding sessions where specific project configurations or system preferences must be remembered. It is also highly effective for knowledge management, allowing agents to ingest complex documentation or meeting notes and surface specific facts when queried months later. Organizations can use it to maintain 'institutional memory' by allowing the agent to continuously learn new internal protocols simply through standard conversation.

Example Prompts

1. "What was the specific API key format we decided on for the project documentation last week?"
2. "Search through all recent notes and retrieve the current server status for the staging environment."
3. "I am tracking a new entity called 'MarketingBudget' with a current limit of $5000, please add this to memory."

Tips & Limitations

To maximize performance, keep your entities list focused, as excessively broad allowlists can increase search latency. If you are operating without an Ollama instance, the skill will gracefully degrade to keyword-only search; while functional, this loses the ability to perform semantic 'fuzzy' lookups. Ensure your dbPath resides in a persistent directory, especially when running inside containerized environments, to prevent losing your database state during volume resets. Finally, monitor the stuckDetection logs if your agent frequently repeats itself, as this indicates a need to tune your recall sensitivity.