Skip to main content

Message Types

Copy MarkdownDownload Markdown

The API sends different message types during streaming to provide rich context.

AI Content​

The main response text from the AI comes in the standard OpenAI format:

{
  "choices": [{
    "delta": {
      "content": "text chunk"
    }
  }]
}

How to handle it:

for chunk in stream:
  # Check for AI content
  if chunk.choices and chunk.choices[0].delta.content:
      print(chunk.choices[0].delta.content, end="", flush=True)

Step Messages​

Shows what the AI is doing in real-time (e.g., "Planning", "Searching documents", "Analyzing results"):

{
  "custom_data": {
    "type": "step_message",
    "title": "Searching IRC Code",
    "status": "in_progress"
  }
}

How to handle it:

if hasattr(chunk, 'custom_data'):
  if chunk.custom_data.get('type') == 'step_message':
      title = chunk.custom_data.get('title')
      print(f"\nStep: {title}")

Status values:

  • in_progress - Currently running
  • completed - Finished successfully
  • failed - Encountered an error

Source Messages​

Shows which documents were referenced to generate the answer. This helps verify the answer's accuracy:

{
  "custom_data": {
    "type": "source_message",
    "content": [
      {
        "node_id": "source_123",
        "text": "Section 179 allows businesses...",
        "metadata": {
          "file_name": "IRC_Code.pdf",
          "page": 42
        },
        "score": 0.95
      }
    ]
  }
}

How to handle it:

if chunk.custom_data.get('type') == 'source_message':
  sources = chunk.custom_data.get('content', [])
  print(f"\nFound {len(sources)} sources:")
  for source in sources:
      file_name = source.get('metadata', {}).get('file_name', 'Unknown')
      page = source.get('metadata', {}).get('page', 'N/A')
      print(f"  {file_name} (page {page})")
Advanced Usage

For detailed information on how to use Page Labels for navigation and Source Origin links, see the Page Labels and Source Origin guide.

Suggestions​

Follow-up questions that users might want to ask based on the current conversation:

{
  "custom_data": {
    "type": "suggestions",
    "suggestions": [
      "What are the dollar limits for section 179?",
      "What types of property qualify for section 179?",
      "How does section 179 differ from bonus depreciation?"
    ]
  }
}

How to handle it:

if chunk.custom_data.get('type') == 'suggestions':
  suggestions = chunk.custom_data.get('suggestions', [])
  print(f"\nSuggested questions:")
  for i, suggestion in enumerate(suggestions, 1):
      print(f"  {i}. {suggestion}")

Human Message​

Echo of the user's question:

{
  "type": "human_message",
  "content": "What is section 169?"
}

AI Metadata​

Additional information about the response:

{
  "type": "ai_metadata",
  "metadata": {
    "model": "bizora-1.0",
    "tokens_used": 150,
    "processing_time_ms": 1250
  }
}

Complete Example​

Here's a complete example that handles all message types:

# Enable streaming for complete message handling
stream = client.chat.completions.create(
  model="bizora-1.0",
  messages=[{"role": "human", "content": "What is section 179?"}],
  stream=True
)

for chunk in stream:
  # AI content - the actual answer
  if chunk.choices and chunk.choices[0].delta.content:
      print(chunk.choices[0].delta.content, end="", flush=True)
  
  # Custom messages - steps, sources, suggestions
  elif hasattr(chunk, 'custom_data'):
      msg_type = chunk.custom_data.get('type')
      
      if msg_type == 'step_message':
          print(f"\nStep: {chunk.custom_data.get('title')}")
      
      elif msg_type == 'source_message':
          sources = chunk.custom_data.get('content', [])
          print(f"\nFound {len(sources)} sources")
      
      elif msg_type == 'suggestions':
          suggestions = chunk.custom_data.get('suggestions', [])
          print(f"\n{len(suggestions)} suggested questions")

Deep Research Mode​

Here's a complete example that handles all message types with deep research:

# Enable streaming with deep research
stream = client.chat.completions.create(
  model="bizora-1.0",
  messages=[{"role": "human", "content": "Analyze the tax treatment of cryptocurrency staking rewards and airdrops."}],
  stream=True,
  extra_body={"askMode": "tax_research_deep_research"}
)

for chunk in stream:
  # AI content - the actual answer
  if chunk.choices and chunk.choices[0].delta.content:
      print(chunk.choices[0].delta.content, end="", flush=True)
  
  # Custom messages - steps, sources, suggestions
  elif hasattr(chunk, 'custom_data'):
      msg_type = chunk.custom_data.get('type')
      
      if msg_type == 'step_message':
          print(f"\nStep: {chunk.custom_data.get('title')}")
      
      elif msg_type == 'source_message':
          sources = chunk.custom_data.get('content', [])
          print(f"\nFound {len(sources)} sources")
      
      elif msg_type == 'suggestions':
          suggestions = chunk.custom_data.get('suggestions', [])
          print(f"\n{len(suggestions)} suggested questions")

Audit Research Mode​

Here's a complete example that handles all message types with audit research:

# Enable streaming with audit research
stream = client.chat.completions.create(
  model="bizora-1.0",
  messages=[{"role": "human", "content": "Analyze the reporting requirements for lease modifications under ASC 842."}],
  stream=True,
  extra_body={"askMode": "audit_research"}
)

for chunk in stream:
  # AI content - the actual answer
  if chunk.choices and chunk.choices[0].delta.content:
      print(chunk.choices[0].delta.content, end="", flush=True)
  
  # Custom messages - steps, sources, suggestions
  elif hasattr(chunk, 'custom_data'):
      msg_type = chunk.custom_data.get('type')
      
      if msg_type == 'step_message':
          print(f"\nStep: {chunk.custom_data.get('title')}")
      
      elif msg_type == 'source_message':
          sources = chunk.custom_data.get('content', [])
          print(f"\nFound {len(sources)} sources")
      
      elif msg_type == 'suggestions':
          suggestions = chunk.custom_data.get('suggestions', [])
          print(f"\n{len(suggestions)} suggested questions")