signafAIde
Documentation menu

Audit evidence from your tracing tool

A trace shows your team what an agent did. A signed record shows it to someone outside your company. Each page here takes the set-up a tracing tool documents and adds one line: the tool-call spans it already receives also become signed, chained records on your machine.

Your tracing tool keeps receiving the same spans. The processor reads them inside your process and exports nothing. With tool-call checks on, it also records what the checks find, Calibrating only: it never blocks a tool. How it works, the verify output and sampling: Audit log from OpenTelemetry traces.

  • Langfusewith Pydantic AI in Python and the Vercel AI SDK in TypeScript, the decision is written before the tool runs.
  • LangSmiththe spans LangSmith writes for a LangChain agent in Python, where the decision is written after the tool ran.
  • Arize Phoenixthe OpenInference spans of a LangChain agent, in Python and TypeScript, where the decision is written after the tool ran.
  • Datadog LLM Observabilitythe OpenTelemetry set-up that sends spans to Datadog, in Python and TypeScript.
  • Traceloopthe OpenLLMetry SDK and its tool spans, in Python and TypeScript, where the decision is written after the tool ran.

Next: why a trace alone is not evidence. Why your traces are not audit evidence