Implementation:Langfuse Langfuse Framework Trace Seed Data

Knowledge Sources	Langfuse
Domains	Seed Data, Framework Integration, Observability
Last Updated	2026-02-14 00:00 GMT

Overview

A collection of nine static JSON files containing real-world trace data from various AI agent frameworks, used as seed data to test Langfuse's rendering and processing of diverse framework-specific trace formats.

Description

These JSON files contain actual trace data captured from different AI agent frameworks integrated with Langfuse via OpenTelemetry or native SDKs. Each file represents a complete trace from a specific framework, including the trace object, its child observations (spans/generations), and any associated scores. They are loaded by the FrameworkTraceLoader class during the seeding process and inserted into ClickHouse.

All files follow the same JSON structure:

{
  "trace": {
    "id": "...",
    "projectId": "...",
    "name": "...",
    "timestamp": "...",
    "environment": "default",
    "input": "...",
    "output": "...",
    "metadata": "...",
    "scores": [...],
    ...
  },
  "observations": [
    {
      "id": "...",
      "traceId": "...",
      "type": "SPAN|GENERATION|AGENT|TOOL",
      "parentObservationId": "...",
      ...
    },
    ...
  ]
}

The nine framework trace files are:

File	Framework	Date	Trace Name	Description
agno-2025-06-11.json	Agno	2025-06-11	Agent.run	Agent using DuckDuckGo search tool to find Twitter trends, via openinference.instrumentation.agno
beeai-2025-08-01.json	BeeAI	2025-08-01	beeai-framework-main	ReActAgent using Wikipedia and OpenMeteo tools for Barcelona trip planning, via openinference.instrumentation.beeai
google-adk-2025-08-28.json	Google ADK	2025-08-28	invocation [hello_app]	Simple hello agent using Google's Agent Development Kit, via openinference.instrumentation.google_adk
koog-2025-08-26.json	AI Koog	2025-08-26	agent.f9f77d1a...	Agent with AGENT-type observations, service name ai.koog, version 0.3.0
langgraph-2025-08-22.json	LangGraph	2025-08-22	LangGraph	Web search agent using LangGraph with tool calls, via langfuse-sdk scope
microsoft-agent-2025-10-14.json	Microsoft Agent Framework	2025-10-14	invoke_agent 99981f6a...	Weather agent using Microsoft's agent framework with get_weather tool
openai-agents-2025-09-30.json	OpenAI Agents	2025-09-30	Agent workflow	Multi-agent workflow via openinference.instrumentation.openai_agents
openai-assistants-2024-05-29.json	OpenAI Assistants	2024-05-29	run_math_tutor	Math tutor assistant solving equation 3x + 11 = 14, using native Langfuse SDK
pydantic-ai-tools-2025-12-04.json	Pydantic AI	2025-12-04	joke-generator	Joke generator with tool calls using Pydantic AI, via langfuse-sdk

Each file captures framework-specific metadata patterns in the metadata field, including resourceAttributes (telemetry SDK info), scope (instrumentation library name/version), and attributes (framework-specific span attributes like openinference.span.kind, gen_ai.operation.name, etc.).

Usage

Use these files when:

• Testing Langfuse's trace rendering for different AI framework formats.
• Verifying that the seeding pipeline correctly loads and inserts framework-specific trace data.
• Adding new framework trace examples for testing (create a new JSON file following the same structure).
• Understanding the trace data shape produced by various AI agent frameworks.

Code Reference

Source Location

• Repository: Langfuse
• File: packages/shared/scripts/seeder/utils/framework-traces/
• Files:
  • agno-2025-06-11.json
  • beeai-2025-08-01.json
  • google-adk-2025-08-28.json
  • koog-2025-08-26.json
  • langgraph-2025-08-22.json
  • microsoft-agent-2025-10-14.json
  • openai-agents-2025-09-30.json
  • openai-assistants-2024-05-29.json
  • pydantic-ai-tools-2025-12-04.json

Signature

// Common JSON structure for all framework trace files:
{
  "trace": {
    "id": "<hex-string>",
    "projectId": "<uuid>",
    "name": "<framework-specific-trace-name>",
    "timestamp": "<ISO-8601>",
    "environment": "default",
    "tags": [],
    "bookmarked": false,
    "input": "<JSON-string>",
    "output": "<JSON-string>",
    "metadata": "<JSON-string with resourceAttributes, scope, attributes>",
    "scores": [{ "id": "...", "name": "...", "source": "EVAL", ... }],
    "latency": <number>
  },
  "observations": [
    {
      "id": "<string>",
      "traceId": "<string>",
      "type": "SPAN|GENERATION|AGENT|TOOL",
      "environment": "default",
      "parentObservationId": "<string|null>",
      "startTime": "<ISO-8601>",
      "endTime": "<ISO-8601>",
      "name": "<string>",
      "input": "<JSON-string>",
      "output": "<JSON-string>",
      "metadata": "<JSON-string>",
      ...
    }
  ]
}

I/O Contract

Inputs

Name	Type	Required	Description
N/A	N/A	N/A	These are static JSON data files. They are read by the FrameworkTraceLoader class during seeding.

Outputs

Name	Type	Description
trace	Object	Complete trace record with id, name, input, output, metadata, scores, and latency
observations	Array of Objects	Child span/generation/agent/tool observations with hierarchical parent relationships

Usage Examples

// Loaded by FrameworkTraceLoader in seeder-orchestrator.ts:
import { FrameworkTraceLoader } from "./framework-traces/framework-trace-loader";

const loader = new FrameworkTraceLoader();
const { traces, observations, scores } = loader.loadTracesForProject("project-123");

// traces, observations, and scores are ready for ClickHouse insertion
await queryBuilder.executeTracesInsert(traces);
await queryBuilder.executeObservationsInsert(observations);
await queryBuilder.executeScoresInsert(scores);