Currai — AI observability skill

This skill teaches you how to instrument an AI application with Currai, an open-source LLM observability platform. Use it whenever a user asks to "add tracing", "log my LLM calls", "see token usage", "debug an agent", or wire up currai.

1. What Currai is

Currai is a workspace-scoped observability platform for LLM apps. A single request is recorded as a trace containing a tree of observations:

• Generation — one LLM call. Carries model, modelParameters, usage (token counts), and input/output.
• Span — a unit of work that isn't an LLM call: tool invocations, retrieval, MCP connect/close, custom workflow steps. Spans can nest under generations via parentObservationId.
• Event — a point-in-time marker with no duration.

The mental model mirrors OpenTelemetry: one trace per logical request, observations are children of the trace (and can be children of each other), every observation has a start and end.

2. Install & environment

pnpm add currai