PAIE Curator

A Local LLM Listener That Learns With You

---

🚀 What This Is

This is the first operational node in a Mesh.
A listener. A watcher. A curator of intelligence.

It connects to a Redis queue and listens for escalated LLM failures — things your models couldn’t classify, or got wrong, or needed help thinking about again. It’s schema-aware. Human-aligned. And model-agnostic.

You don’t need our front-end. You don’t need a massive model.
All you need is a model that knows when it’s unsure, and this listener will do the rest.

---

🧠 Why We Built This

Most current "MCP" systems focus on RAG, vector search, or agent orchestration.
We chose something simpler. Structured escalation — one of the most overlooked tools in LLM engineering.

If you want smarter models, you don’t need bigger ones.
You need better memory.
You need a place for the misses to go.
That’s what this is.

We built it because we were tired of everyone skipping the hard part: structured thought.

---

📦 What’s Inside

• core/: Schema router, fallback handler, structured ingest logic
• mcp/: Real-time Redis listener (mcp_listener.py)
• main.py: Entry point if you want CLI testing
• check_escalations.py: View what your models escalated
• fail_log.jsonl: Output of escalated records

---

🛠 How to Use

1. Start a Redis server locally (default: localhost:6379)

2. Run the listener:

   ./start_listener.sh