OzBrain is a shared brain for knowledge between AI agents and human teams, designed to replace manual context ferrying across tools. The creator, Bubs.co, argues agent-first chat will become the primary software modality, and current note systems weren't built for agents as primary users. OzBrain centralizes agent-reasoned knowledge, stays app-agnostic, tracks changes for audit, handles multi-agent conflicts, refactors knowledge into token-friendly chunks, and supersedes old thinking without erasure.

The project emerged from personal pain: after building a Voice AI using agentic workflows across Cursor, Codex, and Claude, shuffling .md files became unmanageable. A survey of 75 founder friends revealed 26 built custom knowledge systems (Obsidian vaults, markdown repos, cron jobs) while 32 felt the pain but had no solution. The creator compares OzBrain to Vercel versus gBrain's AWS — easy, managed, and secure.

Key features include an MCP connector for instant agent access, encrypted-at-rest storage with per-account keys, full audit logs, row-level security with instant revoke, and automatic conflict detection when agents write contradictory updates. The system routes new knowledge to the right article, splits oversized articles, and exports everything as CSV. Currently in alpha, the maintenance loop isn't running on customer data yet.