Agents need a second brain beyond chat memory

Tonbi's AI Garage shows Gbrain wired into Hermes Agent as a second-brain layer: Markdown in a Git repo, synced to local PGlite, exposed through MCP, and structured around people, companies, meetings, decisions, citations, and relationships. Why it matters: Chat memory only covers what the agent has discussed with you. A second brain gives agents durable world context they can search, cite, update, and share with humans — especially for meetings, decisions, company knowledge, and long-running projects.

Try this: Pick one agent workflow and separate conversation memory from world memory: keep durable facts and decisions in readable Markdown, make the agent search that brain first, require cited answers, and write important decisions back instead of burying them in chat history.