Immersive Fusion proudly announces the open-source release of tracegen, a single-binary distributed trace generator that produces production-realistic OpenTelemetry data at a scale and complexity that pushes APM platforms to their visualization limits. Available today under the Apache 2.0 license, tracegen is both a practical testing tool and an invitation to rethink what observability should look like.

The Problem: APM Evolved Along the Wrong Axis

The APM industry optimized for data collection and query performance, not for how engineers actually experience their systems. The result: platforms that ingest enormous volumes of telemetry and render it as flat panels, waterfall charts, and paginated tables. Engineers pay per host, per GB, per seat, and what they get back is a 2D view of a fundamentally multi-dimensional problem.

Your application is a living distributed system: dozens of services, hundreds of pods, thousands of traces flowing through message queues, caches, databases, and now LLM inference chains. The data is there. The visualization has not kept up.

Meanwhile, the rise of AI workloads created an entirely new observability gap. LLM calls, agent loops, vector searches, and token budgets produce telemetry that most platforms surface in separate views rather than as part of a unified system topology. Specialized LLM observability tools (Langfuse, LangSmith, Helicone) emerged to fill this gap before a standard existed. The OpenTelemetry GenAI semantic conventions are now working to unify this space, but remain experimental, and traditional APM pipelines were not designed with LLM-specific telemetry in mind. The result: engineers still need multiple tools to observe different parts of the same system.

Immersive Fusion built tracegen to make these gaps visible.

What tracegen Generates

tracegen simulates a complete e-commerce platform with AI capabilities, streaming real OTLP data to any backend:

• 28 microservices across API gateway, authentication, order processing, payment, inventory, cart, product catalog, search, caching, notifications, email, shipping, tax, scheduler, user management, analytics, fraud detection, recommendations, configuration, and a web frontend, plus 8 dedicated AI/LLM services (LLM gateway, embeddings, vector DB, agent orchestration, content safety, model registry, feature store, data pipeline)
• 59 pod instances distributed across 5 simulated AKS VMSS nodes with realistic Kubernetes resource attributes
• 40 scenario flows covering checkout, payment processing, search, authentication, inventory sync, and 12 AI agentic patterns including RAG, multi-step agents, and conversational commerce
• Realistic stack traces emulating .NET ecosystem libraries (Npgsql, StackExchange.Redis, Stripe SDK, OpenAI SDK)
• Database, cache, and messaging spans with proper semantic conventions for PostgreSQL, Redis, RabbitMQ, and Kafka

One command. Zero containers. Point it at any OTLP-compatible endpoint, cloud or local:

tracegen -apikey YOUR_KEY -endpoint otlp.iapm.app:443
tracegen -apikey local -endpoint localhost:4317 -insecure

AI Agentic Tracing: Unified, Not Separated

While some APM vendors have begun adding AI observability features (Datadog LLM Observability, New Relic AI Monitoring), these typically appear as dedicated views alongside, rather than spatially integrated with, the distributed system topology. tracegen generates both traditional and AI traces as part of the same topology, because that is how they run in production.

All 12 AI scenarios emit spans following the OpenTelemetry GenAI semantic conventions (experimental, v0.x as of March 2026), using span naming conventions compatible with Microsoft Semantic Kernel and Agent Framework:

• RAG pipelines: Embedding generation, vector search against Qdrant, LLM reranking
• Multi-step agents: Plan/act/reflect loops with tool calls and iteration tracking
• Content moderation: Parallel safety classifiers with guardrail enforcement
• Conversational commerce: Multi-turn sessions with growing context windows

Each LLM span carries proper attributes: gen_ai.usage.input_tokens, gen_ai.usage.output_tokens, gen_ai.request.model, gen_ai.response.finish_reasons, and agent-specific fields like gen_ai.agent.id and gen_ai.tool.name.

In Immersive APM, these AI traces appear alongside traditional distributed traces in the same 3D environment. Services are rendered as spatial objects with edges encoding latency and throughput. AI agent loops, LLM inference chains, and vector searches are visible in context with the API calls and database queries that surround them. No tab switching. No separate dashboards. One unified view of the entire system.

Chaos Engineering That Shows the Blast Radius

tracegen includes failure injection modes with a tunable error rate (0-10 scale):

• Lost messages: Configurable message loss per queue hop (default 5%) simulating unreliable consumers
• Dead consumers: Producers send messages that are never consumed (also available via -no-consumers flag)
• Retry storms: Payment retries 3x with exponential backoff
• Timeout cascades: Circuit breakers serving stale cache data
• Saga compensation: 4-way parallel rollback after mid-transaction failure
• LLM rate limits: OpenAI 429 errors with token budget details
• Token budget exceeded: Agent exceeds iteration token limit mid-conversation
• Hallucinated tool calls: Agents request tools that do not exist
• Content filter blocks: Safety classifiers reject generated content

These are the failure modes that cause 3 AM pages. The question is not whether your APM platform detects them. The question is whether it shows you the cascade: which pods are affected, which downstream services are degrading, and where the blast radius ends. In Immersive APM, failure propagation is rendered spatially, with affected nodes, edges, and pods highlighted in the same 3D view where you see healthy traffic. tracegen generates the chaos so you can evaluate the difference.

Trace Generator Comparison (as of March 2026)

This table compares open-source trace generators, not APM platforms. Sources: telemetrygen, Astronomy Shop (OTel Demo), HotROD. Each tool is valuable for its intended scope. tracegen fills a different gap: generating the volume, variety, and failure complexity needed to stress-test how platforms visualize distributed systems.

Configurable Throughput

Aggressiveness levels from 1-10 control generation rate:

• Level 1 (whisper): ~2 traces/second for careful evaluation
• Level 5 (steady): ~7 traces/second for realistic production load
• Level 10 (full blast): ~350 traces/second for pipeline stress testing (measured on a 4-core workstation with 8 GB RAM; results vary by hardware and network conditions)

Scenario filtering flags (-ai-only, -no-ai-backends, -no-consumers) let teams isolate specific workload types.

tracegen -apikey $KEY -level 10 -errors 10

Compatible With Every OTLP Backend

tracegen speaks OTLP over gRPC. Use it to evaluate any OpenTelemetry-compatible platform, including ours:

• Immersive APM (3D real-time visualization)
• Jaeger
• Grafana Tempo
• Honeycomb
• New Relic
• Datadog
• Splunk Observability
• Elastic APM
• Any OpenTelemetry Collector

We built tracegen to be platform-agnostic because we believe the best way to demonstrate Immersive APM's spatial visualization is to let engineers compare it directly against whatever they use today.

Open Source. Apache 2.0. No Strings.

tracegen is available on GitHub under the Apache 2.0 license. Built in Go, it compiles to a single binary for Linux, macOS, and Windows (amd64 and arm64). Download the latest release, run it, fork it, contribute.

See What Observability Should Look Like

"We built tracegen because we needed the world to see what happens when you stop treating observability as a spreadsheet and start treating it as an experience. The 3D visualization is not a gimmick. When you render services as spatial objects, cluster pods by node, encode latency as edge thickness, and animate trace flow in real time, patterns emerge that flat dashboards structurally cannot show. That is the gap we are closing." - Dobri Kowalski, Founder, Immersive Fusion

28 services. 59 pods. 40 flows. AI agents. Chaos. All rendered in immersive 3D. This is what APM should look like.

About Immersive Fusion

Immersive Fusion LLC builds Immersive APM (IAPM), an application performance monitoring platform that renders distributed systems in real-time 3D. Founded on the belief that engineers deserve to see their applications the way those applications actually run, Immersive Fusion combines spatial visualization, AI-guided diagnostics, and full OpenTelemetry support to deliver observability that goes beyond dashboards.

Media Contact: Dobri Kowalski, dkowalski@immersivefusion.com

Download tracegen on GitHub. Free, open source, no signup required.

Start Free. Immersive. AI-guided. Full-stack observability. Enter the World of Your Application®.