Honeycomb Distributed Tracing
Product · Developer tool by Honeycomb
Follow requests across microservices and AI agents in a single trace.
Honeycomb Distributed Tracing is an observability solution for engineering teams to follow requests across microservices and AI agents. It provides a unified waterfall view of latency, errors, and LLM calls to resolve production issues in minutes.
- Languages
- Java, Ruby, Python, Go, NodeJS, .NET, JavaScript, Android, iOS
Category: Observability & Monitoring
- Platforms
- Web, API, AWS, Google Cloud, Microsoft Azure
- Deployment
- SaaS, Honeycomb Private Cloud (AWS)
- Supported Languages
- 40+
- Data Ingestion Standard
- OpenTelemetry (OTLP)
- Query Performance
- Sub-second
- Event Size Limit
- 1MB
- Follow requests across microservices and AI agents in a single trace
- Sub-second query times across high-cardinality production data
- OpenTelemetry-native with support for 40+ programming languages
- Trace LLM calls, tool invocations, and token cost inline
Overview
Honeycomb Distributed Tracing is an observability tool designed for engineering teams to follow requests across microservices and AI agents. It provides a unified waterfall view that combines latency data, errors, retries, and LLM calls to help teams isolate the root cause of system slowdowns. The product is built on an OpenTelemetry-native foundation, allowing it to ingest high-cardinality data from over 40 programming languages without vendor lock-in. It enables developers to trace multi-agent conversations end-to-end, tracking token cost and latency per model alongside traditional system spans. By retaining high-value relationships across production data, Honeycomb delivers sub-second query times even when analyzing billions of log lines.
Analysis and Debugging
- End-to-End Trace Visualization: Engineers can follow a single request as it processes across microservices, queues, retries, and AI agents in one unified view.
- Waterfall Analysis: The waterfall view displays latency, errors, retries, and downstream dependencies together to isolate the cause of slowdowns.
- Interactive Trace Filtering: Users can filter by any field or click any spot in a graph to jump directly to the associated trace for deep investigation.
- High-Cardinality Querying: The platform allows querying and graphing over any attribute in trace spans, such as database calls or specific request URLs.
AI and Agent Observability
- Agent Timeline View: Honeycomb captures one GenAI conversation per request with child spans for every LLM call, tool invocation, and handoff.
- Loop and Retry Detection: The platform surfaces agent loops, retries, and runaway behaviors the moment they start to prevent cost spikes.
- Token Cost Tracking Explorer: Engineers can read token usage from provider responses and tag it with features, environments, and outcomes for unit-economic analysis.
- Multi-Agent Workflow Tracing: Teams can trace multi-agent conversations end-to-end to see LLM calls and system failures in a single view.
Instrumentation and Ecosystem
- Broad Language Support: The platform supports over 40 programming languages, including Java, Ruby, Python, Go, and.NET.
- OpenTelemetry Native Foundation: Honeycomb uses OpenTelemetry-native instrumentation to avoid proprietary agents and vendor lock-in.
- AWS Lambda Integration: Official AWS-managed OpenTelemetry Lambda layers and Honeycomb extensions allow for tracing serverless functions.
- Google Cloud Integration: The platform ingests structured telemetry from GCP services including Cloud Functions, Pub/Sub, and BigQuery.
Use cases
- Engineers trace a request across microservices to identify which specific service is causing a 500 error or high latency.
- AI developers monitor multi-turn agent conversations to debug hallucinations or tool call failures in production.
- SRE teams use the waterfall view to identify N+1 queries or slow database operations within complex distributed traces.
- Platform engineers track token spend per feature and model to control costs as AI usage scales across the organization.
Specifications
- Platforms
- Web, API, AWS, Google Cloud, Microsoft Azure
- Programming languages
- Java, Ruby, Python, Go,.NET, NodeJS, JavaScript, Android, iOS, React Native
- Integrations
- AWS Lambda, Amazon EKS, Google Cloud Functions, Google BigQuery, Kubernetes, Claude Code, Cursor, Amazon Q Developer
- Deployment
- SaaS, Honeycomb Private Cloud (AWS)
- Supported Languages
- 40+
- Data Ingestion Standard
- OpenTelemetry (OTLP)
- Query Performance
- Sub-second
- Event Size Limit
- 1MB
- Maximum Columns per Event
- 2000
- Data Availability
- Under 90 seconds
- Retention (Usage Page)
- 60 days of volume history
How it works
- Instrument Systems: Instrument your applications using OpenTelemetry SDKs or Honeycomb-specific extensions for platforms like AWS Lambda.
- Ingest Telemetry: Send structured telemetry data as events to Honeycomb's high-performance columnar data store.
- Analyze and Debug: Use the waterfall view and BubbleUp to identify outliers and trace requests across your entire stack.
Pricing
- Free Plan
- Free forever (up to 20M events/month)
- Pro Plan
- Starting at $150 per month
- Enterprise Plan
- Custom pricing (starts at 10B events/year)
FAQ
How does Honeycomb help teams trace AI agent behavior?
Honeycomb renders multi-agent workflows as a single coherent view in the Agent Timeline, connecting LLM calls, tool invocations, and handoffs in one trace.
Can I analyze traces alongside logs and metrics?
Yes, Honeycomb allows engineers to analyze traces alongside logs and metrics in a single workflow to pinpoint root causes faster.
Added by AntSpace from public sources (honeycomb.io).
Related
- Intercom: Scaling Fin to 2x Engineering Productivity
- Scribe: 75% Reduction in Observability Costs
- Homeaglow: 40x Performance Gain in Critical Booking Flow
- From 93% to 99%: How hipages Found (and Fixed) Hidden Failures
- Honeycomb Observability Platform
- Refinery (Honeycomb Telemetry Pipeline)
- Honeycomb Private Cloud