n8n Integration Guide

Quick Start

LLM Resayil is OpenAI-compatible — use the HTTP Request node in n8n to call it directly.

Method:         POST
URL:            https://llmapi.resayil.io/v1/chat/completions
Authentication: Header Auth
  Header Name:  Authorization
  Header Value: Bearer YOUR_API_KEY
Content-Type:   application/json

{
  "model": "qwen3:14b",
  "messages": [
    {
      "role": "user",
      "content": "Summarize: {{ $json.text }}"
    }
  ],
  "max_tokens": 1000,
  "temperature": 0.7
}

Streaming Configuration

Streaming sends tokens as they're generated, reducing perceived latency. Essential for large models that take time to produce the first token.

Enable Streaming

Add "stream": true to your request body:

{
  "model": "deepseek-v3.2",
  "messages": [
    {"role": "user", "content": "Write a detailed analysis..."}
  ],
  "stream": true,
  "max_tokens": 4000
}

n8n Node Settings for Streaming

• In HTTP Request node Options, set Response Format to Stream
• Set Timeout to at least 300000 ms (5 min) for standard models
• For 200B+ models, set timeout to 600000 ms (10 min)

Timeout Recommendations

Larger models need more time for first-token generation. Our API supports up to 10 minutes for the largest models. Set your n8n timeout accordingly.

Rate Limits

LLM Resayil enforces per-minute and per-burst limits. Exceeding either returns 429 Too Many Requests.

See the full Rate Limits documentation for response headers and backoff strategies.

Error Handling

Build resilient workflows by handling common error codes. Use n8n's Retry on Fail option or add explicit error handling nodes.

Retry Configuration

In the HTTP Request node settings:

• Retry on Fail: Enabled
• Max Tries: 3
• Wait Between Tries: 10,000 ms
• Continue on Fail: Enable to handle errors gracefully downstream

Best Practices

Choose the Right Model

• Simple tasks (summarization, classification): qwen3:14b or llama3.2:3b — fast, cheap (0.5x credits)
• Complex reasoning (analysis, multi-step): deepseek-v3.2 or llama3.1:70b
• Code generation: qwen2.5-coder:14b or devstral-2:123b

Optimize Costs

• Set max_tokens to limit response length — saves credits on long outputs
• Small models (3B–14B) use 0.5x–1x credits; XL models use 3.5x
• Process batch items sequentially with delays, not in parallel

Workflow Design

• Always set a timeout matching the model size (see table above)
• Use streaming for responses longer than a few sentences
• Don't fire parallel requests beyond your burst limit — use Split In Batches with delays
• Validate responses — check that choices[0].message.content exists and is non-empty
• Log errors to a Google Sheet or database for debugging

Example: Text Summarizer

Complete HTTP Request node configuration for a summarization workflow:

{
  "model": "qwen3:14b",
  "messages": [
    {
      "role": "system",
      "content": "You are a professional summarizer. Output only the summary."
    },
    {
      "role": "user",
      "content": "Summarize in 3 bullet points:\n\n{{ $json.input_text }}"
    }
  ],
  "max_tokens": 500,
  "temperature": 0.3
}

Troubleshooting