Claude-Mem OpenClaw Plugin — Setup Guide

Persistent memory across sessions is achieved by running a claude-mem worker on port 37777 to store agent observations and inject the timeline into system prompts for context retention. This setup ensures agents recall past interactions without manual data management.

System prompt context injection works by using the before_prompt_build hook to automatically insert stored agent observations into the prompt context. This mechanism provides governance-aware longitudinal memory so agents access prior session data seamlessly.

Yes, you can stream agent observations to Discord, Slack, or Telegram by connecting the worker's SSE endpoint to your messaging channels. This setup provides real-time visibility into what agents learn during their operations.

The best way to set up the Claude-Mem plugin is to run the one-liner installer and follow the interactive prompts. This process configures the worker connection, enables the plugin entry, and sets project and feed parameters on your OpenClaw gateway.

Yes, an OpenClaw deployment is required because the plugin is designed specifically for the OpenClaw gateway environment. It relies on OpenClaw hooks to inject observation timelines and stream new data via SSE to configured messaging channels.

Verify the persistent memory worker and SSE feed by running the /claude_mem_status command to check worker health and the /claude_mem_feed command to inspect the feed connection state. These operations confirm the system is actively recording and streaming observations.