How to Fix ETIMEDOUT Errors When Calling a Large Language Model in n8n

The default HTTP timeout in n8n is 300 seconds (5 minutes). For self-hosted instances, increase this by setting the N8N_HTTP_TIMEOUT environment variable. Set it to 600000 (10 minutes in milliseconds) or higher for workflows that make complex LLM calls. In Docker, add this to your docker-compose.yml environment section. For n8n Cloud, this setting is managed by n8n and cannot be changed directly — use the HTTP Request node with a custom timeout instead.

1# Docker Compose environment variable
2services:
3  n8n:
4    image: n8nio/n8n:latest
5    environment:
6      - N8N_HTTP_TIMEOUT=600000
7      # Also set the generic Node.js timeout
8      - NODE_OPTIONS=--max-http-header-size=16384

The simplest way to avoid timeouts is to make the LLM respond faster. Reduce the max_tokens or 'Maximum Number of Tokens' setting in your LLM node to limit response length. For OpenAI, set it in the node options. For Anthropic Claude, set max_tokens in the Anthropic Chat Model sub-node. Also simplify your system prompt — remove unnecessary instructions, examples, and formatting requirements. Each additional instruction increases processing time. If your prompt includes few-shot examples, reduce them to 1-2 instead of 5-6.

For maximum control over timeout behavior, bypass the dedicated LLM nodes and use the HTTP Request node to call the API directly. The HTTP Request node has an explicit timeout setting in its Options. Set it to the maximum you need. This approach also lets you set custom headers, handle streaming responses, and implement more granular error handling. Configure the request body to match the API's expected format.

1// HTTP Request node configuration for OpenAI
2// Method: POST
3// URL: https://api.openai.com/v1/chat/completions
4// Authentication: Header Auth
5//   Name: Authorization
6//   Value: Bearer {{ $credentials.openAiApi.apiKey }}
7// Body (JSON):
8{
9  "model": "gpt-4o",
10  "messages": [
11    {
12      "role": "system",
13      "content": "You are a helpful assistant."
14    },
15    {
16      "role": "user",
17      "content": "{{ $json.user_message }}"
18    }
19  ],
20  "max_tokens": 2048,
21  "temperature": 0.7
22}
23// Options → Timeout: 120000 (120 seconds)

Timeouts are often transient — the same request may succeed on a second attempt if the API server is less loaded. On your LLM node or HTTP Request node, go to Settings and enable 'Retry On Fail'. Set Max Tries to 3 and Wait Between Tries to 5000 ms. This gives the API a chance to recover between attempts. For production workflows, also consider adding an exponential backoff pattern using a Code node and a Loop.

1// Code node for exponential backoff retry (advanced)
2// Use this inside a loop workflow for custom retry logic
3
4const maxRetries = 3;
5const baseDelay = 2000; // 2 seconds
6const currentRetry = $json.retry_count || 0;
7
8if ($json.status === 'timeout_error' && currentRetry < maxRetries) {
9  const delay = baseDelay * Math.pow(2, currentRetry);
10  // Use the Wait node after this Code node with the calculated delay
11  return [{
12    json: {
13      ...($json),
14      retry_count: currentRetry + 1,
15      wait_ms: delay,
16      should_retry: true
17    }
18  }];
19}
20
21return [{
22  json: {
23    ...($json),
24    should_retry: false,
25    final_status: $json.status === 'timeout_error' ? 'failed_after_retries' : $json.status
26  }
27}];

If your workflow is triggered by webhooks and needs to respond within a tight window, consider using a faster model. GPT-4o-mini is significantly faster than GPT-4o. Claude 3.5 Haiku is faster than Claude 3.5 Sonnet. Gemini Flash is faster than Gemini Pro. You can also implement a tiered approach: try the fast model first, and only fall back to the larger model if the task requires it. Use an IF node to route simple vs complex requests to different LLM nodes.