TheLoosely Organized Research Facility
Investigating the bleeding edge of agentic communication — A2A, MCP, agent coordination, and the infrastructure that makes multi-agent systems actually work.
Active Projects
Project: Nexus
Status: Active • Classification: Open Source
A coordination server for multi-agent engineering teams. Distributed file locking, institutional memory, crash recovery, and real-time coordination via Redis pub/sub and WebSockets.
FILE CLAIMS
Redis-backed distributed locks on file paths with TTL-based lease expiration
LEARNINGS
Append-only knowledge base for AI agents across sessions and features
CHECKPOINTS
Session state snapshots for crash recovery and progress tracking
REAL-TIME
Redis pub/sub + WebSocket broadcast across all connected instances
Tech: Bun • Hono • PostgreSQL • Redis • Drizzle • Zod
Latest Thoughts
Why Your AI Agents Keep Overwriting Each Other
The problem statement that motivated Nexus — what happens when multiple agents touch the same codebase without coordination.
Designing Agent-Native Interfaces
How Claude Code skills expose coordination as natural language commands — and why the proactive agent pattern changes everything.
Latest Research
From Aesthetic to Algorithm: Building Design Systems as Agent Skills
How we turned a visual aesthetic into a machine-readable skill — from stock ticker inspiration to character bibles to executable design systems that agents can follow consistently.
Distributed Locking for AI Agent Coordination
How Redis-backed file claims solve the coordination problem when multiple AI agents work on the same codebase simultaneously.
Building Institutional Memory for AI Agents
AI agents are stateless across sessions. Here's how we built an append-only knowledge base that gives agents persistent memory tied to features and decisions.