mnemon

What This Skill Does

mnemon is a robust, persistent memory subsystem for OpenClaw AI agents, functioning as a structured long-term storage and retrieval layer. Unlike ephemeral context windows, mnemon provides a CLI-driven backend that enables agents to store factual data, track causal relationships between memories, and manage their knowledge lifecycle through semantic indexing. It acts as an extension of the agent's cognition, allowing it to remember specific entities, assign importance levels, and execute garbage collection to prune stale or redundant information. By integrating via OpenClaw hooks, mnemon ensures that relevant past knowledge is automatically recalled or nudged into the agent's active processing stream.

Installation

Installation is streamlined for both macOS/Linux via Homebrew or standard Go workflows. To install the binary, run brew install mnemon-dev/tap/mnemon or go install github.com/mnemon-dev/mnemon@latest. Once the binary is present, initialize the integration with mnemon setup --target openclaw --yes. This command deploys the necessary hooks, plugins, and prompt management files into your ~/.openclaw directory. After setup, restart the OpenClaw gateway to enable the persistent memory loop. You can further customize the agent's behavior by editing the guide.md prompt file or toggling features like remind and nudge within your openclaw.json configuration.

Use Cases

mnemon is designed for complex, long-running agent tasks. Use it to maintain context for large-scale software engineering projects where requirements evolve over weeks, or to keep track of user preferences and recurring project-specific constraints. It is ideal for research workflows where linking related facts—such as causal connections between documentation updates and bug fixes—is critical for the agent's reasoning. By utilizing the link and gc commands, users can ensure their agent remains performant, focused, and historically aware, avoiding the common pitfalls of context degradation.

Example Prompts

1. "mnemon remember 'The production API endpoint uses a bearer token strategy for authentication' --cat 'infrastructure' --imp 5 --entities 'api,security'"
2. "mnemon recall 'How do we handle production authentication?' --limit 5"
3. "mnemon link 101 105 --type causal --weight 0.9 --meta '{"reason": "Authentication fix caused secondary service latency"}'"

Tips & Limitations

Always use judgment when linking memories; causal signals are based on patterns and can produce false positives. Focus on semantic relevance rather than keyword frequency to maintain high-quality retrieval. Regularly utilize the garbage collection command mnemon gc with a threshold to prevent the database from ballooning with low-priority, redundant data. The system excels at structured data, so take advantage of categories and entity tagging to keep your memory store organized and searchable.