Honeycomb

Honeycomb is an observability platform built on distributed tracing, combining log analytics, time series metrics, frontend observability and a telemetry pipeline, with a distinct product line for observing AI agents and large language model applications.

BubbleUp is the capability Honeycomb is best known for and it inverts how debugging usually works. The normal process is to form a hypothesis about what the slow or failing requests have in common and then write a query to test it, which only finds causes someone already suspected. BubbleUp takes a selected set of anomalous events and automatically compares every dimension against the baseline, surfacing what actually distinguishes them. That finds the causes nobody thought to look for, which are the ones that take days.

The AI agent observability line is genuinely purpose-built rather than relabelled tracing. Agent Timeline, LLM Observability, an Agentic Intelligence Canvas, Model Context Protocol support and MCP Skills address a workload where a single user request fans out into many model calls and tool invocations, and where the failure is often a wrong answer rather than an error code. Conventional monitoring has nothing to say about that.

Service level objectives, a service map, anomaly detection and OpenTelemetry support round out the platform, with private cloud deployment available. Honeycomb's team also wrote Observability Engineering, now in a second edition with 27 new chapters. The free tier covers up to 20 million events and 100 million time series data points, Pro starts at 150 dollars a month for 50 million events scaling to 750 million, and the telemetry pipeline starts at 0.10 dollars per gigabyte.