@error-bar/tracing for Node
and errorbar-tracing for Python, stream your LLM traffic to
errorbar, where you grade it, calibrate a judge against your own standards, and
find out with confidence intervals whether a cheaper model holds up on
your production traffic. Your inference does not move: requests keep
going to your current provider; only trace telemetry flows to errorbar.
No instrumentation code of our own
Each package contains no instrumentation code of its own. It pins and configures the ecosystem’s standard OpenTelemetry instrumentations, which gives it a property no other tracing SDK offers: you can uninstall it without losing your instrumentation. The identical setup in vanilla OTel is documented on each SDK page (the eject guarantee); your spans are byte-for-byte the same either way, and nothing proprietary ever goes on the wire.When to use the SDK, and when not to
The SDK is one of four integration paths:- Already routing through the gateway or bringing your own key? You don’t need an SDK; the gateway captures your traffic in the request path.
- Already emitting OpenTelemetry (Vercel AI SDK telemetry, Pydantic AI, an existing OTel setup)? You don’t need this package at all; three env vars point your existing exporter at errorbar. See OTLP trace ingest.
- No OTel yet, and your inference must not move? Use the SDK: one install, one line, standard OpenTelemetry underneath.
Privacy
Span structure is always stored. Model-call content (prompts/completions) is stored only if your errorbar workspace has request logging enabled, under your retention window, with the same scrubbing and size caps as gateway traffic.Packages
Both are Apache-2.0.