What I built
A knowledge-graph memory plugin for OpenClaw that turns conversation history into persistent entities and relationships.
The useful unit is not another note. It is a navigable map of people, projects, tools, and decisions.
Back to work
Memory
AgentSense
Knowledge-graph memory for OpenClaw.
AgentSense watches OpenClaw conversations, extracts people, projects, tools, and decisions, then stores them as a relationship graph the agent can query later.
Links
Problem
Chunked markdown memory can find what was said, but it struggles to answer who is connected to what, which tools belong to which project, and how past decisions relate.
Story
How it works
OpenClaw runs with AgentSense enabled, the extraction cron observes memory, entities are written into SQLite, and the agent can query the graph when context is needed.
The plugin keeps the existing memory tools intact while adding graph_search, a Telegram /graph command, and auto-recall behavior.
What it proves
The output is a relationship-aware memory layer that helps an agent recover context like project ownership, tool relationships, and decision history.