How to Handle Incomplete Responses from Cohere in n8n

1const json = $json;
2const text = json.text || json.output || json.message?.content || '';
3const finishReason = json.finish_reason || json.meta?.finish_reason || '';
4
5// Check for obvious incompleteness indicators
6const indicators = {
7  max_tokens_hit: finishReason === 'MAX_TOKENS' || finishReason === 'COMPLETE_LENGTH',
8  ends_mid_sentence: text.length > 0 && !text.trim().match(/[.!?}\]"']$/),
9  unclosed_json: (text.match(/\{/g) || []).length > (text.match(/\}/g) || []).length,
10  unclosed_array: (text.match(/\[/g) || []).length > (text.match(/\]/g) || []).length,
11  very_short: text.length < 50 && !json.error,
12  has_error: !!json.error
13};
14
15const isIncomplete = indicators.max_tokens_hit ||
16  indicators.unclosed_json ||
17  indicators.unclosed_array ||
18  indicators.has_error;
19
20return [{
21  json: {
22    text: text,
23    is_complete: !isIncomplete,
24    finish_reason: finishReason,
25    completeness_indicators: indicators,
26    char_count: text.length
27  }
28}];

1// HTTP Request node body for Cohere Chat API
2// POST https://api.cohere.ai/v1/chat
3{
4  "model": "command-r-plus",
5  "message": "{{ $json.user_message }}",
6  "preamble": "You are a helpful assistant.",
7  "max_tokens": 4096,
8  "temperature": 0.3,
9  "chat_history": []
10}

1# Environment variable for self-hosted n8n
2N8N_HTTP_TIMEOUT=600000  # 10 minutes in milliseconds
3
4# Or in the HTTP Request node Options:
5# Timeout: 120000 (120 seconds)

1// Code node: Build continuation prompt
2// This runs when the IF node detects an incomplete response
3
4const partialText = $json.text || '';
5const originalPrompt = $json._original_prompt || '';
6const continuationCount = ($json._continuation_count || 0) + 1;
7const maxContinuations = 3;
8
9if (continuationCount > maxContinuations) {
10  // Reached max continuations — return what we have
11  return [{
12    json: {
13      final_text: $json._accumulated_text + partialText,
14      status: 'max_continuations_reached',
15      parts: continuationCount
16    }
17  }];
18}
19
20// Build chat history with the partial response
21const chatHistory = [
22  { role: 'USER', message: originalPrompt },
23  { role: 'CHATBOT', message: partialText }
24];
25
26return [{
27  json: {
28    message: 'Continue your response from exactly where you left off. Do not repeat any content.',
29    chat_history: chatHistory,
30    _original_prompt: originalPrompt,
31    _accumulated_text: ($json._accumulated_text || '') + partialText,
32    _continuation_count: continuationCount
33  }
34}];

1// Code node: Split a long content task into chunks
2// Example: Generate a report with multiple sections
3
4const topic = $json.report_topic;
5const sections = [
6  'Executive Summary',
7  'Market Analysis',
8  'Competitive Landscape',
9  'Recommendations',
10  'Conclusion'
11];
12
13const chunks = sections.map((section, index) => ({
14  json: {
15    section_name: section,
16    section_index: index,
17    total_sections: sections.length,
18    prompt: `Write the "${section}" section of a report about ${topic}. ` +
19      `This is section ${index + 1} of ${sections.length}. ` +
20      'Write 200-400 words. Return ONLY the section content, no headers.',
21    max_tokens: 1024
22  }
23}));
24
25return chunks;