How to stop n8n from cutting off long language model responses?

The most reliable way to stop n8n from cutting off long language‑model responses is to increase its internal payload limits (especially N8N_PAYLOAD_SIZE\_MAX) and make sure the AI node or HTTP Request node asking the model is not running into its own timeout or token limits. In real workflows, the cutoff almost always happens because n8n hits a payload/timeout ceiling, not because the model stops.

 

Why n8n cuts long LLM responses

 

n8n enforces limits to protect itself from huge JSON blobs. Two things matter the most:

• Payload size limit — n8n will silently truncate the incoming data if it exceeds N8N_PAYLOAD_SIZE\_MAX. This is the most common reason long LLM outputs get cut.
• Timeout limits — the node calling the model (AI node or HTTP Request node) may time out before the model finishes generating, causing early termination.

These are n8n server limits, not limits inside the workflow.

 

Fix Step 1: Increase N8N_PAYLOAD_SIZE\_MAX

 

In production (Docker or server), set a higher payload limit. For large LLM outputs, 16–64 MB is typical.

Example for Docker Compose:

environment:
  - N8N_PAYLOAD_SIZE_MAX=64mb // allows large JSON responses

Example for a plain environment variable:

export N8N_PAYLOAD_SIZE_MAX=64mb

Restart n8n after changing this. Without this step, n8n continues to cut responses no matter what you do inside the workflow.

 

Fix Step 2: Increase timeout for the node generating the response

 

If you use:

• HTTP Request node: Increase the Timeout field. If left too low (like 30s or 60s), long LLM outputs fail mid‑stream.
• OpenAI node / other AI nodes: These generally follow the global node timeout set in the n8n settings or environment variable N8N_DEFAULT_TIMEZONE. If using proxy model endpoints via HTTP Request, the HTTP timeout is the actual limit.

Example of setting HTTP timeout to 180 seconds:

// in the node UI:
// Timeout: 180000  (milliseconds)

 

Fix Step 3: Make sure your model itself is allowed to produce long outputs

 

Even if n8n can accept large responses, the model may be capped by:

• max\_tokens setting
• provider‑level request size limits

For example, if you call a model via HTTP Request node:

{
  "model": "gpt-4.1",
  "max_tokens": 8000,
  "messages": [
    { "role": "user", "content": "Write a long report..." }
  ]
}

If max\_tokens is too small, the model stops early regardless of n8n settings.

 

Fix Step 4: Use streaming carefully (or avoid it)

 

n8n does not support incremental-streaming LLM responses inside normal workflows. When you turn on streaming with some APIs, the server may send partial chunks and close the connection early. n8n often treats this as a complete response, which looks like a cutoff.

If you need truly long responses:

• Turn stream = false for OpenAI‑like APIs, so the full response arrives as one JSON.

 

Fix Step 5: If response becomes extremely large, store outside n8n

 

Once an LLM output gets bigger than ~30–50 MB, it’s safer to store the raw text in S3, database, or filesystem instead of flowing it through the n8n UI. n8n will accept the payload, but the UI might struggle to display it.

Pattern for this:

• LLM response arrives in n8n.
• Immediately save content to an external store via HTTP Request, S3 node, or database node.
• Pass only a link or an ID between steps.

 

Summary

 

If n8n is cutting off long LLM responses, increase N8N_PAYLOAD_SIZE_MAX, increase node timeouts, disable streaming, and ensure your model max_tokens is high enough. These are the real production‑level fixes. Once these limits are raised, n8n will reliably pass the full response through the workflow without truncation.