Step 2: Initialize and Iterate the LLM Index

These nodes set a starting model index and create a loop to try different models if needed.

1. Open Assign LLM Index and set llm_index to {{ $json.llm_index || 0 }}.
2. Open Increment Model Index and set llm_index to {{ $('Assign LLM Index').item.json.llm_index + 1 }}.
3. Ensure Increment Model Index connects to No-Op Step and back to Assign LLM Index to form the retry loop.

Step 3: Set Up LLM Selection and Response Composition

This step routes the request to a specific model and generates the response text.

1. Open Select LLM Engine and keep the code as provided (it selects the language model based on llm_index).
2. Connect OpenAI 4o Mini, OpenAI 4o Core, and OpenAI o1 Core to Select LLM Engine as ai_languageModel inputs.
3. In Compose Reply Text, set Text to {{ $('Incoming Chat Trigger').item.json.chatInput }} and keep Prompt Type as define.
4. Confirm Compose Reply Text includes the system message: You’re an AI assistant replying to a customer who is upset about a faulty product and late delivery. The customer uses sarcasm and is vague. Write a short, polite response, offering help.

Credential Required: Connect your openAiApi credentials in OpenAI 4o Mini, OpenAI 4o Core, and OpenAI o1 Core. OpenAI Chat Engine is connected as the language model for Assess Response Quality—ensure credentials are added to OpenAI Chat Engine.

Step 4: Evaluate Response Quality and Format Output

The workflow analyzes the generated response and either outputs it or triggers the next model attempt.

1. Open Assess Response Quality and set Input Text to {{ $json.text }}.
2. Verify the System Prompt Template specifies categories pass, fail and the evaluation criteria for response quality.
3. Open Send Output Result and set output to {{ $json.text || $json.output }} so the workflow returns the final response or fallback text.

Execution Flow: Compose Reply Text → Assess Response Quality → Send Output Result for successful responses, while failures move to Increment Model Index and loop back.

Step 5: Add Error Handling Logic

This branch captures expected errors from LLM selection and provides safe fallback messages.

1. Open Validate Expected Error and confirm the condition checks {{ $json.error }} equals Error in sub-node Select LLM Engine.
2. In No Match Loop End, set output to The loop finished without a satisfying result to handle known errors gracefully.
3. In Handle Unexpected Error, set output to An unexpected error happened for all other error cases.

Step 6: Test and Activate Your Workflow

Validate each stage to ensure the model routing and quality checks work as expected.

1. Click Execute Workflow and send a sample message to Incoming Chat Trigger.
2. Confirm Compose Reply Text produces a short, polite response and Assess Response Quality returns pass when criteria are met.
3. Verify Send Output Result outputs the final response text.
4. Toggle the workflow to Active to enable production usage.