How to Structure Fallback Messages if a Model Fails in n8n

1// When Continue On Fail is enabled, a failed node outputs:
2// {
3//   "json": {
4//     "error": {
5//       "message": "429 Too Many Requests",
6//       "description": "Rate limit exceeded. Please retry after 60 seconds."
7//     }
8//   }
9// }
10//
11// A successful node outputs normal data:
12// {
13//   "json": {
14//     "choices": [{"message": {"content": "Hello! How can I help?"}}]
15//   }
16// }

1// IF node configuration
2// Condition: Check if the response contains an error
3//
4// Value 1: {{ $json.error }}
5// Operation: is not empty
6//
7// True branch  → Fallback message path (error occurred)
8// False branch → Normal processing path (success)

1// Set node on the error branch — create a friendly fallback
2// Field: response
3// Value:
4
5I'm sorry, I'm unable to process your request right now. Please try again in a moment. If the issue persists, our team has been notified and is looking into it.
6
7// Or use a Code node for more control:
8const errorMessage = $json.error?.message || 'Unknown error';
9
10let fallbackResponse;
11if (errorMessage.includes('429') || errorMessage.includes('rate limit')) {
12  fallbackResponse = 'I\'m receiving a lot of requests right now. Please try again in about 30 seconds.';
13} else if (errorMessage.includes('timeout')) {
14  fallbackResponse = 'My response took too long to generate. Please try a shorter question.';
15} else {
16  fallbackResponse = 'I\'m temporarily unable to respond. Please try again shortly.';
17}
18
19return [{
20  json: {
21    response: fallbackResponse,
22    isError: true,
23    errorType: errorMessage
24  }
25}];

1// Backup model pattern:
2// Primary: OpenAI GPT-4 → IF (error check) → True: Backup Model
3//                                           → False: Normal response
4
5// Backup Code node to prepare the prompt for the fallback model:
6const originalPrompt = $('Webhook').first().json.body.message;
7
8return [{
9  json: {
10    prompt: originalPrompt,
11    model: 'gpt-3.5-turbo',
12    systemMessage: 'You are a helpful assistant. The user\'s question may have been intended for a more capable model. Do your best to provide a useful response.'
13  }
14}];
15
16// Connect this to a second OpenAI node configured with GPT-3.5-turbo
17// Add another IF node after the backup model in case it also fails

1// Code node: Prepare error log entry (connect in parallel to fallback)
2const errorData = $json.error || {};
3const userInput = $('Webhook').first().json.body?.message || 'unknown';
4
5return [{
6  json: {
7    level: 'error',
8    source: 'llm-fallback',
9    message: errorData.message || 'Unknown LLM error',
10    userInput: userInput.substring(0, 200),
11    timestamp: new Date().toISOString(),
12    executionId: $execution.id,
13    workflowName: $workflow.name
14  }
15}];
16
17// Connect this to:
18// - Slack node (post to #errors channel)
19// - Send Email node (alert the team)
20// - PostgreSQL node (insert into error_logs table)