LumiqTrace is an AI agent observability platform that tracks what happens inside multi-agent systems during production execution.
When an agent runs, LumiqTrace captures every decision, tool call, sub-agent handoff, and planning span as a structured trace. Each span carries agent identity you can see which agent owns each step, what context was passed during delegation, what came back, and the
latency and cost of each sub-execution. An agent map is built automatically from live execution data without manual configuration.
Setup uses provider auto-patching: initializing LumiqTrace silently instruments OpenAI, Anthropic, Gemini, Bedrock, and Mistral calls. Framework-level tracing adds one handler per framework LangChain, CrewAI, Google ADK, and OpenAI Agents SDK are supported.
The platform includes 12 built-in LLM-as-judge evaluation templates faithfulness, relevance, toxicity, groundedness, and others that run automatically on every trace. No custom scoring functions required.
A cost optimizer analyzes trace data to surface token waste, inefficient prompt patterns, and model swap opportunities with estimated dollar savings.
LumiqPilot is an AI ops assistant within the platform. It reads live trace data to answer questions about cost spikes, failure patterns, or latency regressions. From the same interface, users can create alerts, switch models, or roll back prompts.
Free tier includes 10,000 traces per month with no credit card required.
