Step 2: Connect LLM Providers in Parallel

Configure the three model builders that run simultaneously from the same chat input.

1. Connect Inbound Chat Trigger to OpenAI Response Builder, Gemini Response Builder, and Anthropic Response Builder. Inbound Chat Trigger outputs to both OpenAI Response Builder and Gemini Response Builder and Anthropic Response Builder in parallel.
2. In OpenAI Response Builder, set Model to gpt-4.1 and set the first Messages → Content to {{ $json.chatInput }}.
3. Credential Required: Connect your openAiApi credentials in OpenAI Response Builder.
4. In Gemini Response Builder, set Model to models/gemini-2.5-flash and set Messages → Content to {{ $json.chatInput }}.
5. Credential Required: Connect your googlePalmApi credentials in Gemini Response Builder.
6. In Anthropic Response Builder, set Model to claude-sonnet-4-5-20250929 and set Messages → Content to {{ $json.chatInput }}.
7. Credential Required: Connect your anthropicApi credentials in Anthropic Response Builder.

Step 3: Normalize and Merge Model Responses

Standardize each model’s response payload and merge all three into a single stream for evaluation.

1. In Normalize OpenAI Output, set provider to OpenAI and response to {{ $json.message.content }}.
2. In Normalize Gemini Output, set provider to Gemini and response to {{ $json.content.parts[0].text }}.
3. In Normalize Anthropic Output, set provider to Anthropic and response to {{ $json.content[0].text }}.
4. Connect all three normalize nodes into Merge Model Replies and set Number of Inputs to 3.

Step 4: Attach Tests, Score, and Batch

Generate evaluation checks for each response, batch them, and score each test with Contextual AI.

1. In Attach Evaluation Checks, keep the provided JavaScript that expands each response into tests like Is the response clear and easy to understand?.
2. Send output to Batch Through Tests to process tests sequentially; keep default batch options.
3. From Batch Through Tests, connect one output to Compose Summary Message and the other to Pause Three Seconds to throttle scoring.
4. In Pause Three Seconds, set Amount to 3 seconds, then connect to Execute LMUnit Scoring.
5. In Execute LMUnit Scoring, set Resource to LMUnit, Query to {{ $('Inbound Chat Trigger').first().json.chatInput }}, Response to {{ $json.response }}, and Unit Test to {{ $json.unit_test }}.
6. Credential Required: Connect your contextualAiApi credentials in Execute LMUnit Scoring.
7. In Map Scores to Replies, map provider to {{ $('Attach Evaluation Checks').item.json.provider }}, response to {{ $('Attach Evaluation Checks').item.json.response }}, unit_test to {{ $('Attach Evaluation Checks').item.json.unit_test }}, and score to {{ $json.score }}.

Step 5: Aggregate Scores and Deliver the Reply

Aggregate test results per provider, format a summary message, and reply to the chat.

1. In Aggregate Evaluation Sets, keep the provided JavaScript that groups results by provider and unit_test.
2. Connect Map Scores to Replies to Aggregate Evaluation Sets and then to Batch Through Tests to continue the loop until all tests are processed.
3. In Compose Summary Message, keep the JavaScript that builds the final message string in message.
4. In Deliver Chat Reply, set Message to {{ $json.message }} so the user receives the evaluation report.

Step 6: Test and Activate Your Workflow

Validate end-to-end behavior in the editor, then activate the workflow for production use.

1. Click Execute Workflow and send a test message to Inbound Chat Trigger.
2. Confirm that OpenAI Response Builder, Gemini Response Builder, and Anthropic Response Builder all return outputs and that Merge Model Replies completes.
3. Verify scoring runs through Execute LMUnit Scoring and that Deliver Chat Reply returns a message starting with Here are the evaluation results:.
4. Toggle the workflow to Active to enable production execution.