soul
.py
v2.0
RAG + RLM hybrid — automatic query routing
📦 PyPI
v0.1 demo
★ GitHub
Architecture
Retrieval mode:
⚡ Auto (router decides)
🔍 RAG only
🔄 RLM only
Session 1
↻ New Session
Query routing in action
Auto mode uses a lightweight LLM classifier to route each query. Watch the Router panel to see why.
Step 1
— Tell the agent several things about yourself
Step 2
— Ask a specific question → routes to RAG
Step 3
— Ask for a summary/pattern → routes to RLM
Step 4
— Try forcing RAG/RLM mode to compare
My name is Prahlad and I run an AI research lab in Pittsburgh.
→ RAG
I enjoy driving fast cars and flying small planes.
→ RAG
What car do I own?
→ RAG
Summarize everything you know about me across all our conversations.
→ RLM
What patterns appear in the things I've told you?
→ RLM
Send
⚡ Router Decision
waiting for query...
Query classified as
—
Router
—
Retrieval
—
Total
—
RLM sub-summaries
Retrieved Context
Full MEMORY.md
Ask something to see what was retrieved...
# MEMORY.md (No memories yet)
↺ Reset everything