Skip to main content
RunAgain is an observability + evaluation platform for LLM agents. Point any OpenTelemetry-instrumented agent at RunAgain and get full-fidelity traces, eval scores, drift detection, and regression diffs — for agents built on the Claude Agent SDK, the Vercel AI SDK, or driven from Claude Code in your terminal.

Quickstart

Send your first trace in a couple of minutes.

Trace Claude Code

Capture your terminal Claude Code sessions as traces.

Observability

How traces, spans, and observation types work.

Evaluations

LLM-judge, assertion, rubric, and dataset scoring.

​
What you get

Observability

Full-fidelity traces of every run, tool call, token and latency.

Evaluations

LLM-judge, assertion, rubric and dataset scoring — or bring your own.

Drift detection

Catch quality, latency and cost drifting from baseline.

Tool-call mocking

Freeze flaky or paid APIs so eval runs stay deterministic.

Tool simulation

Stress-test how your agent reacts to simulated tools.

Datasets & experiments

Version datasets and score experiments against a baseline.

​
How it works

1

Instrument your agent

Add the RunAgain SDK, the Vercel AI SDK telemetry exporter, or enable Claude Code’s OpenTelemetry — all speak OTLP.
2

Traces stream to RunAgain

Spans are ingested over OTLP/HTTP at https://ingest.runagain.ai, authenticated with a project API key.
3

Observe, score, and iterate

View traces, run evals, watch for drift, and compare experiments in the dashboard at app.runagain.ai.
Ready? Head to the Quickstart.