Trace Format Reference

When you send a trace to Mibo, the data field is passed to a platform-specific parser that extracts the response text, node executions, and tool calls. The parser is selected based on the platform type the trace is routed to, not by inspecting the data.

Tip

If you’re building a custom integration that isn’t n8n or Flowise, connect it as a Custom API platform. The Custom API parser is the most flexible and has the simplest format requirements.

The recommended pattern

Regardless of the platform, the same principle applies: structure your trace data so Mibo can find what it needs to evaluate.

At a minimum, include the response text in a common key:

{
  "data": {
    "text": "The AI responded with this message."
  }
}

This is enough for semantic assertions and basic checks like response_regex or json_match.

Adding tool calls

If your AI uses tools or functions, add a tool_calls array:

{
  "data": {
    "text": "I booked your appointment for tomorrow at 9 AM.",
    "tool_calls": [
      {
        "name": "create_booking",
        "arguments": {
          "date": "2026-03-09",
          "time": "09:00"
        }
      }
    ]
  }
}

Adding node-level detail (for node_call assertions)

If your system has multiple steps or nodes (like a workflow), the best approach is to use each node name as a key mapping to its output data. This structure lets Mibo match nodes by name, which is what makes node_call assertions work.

{
  "data": {
    "Fetch Data": {
      "output": { "items": 42 },
      "type": "http-request",
      "status": "success"
    },
    "AI Agent": {
      "output": { "text": "Based on the 42 items found..." },
      "type": "ai-agent",
      "status": "success",
      "tools_called": [
        {
          "name": "summarize",
          "input": { "count": 42 },
          "output": { "summary": "..." }
        }
      ]
    }
  }
}

With this structure, you can write assertions like:

{
  "target": "node_call",
  "condition": "MUST_CALL",
  "expected_name": "AI Agent",
  "expected_arguments": { "text": { "matcher": "contains", "value": "42 items" } }
}

The node names in your trace ("Fetch Data", "AI Agent") are the same names you reference in expected_name.

Platform-specific formats

Each platform parser has its own expectations for the exact data structure. See the detailed format for your platform:

Custom API The simplest format. Response text + optional tool calls.

Flowise Flowise prediction response format with agentFlowExecutedData.

n8n Three auto-detected formats: optimized, manual node names, and API mode.

Common fields (all platforms)

These fields go in the request body alongside data, not inside it:

{
  "data": { ... },
  "platformId": "550e8400-e29b-41d4-a716-446655440000",
  "externalId": "your-tracking-id",
  "metadata": {
    "workflowId": "wf-123",
    "chatflowId": "cf-456",
    "environment": "production",
    "version": "1.2.0"
  }
}

Field	Required	Description
data	Yes	The trace payload (format depends on platform type)
platformId	No	Explicitly route to a platform
externalId	No	Your own tracking identifier (max 255 chars)
metadata	No	Extra context stored with the trace. Can include platform identifiers for auto-routing (e.g., chatflowId, workflowId), environment info, or any other key-value pairs.

Headers

Header	Required	Description
x-api-key	Yes	Your Mibo API key
Content-Type	Yes	application/json
Content-Encoding	No	gzip for compressed payloads

Quick decision guide

Not sure which format to use?

1. Is your system n8n?

   Use the Mibo Testing Node. It sends traces automatically in the optimized format. No manual formatting needed.

2. Is your system Flowise?

   Send the Flowise prediction response as-is in data. Include agentFlowExecutedData if you want node-level assertions.

3. Is it anything else?

   Connect as Custom API. Put your response text in a common key (text, message, output, etc.) and add tool_calls if your AI uses tools.

Tip

Start with the simplest format (just text) and add more structure as you need more assertion types. Semantic assertions only need the response text. Node and tool assertions need the full trace structure.