neutron-agent-memory

What This Skill Does

The neutron-agent-memory skill provides a robust, persistent memory layer for OpenClaw agents. It bridges the gap between ephemeral sessions and long-term knowledge retention by utilizing the Neutron API to store, index, and semantically retrieve information. The skill operates via dual-storage: it maintains structured Agent Contexts for session metadata and full-thread snapshots as Seeds. This ensures that your agent can recall not just isolated facts, but the full narrative flow of past interactions, enabling a highly context-aware experience that improves over time through append-only data updates.

Installation

To integrate this memory layer into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/naeemmaliki036/neutron-agent-memory

Ensure that you have your Neutron credentials configured. You can either export the NEUTRON_API_KEY, NEUTRON_APP_ID, and NEUTRON_EXTERNAL_USER_ID as environment variables or populate the ~/.config/neutron/credentials.json file. Once configured, verify connectivity by running ./scripts/neutron-memory.sh test.

Use Cases

• Long-Term Relationship Building: Allow your agent to remember user preferences, names, and prior conversation topics across distinct, non-contiguous chat sessions.
• Complex Task Tracking: Use procedural memory to save multi-step project progress, ensuring the agent doesn't lose track of active workflows when restarted.
• Knowledge Retrieval: Offload massive documentation or past technical discussions into semantic seeds, allowing the agent to perform similarity searches to answer "what do I know about [topic]" questions effectively.
• Automated Logging: Automatically snapshot every interaction thread to create a searchable "second brain" of all agent activities.

Example Prompts

1. "Store this project summary as a seed so I can reference it later during our sprint planning."
2. "Search your memory for any discussions we had about blockchain integration and summarize the main pain points."
3. "Create a new episodic context for our current session, storing the active user preferences as a JSON object."

Tips & Limitations

• Storage Growth: Since seeds are append-only, ensure you periodically review your stored data to prevent excessive token costs or noise if you are managing high-volume traffic.
• Context Truncation: While agent contexts are great for metadata, avoid stuffing them with large amounts of raw text; rely on the semantic search of seeds for bulky data retrieval.
• Semantic Precision: When performing searches, experiment with the sensitivity parameter (default 0.5) to balance between broad conceptual matches and exact keyword hits.