Integration gallery¶
One short recipe per trace source. Each one shows the command or code, the real output it produced on a trace from the repository's fixtures, what you get, and its limits.
-
Any agent that emits the OpenTelemetry GenAI semantic conventions, exported as OTLP-JSON.
-
Spans from an OpenInference instrumentor, the format Phoenix collects.
-
Spans from Traceloop's OpenLLMetry instrumentations.
-
Ingest a Strands OTel export, or let the toolkit drive the topology.
-
.evallogs from Inspect runs, one trace per sample. -
Datasets: JSONL, parquet, HuggingFace
A table of captured traces, one trace per record.
-
Read every trace in a Langfuse project, or use Langfuse as the store.
-
An agent you can only call: prompts in, replies out, no steps.
-
Run an agent deployed to AgentCore Runtime over generated inputs, read a runtime's traces for a time window, or probe it live. The full integration has its own page.
Not listed? If your framework emits OpenTelemetry spans, or has an OpenLLMetry or OpenInference instrumentation you can turn on, start with OpenTelemetry GenAI spans: the same command detects the convention.