Model Switcher

Added by Justin Sheehan Justin S.
Column
Done
Assigned to
Troy Pastoral Troy P.
Notes
ChatGPT
Claude
Gemini

With Knowledgebase, without knowledge base 
Subtasks
Configure and test knowledgebase toggle (see comment) *APP *APP, Justin Sheehan Justin S., Troy Pastoral Troy P. Tue, May 12
Update configuration for model switcher Troy Pastoral Troy P. Mon, May 18
Justin Sheehan
Justin Sheehan Chief Journey Officer
Troy can you make sure it defaults to have Knowledgebase turned on?

Also, I did some testing (with Knowledgebase turned OFF) and it was still pulling from its sources. 

Right now, the Knowledgebase is referring to the Daryle AI sources. 
Troy Pastoral
Troy Pastoral AI Whisperer
image.png Download

Auto resolution:
  • Provider Auto → OpenAI (grounded/auto modelOverride routes to OpenAI)
  • Tier Auto in pinecone-rag-chat → instant (when modelOverride is grounded/auto), else thinking
  • Tier Auto in route-orchestrator → thinking

Tier modifiers:
  • OpenAI reasoning_effort: instant=low, thinking=medium, pro=high
  • Anthropic extended thinking: thinking=1024 tokens, pro=2048 tokens (skipped on Haiku/Instant)
  • Instant tier: auto-enables fastMode, skips semantic cache, lighter Pinecone path, 800 max tokens vs 4000
Troy Pastoral
Troy Pastoral AI Whisperer Edited
KB Toggle:


Logs confirm:
knowledgeBaseEnabled: false, skipRAG: true, sourceCount: 0
🔕 [SKIP-RAG] Knowledge base bypassed
So the "3Cs / refine or define" content came from the raw model (GPT-4.1), not from Daryle's KB. No Pinecone retrieval, no citations. The session replay also shows a "Knowledge base off" divider at 02:51 PM.

If you expected KB-grounded answers, toggle the KB back on in the chat header.


Justin   try different modes + response, then send me the chat ID's to verify.