Walkthrough · LLM routing
The right model
for each step. No code.
Four LLM calls in one Reaction. Each one resolves through the precedence chain to a different Profile - Haiku for the cheap summary, GPT-4o for the hard reasoning, Sonnet for the recommender. The 30-day spend dashboard groups the bill by profile, engine, and tier.
Walkthrough · 1 of 4
A platform admin sets up five LLM Profiles. Each is a named bundle of provider, model, temperature, and the Connection that holds the key.
quadrazene.app/admin/llm/profiles
quadrazene v0.2.1
Quadrazene
ReactorComposeRecipesInboxGovernanceRecords⌘K
LLM Profiles · 5
Haiku · fast + cheap
anthropic · claude-haiku · temp 0.2
Sonnet · balanced
anthropic · claude-sonnet · temp 0.4
GPT-4o · deep-reasoning
openai · gpt-4o · temp 0.3
Gemini 1.5 Pro
google · gemini-1.5-pro · temp 0.4
Ollama · on-prem (regulated)
ollama · llama-3-70b-instruct · temp 0.3
Live walkthrough — click a button or use the rail to navigate.
What you just saw